Agents@Work — How Enterprise Teams Deploy AI Agents in Production

Welcome to Agents at Work.

I'm Daniel from Relevance AI. With this series, we had one goal: bring you content from the world's top companies on how they're winning with their agents. We're going to be hearing from companies like Canva, KPMG, Autodesk, and Lightspeed on all the things they're doing to see success. We're going to hear from the people breathing life into the agents and training them, all the way through the executives enabling their teams to win. I'm very excited to have you join us. Let's get into it. Hello, and I'm super excited to introduce Alan Roe from Autodesk with us today to share a little bit about how agents are being used in production. Thanks so much for joining us.

Hey, thanks for having me.

Now, Autodesk is a phenomenal company. A lot of us know what it does, but in case one of our viewers doesn't, can you start off by sharing a little bit about Autodesk?

Yeah, of course. So Autodesk, we build software that helps people design and make like buildings, products, films, you name it. And we work, we serve a lot of different industry, a lot of companies from small firms to global enterprises. And yeah, we aim to bring value to everyone who's in that industry.

And I feel like Autodesk brings together design, engineering, innovation, and I feel a little bit of that is actually encapsulated within your team as well. Do you mind sharing a little bit about what kind of scope you and your team have within the broader go-to-market engine in Autodesk?

Of course. So our goal is to bring value to everyone. Now what that means is that we look at data points, data intent through different channels like email, chatbot, webinar, and we identify, hey, what is the right message for these people? What is the right content? What's the right timing? So our goal is to really bring value to these people and bring that intent into action. Also, we work very closely with all teams, mainly GTM teams, but also product marketing, marketing operations, sales operations, BD and sales to get their feedback, iterate, but also making sure that we have clear KPIs that cross the whole teams and that we're supporting that KPI, but also making sure that people are getting most of their product journey.

And working in marketing automation, I think Obviously AI agents are becoming more and more relevant to that role, like many others. When you first came across agents, right, what was your initial reaction? Skepticism, optimism, pessimism? What were you thinking?

I was just very curious because anyone who used AI agents, they will know the speed. It's phenomenal. Things that would take hours on me to create content that is suited for one-to-one personalization, that'll take hours. But With AI agents, it will take minutes if not seconds, and that is such a powerful, powerful tool, and that got me excited. But on the other side, there's caution because when we— our desk, our industry knowledge, institutional knowledge is our core, and we want to make sure that whatever AI is generating or help us generate meets our standard, and we assume nothing until we saw the results. And we made sure that the results guide our next steps.

And so when you started first, I guess, working with Relevance and building agents, what were your initial expectations? Did you expect it to show success immediately? Was it more of an experience for you and the team to evolve? Where did you sit in that spectrum?

Yeah, initial expectation was that we wanted to learn quick because we knew that with the power and speed of AI agents, we knew that this was the future. But the best path right now for us to experiment quickly, learn quickly, see what sticks, making sure that we have the processes that supports the AI workforce as part of our GTM motion. So initially we started from there. We started, hey, what can we really bring value? What is that key point that is close to revenue? What is that one job that's related to that key point? And how can AI agent can help us supercharge that process so we can scale faster? That was the initial goal.

And looking back now, how do those early expectations compare to what actually happened?

Everything we expected, but also there are surprises down the road. So we expected, hey, this is going to supercharge everything because as from our lifecycle automation team, we really try to aim value. I keep saying this, trying to add value at a 1-to-1 personalized scale. And AI agents help us do that. Now, the Something that we did not quite expect was that how powerful it can be, which means that you need a lot of processes, a lot of setting, especially it is any, like any workforce, you have to kind of train your agent to make sure that they understand your north star and your brand tone and voice. Like any AI work, any workforce, AI agents needs to be trained on that goal. What your goal is, what your KPIs are. So, uh, it, when it creates the content or any processes, helps with the processes, it's up to our standard. So that was, uh, it took a lot of iterations and training.

Yeah, absolutely. And that's an interesting point that I tend to see time and time again, that treating your AI workforce in a similar way to how you treat a regular workforce can be really powerful in order to iterate and get the results that you want. So let's talk about the first pilot program. You and your team, as you mentioned, took a very experimental approach. You iterated on different agents to see what's most effective. Can you share a little about the process? Like if someone else was starting a journey today, how could they most effectively get started and iterate their way to success as well?

Absolutely. So for AI agents, the power is immense, but you have to have a, not necessarily strict, but very solid guideline of where you want to be. Initial point was that, okay, AI agents is really good at scaling. How can we really use its strengths in our process? So in our whole lifecycle journey, we identify, hey, what is that point where it's close to revenue, but that has a clear objective or goal that we can replicate and scale in a meaningful manner? So to achieve that, it takes a lot of cross-team collaboration. We work with product marketing team, we work with BD, we work with sales, we work with, you know, legal. Even making sure that everything's compliant, but also we are really doing things the right way. When we first started, that was the case, but we did not assume anything because it's always data first and confidence second. In this case, we didn't assume anything, but we had a strong goal, north star, to start with. But we let the results that came in to drive the next steps. And that was very important because when you are scaling these one-to-one message that goes one-to-one to 1 to 10,000, 1 to 100,000, that becomes a bit of a challenge because everybody has different firm types or different industry, different geolocation. And this is why you have to really interlock with other teams who are getting those feedback from our customers, our prospects, and making sure that we iterate on that feedback and also that data we're getting to make sure that everything that we do, every iteration of the prompt, every new addition to the agents, are actually moving the pipeline so that we're hitting our goal. That was very important.

You mentioned kind of this idea of a North Star metric. I think a lot of leaders think a lot around how to best incentivize their team and how to best align and give them a goal. How did you set that target? Like, what were you thinking about when you were picking a North Star metric?

Great question. So in our North Star, It's about clarity.

You have—

it has to be clear. Like, for example, sure, you want to increase pipeline.

That's great.

Awesome. But that's a bit of a broad term. Where you want to start with is something actionable. For example, instead of a pipeline, how can we increase trial to conversion rates specifically through these products? That's a great starting point. And this is very important because it gives a bit of a frame or a more narrower scope where where you can really focus on a first iteration of the model. After that, after initial phase, what does happen is that now you can see, okay, the AI agent that we train based on our knowledge base, based on our experience, is really good at this thing. How can we replicate the successful parts to other products or other areas? But also, let's look at where it falls a little bit short to our standards and how can we further erase it, further test it. And that's the key. You got to start small, but also never assume the next steps. Oh, this nurture program really worked with this relevant AI agent or any agents. It's different because once you get the feedback from BD sales, from customer support, customer success, you'll see, oh wait, so this is how people are seeing, interacting with these custom content. How can we make this even better or how can we really fill the gaps in? So make sure the data needs to speak the next steps.

And so you kind of mentioned about that process of embedding Autodesk's institutional knowledge, your custom processing processes, doing the things that you guys are best at. I'm curious to understand what that actually looked like. How did you do that in practice? How did you train the agents? How did you embed that, those processes and knowledge into the workflows they have?

Absolutely.

Another great question. So Autodesk, we really like institutional knowledge and the, the playbooks are our moat. It is our core of like why we are here in the industry. And also we really want the best. We really want to truly serve the best to our customers and our prospects. So the key for us was how can we really modularize the industrial knowledge and also the best playbooks? And getting that modularized and how can we apply to not— it could be mainly for us as AI agents, but this can be applied to everywhere by making sure that we make it modular, those knowledge, make it as a shared database and shared knowledge base so that other agents can use. But also this doesn't— it's not just our team who created this. It took a lot of cooperation with the product marketing. They would do research on, hey, how can we drive a better user behavior? Or this is what we need. Getting feedback from our BD sales. Hey, this message works, did not work. Having close, you know, close relationship with those people, very important. But most importantly, the data. At the end, the data will speak if your model is working or not, because our strength is the knowledge in the industry and the best playbooks. And you can see from the data, hey, like, are we really seeing a lift on reducing churn or are we seeing a lift on the pipeline? That data will see everything, and it's very important to track every single data point from that data coming in. QLs, what's the conversion rate? What's the response rate? Based on that, you have to update your prompt, and there's no magical prompt here. There's no one prompt that fixes everything. Even within that, you'll find, wait a minute, if we break this prompt into separate agents and build a separate workforce that works with each other, it's going to better deliver our brand voice, our brand messaging. And our North Star. So overall, it's a bit of a long answer, but there's no one simple answer. It's just a lot of trial and error.

And I guess what was the point? Do you know when it was that point where you reached an output that you were happy with or results that you were happy with? What were you looking for to know that, okay, the agent's able to do the work that we needed to?

Of course, on the demand gen and growth marketing side, when we saw the data actually increase, so The engagement went up ever since AI agents started personalizing the data or the content one-to-one at scale. We got direct feedback from BD sales. In addition to the stats, we saw the pipeline increase, but also getting BD feedback, listening to their calls, seeing, oh, this is a conversation much smoother because people are better educated, better nurtured when they arrive at the BD queue or people from operations. When we identify, when we try to use AI agents as part of the enrichment or duplication process, we saw efficiencies that enables people to do better work and more meaningful work. So it's these data points plus feedback. AI agents are only good as the human that's guiding them. So it's very important to keep close on the results. And lastly, I do want to mention as part of it is The prompts age, the world change prompts age, so you shouldn't— where we got to a point where we were happy, where the blind results met our standards, but as the world changes, society changes, we knew that and we saw it, it does get outdated, so you have to constantly update it, treat it like a product. I wouldn't say never-ending, but we're not going to be at the 100% point where, oh, We create this, it's done. We don't have to touch it at all. That's not the case.

Absolutely. And going back to your comment about AI workforces being like workforces, no team ever is. Every go-to-market playbook gets evolved. We all want to get better. Exactly. That's a great piece of insight. So fast forwarding to today, you've got multiple agents working for you in your organization, iterating them. How has building and deployment changed for you from that initial agent that you put together?

Absolutely. When we first started, we had one goal, one goal of how can we bring speed and efficiencies to this single data point that we mentioned. For example, it will be something like a trial to conversion to make sure that AI agent is— we're trusting enough that AI agent is creating content, creating processes, bringing efficiencies at our standard. That was the initial part. Now it's more when we— now we're loaded into the whole process where we know we have a more working model. So based on that successful working model, we are scaling to other products, other line of products. We have a different architecture, different engineering, different construction, operational product that serves everyone in industry. So based on one working model, we want to make sure that, hey, let's take the successful parts out of the working model and scale it to other products. But let's make sure we are not just doing bursts of experiments. Let's see what sticks anymore. Scale the working ones. If something's falling short, let's do more experiments. So I would say, hey, like, experience never stops. We did a lot of experiments in the initial beginning. It never stopped. It's just we're just doing more. But what was addition to that was let's— how can we really scale the working model to different parts of the company?

Makes sense. And one of the things that I think you guys were really good at was that intercompany collaboration. You mentioned a lot of teams are required to really get the most out of your agents. Can you tell us a little bit about what you did to be so successful working with your partners across the business to build really great agents?

Absolutely. To have, and this applies for not just AI agents, but to have a successful campaign process program, everybody has to be interlocked and aligned. And the best part about the power of AI agents is once you find a working model, once you know that what sticks, the data shows your lift. And when we show that data across teams, they're excited. They see the true potential of this. So the collaboration came naturally, but specifically on that part where, like I mentioned, AI agents are only good as the human that's involved.

It has—

you have to stay close to that whole process. So that means that hearing feedback from BD sales, getting, you know, results or guidance or any product interactions from our data team or data science team, or working with the product marketing to identify, hey, what's the best user experience? And this all comes down to, hey, what is our overarching or star? Of these. And as long as you have that collaboration and a really high focus on where we want to be, it's going to be successful. But also, yeah, the data, I'll mention again, it's very important, making sure that everybody has shared analytics, KPIs to really kind of take them onto the journey. They're not just bystanders of the AI journey. People might think, oh, I did a prompt, I'm done. No, everybody's in on this. This is the future, this is the next working model, and to be successful, everybody has to be on board.

And maybe we can finish then on what other advice would you have for a leader, a marketing leader, a go-to-market leader who's looking to embark on this journey of agents and bringing them into their own companies? What are some of the things they should be thinking about or doing to be as successful as you guys? Absolutely.

AI agents are powerful, extremely powerful. I mean, you probably saw what it can do. My advice is, hey, find that data point as a starting point. Find a point where it's close to revenue, but also has a clear job that can support that data point into conversion. That's very important because when you're first starting out, it could look very massive. Oh my God, I want to be successful. I want the AI agent to do everything, but you can get there, but starting point needs to be very, very focused. From there, you'll see things like, oh, okay, if I want to drive this specific experience, I need to draft or create an agent or create an AI workforce in a certain way. And from there, you'll find a model that's going to work for your company., and this is going to look very different for different industry, different company. So you have to find your working model, and from there you're not alone, making sure you bring other teams in, get people excited about what content, what data it's producing, brand, bring brand in to make sure that your AI agents are creating content that matches your tone. The whole part is exciting, and that's why I want to tell No, people are trying out. It's not scary. It's super exciting what AI can do and the potential of it. You got to get on board, but start with something narrow and focused that's close to revenue so you can really see the results quickly, see data points, you know, more clearly, and also use the data points to drive your next actions.

That is my recommendation. Well, that's brilliant advice. Thank you so much, Alan, for joining with us, and I'm sure everyone listening will have taken so many notes because that was a wealth of knowledge. Really appreciate your time.

Thanks for sharing with us. Thank you.

Appreciate it. Great to meet you today, or I guess we've met, but maybe a quick intro from you.

You know, what what, uh, what's your name, what do you do?

Yeah, uh, my name is Michael Denari. Most people call me Denari, but I'm the head of IT at Canva. Next month will be 5 years at Canva, so it's been a wild 5 years. And I'm originally from California, but been out in Sydney the last 4 years and loving it.

So, Denari, you're joining us today from our Relevance AI HQ in Sydney, Australia, which sits right next to, uh, Canva. It's great having our customers also be our next-door neighbors. Especially with how much we've been collaborating. It's really saved on travel time.

Normally we have to fly out and maybe normally you have to fly out as well, but in this case we don't have to do as much.

Totally. But to start us off, how did you first come across Relevance AI?

And was it from walking past the sign out in front of the office? Was it from colleagues, etc.?

How did you first hear about us? Yeah, I mean, I think first hearing about you all is definitely walking by. So I'm in the office most days here at our Sydney headquarters and I think there's lots of startups with.ai at the end of the name and you're always curious about what they're doing. So definitely some Googles in the beginning, uh, Google searches to kind of find out what you all were about. But when I think when it got a little bit more serious and when you start to see a little bit more inbound from inside of Canva of people interested about the platform, wanting to use it, I think it's pretty natural being in an IT position. Lots of people want to use lots of tools across the company and you know, you kind of have to feel, well, we already have a platform for that. Uh, we're not that interested. Kind of a lot of politely swatting things down, but, you know, looking into your guys' growth and kind of the progression of the product, you know, and that kind of agent-first build, kind of intuitive, easy-to-use builder, uh, was super appealing for kind of our internal AI efforts where we're trying to really be at the forefront. So, you know, once we got on a demo and started talking to you guys, it moved pretty quickly from there, but that was a little bit of the, the origin story.

Awesome. Yeah, normally companies will park like vans or have like a billboard We parked our office right next to you guys.

Hard to deny.

Yeah.

Your role at Canva has spanned from like, you know, Head of Procurement to Head of Global Spend and now, you know, Supergroup Lead, Global Head of IT. Can you please give us an overview of your role currently?

Yeah, so we started the IT Supergroup about 4 years ago. So largely the last 5 years I've been at Canva has been focused on the IT side of things and Today, that's kind of the way that I would frame it is it's the non-product side of technology. So it's internal technology. You've got your traditional tech support devices, our space tech in our rooms, kind of your more traditional core IP. You also have our operational teams, whether it's HR or finance, the systems that are running those things. So whether it's Workday, NetSuite, all the portfolio of systems that kind of run Canva internally. And then we have a bit more of the exciting side, probably where relevance fits. But which is our teams of internal AI and automation, the solutions we deliver. We have an internal applications team that's more of a software engineering product team building internal products, but your collaboration system said that's really probably more where we view our opportunity to really transform the company to be on the forefront of innovation and try to make as many innovative tools available to Canva Knots to automate processes and really move the needle.

Awesome.

Canva was one of the first like major large companies to Essentially announced becoming AI first.

Um, and what has it kind of had in terms of the impact of your role and working with the team?

Yeah, my team likes to refer to the last, like, couple years as the AI fire tornado. It's been chaotic, fun, a big mix of emotions. And so I guess as a head of IT, it's hard to not have it the majority of my brain space and time these past couple years be really consumed by AI and are we at the forefront? I think we have purposely chosen to really make lots of tools available to Kanbanauts to try to be at that leading, you know, that AI-first kind of mentality that you talked about. And so that naturally creates lots of noise of new features, products leapfrogging each other, onboarding really new game-changing providers. And, and so yeah, I think that it's been chaotic and just really trying to make sure that we're keeping up, but, and also doing that in a secure way.

Makes sense.

And one of the things I actually hear about, even outside of this initiative, but one that's like, I hear people talk about is like the AI week, that one week where you guys gave the core company essentially time to learn and educate themselves about these AI tools. What are your thoughts on how other companies who are aspiring to become AI-first should adopt similar kind of AI weeks, et cetera?

Yeah, I mean, AI Week, uh, you know, was, was 3 days of trainings and then a 2-day hackathon for the entire company. So it was really, and, and if, if you back up, like it's just about that leadership intent. Like I talk to a lot of my peers and you might struggle to get approval for budget. Leaders are still skeptical about AI in general. Maybe they're only allowing it for certain use cases. And so I think at Canva, it's really a good indication for us of kind of being on the opposite end, end of the spectrum and. That creates its own set of challenges as well of like expectations. There's funding, there's demand, there's, there's a push to really innovate. Um, but just AI Discovery Week for us is probably a really good model to follow. Maybe not all companies can give everybody a full week off to really focus and experiment, but it's, it's a real neat needle mover to let people kind of step away from their day-to-day to experiment, to learn, to be inspired, and then to hopefully put some of those things into practice in the hackathon side of things and put some of those ideas into production for something that might change their day-to-day work.

But I will say that AI Week, actually, other people have been talking about as like the golden example, which is really, really, really awesome.

And maybe, again, we'll go a little bit off script here, but like, you know, AI Week is a really good indication of that company-wide sponsorship from a leadership perspective. But I think one of the interesting things at Canva is we look at every, what we call a supergroup or a department across the company is All of our, our department leads are trying to think about how we also enable an AI-first culture within the bounds of our department. And I think that's one of the learnings that I try to share with other companies as well, which is what it looks like for an IT team versus a finance team versus sales team. Like, yes, a week at a company level is amazing and incredible and really shows our senior leadership support to be able to experiment, but there needs to be also be this ongoing focus at that department level that's a little bit more tailored to how each individual team works. And so I think that's been a really big unlock and there's been lots of creative ideas to tailor whether for IT, we have a Monday hour of power for an hour every single Monday morning, and that might look different in finance. He does it quarterly for several days. Like, it's been really interesting to see the different styles across the company for that more local embedment of enablement of AI. An hour for power.

I like that. I'm going to steal that. That's actually such an awesome saying. My co-founder Daniel, he recently spoke with Canva's Chief Customer Officer Rob to understand how agents fit into Canva's AI-first strategy within the go-to-market team specifically. Rob highlighted how strong the partnership has been between go-to-market and IT specifically in making this all possible. When did IT first get bought into that process?

Yeah, I guess if I back out like our go-to-market and IT teams, I think this is probably true at most companies are partnering on lots of bigger system implementations, what we'll say more traditional, your Salesforce, your NetSuite, you know, maybe implementing a CPQ. And so, you know, you form these relationships early that I think are really important between whether it's systems teams or business process transformation teams, delivery teams. And so that was really, I think, the beginnings over the last couple of years is forming strong relationships there. And so then we're all experiencing this pressure, excitement, etc., to deliver AI transformation and we're all trying to figure out what this looks like and agents are becoming a thing and how is this different than more traditional automation? And so I think we were able to really build on the relationships that we had built through some of those more traditional implementations to think about how we can transform some business processes with AI, with agents, et cetera. And so yeah, we were already having some early conversations with GTM trying to look at some of our existing stack. Relevance came in, we were like, this, this could be perfect, let's run with it. And we really went on a journey and as you know, go-to-market was kind of the, the first use case that we used with Relevance, and it was a great learning for both of us on how, how do we deliver solutions in this AI-first world.

Yeah, 100%. What were the critical things your team knew had to be right from the very beginning when it comes to working with these different teams?

I think the, you know, number one, we need to have confidence that we can unlock and use data in a safe way in these platforms if we want to be able to deliver them. So I think You know, there's, it's so easy to vibe code something to get something into a prototype. And I think that's one of the awesome things about AI and it's a really great use case. It's a completely different story to get it to production. And I think IT really needs to be able to see ahead to think about what type of data are we gonna allow in different platforms? What use cases are we going to enable? And then how do we make it as frictionless as possible for a team to go from idea to production? Um, and so I, I think Really being able to think about that end-to-end, which has lots of different considerations, I guess, in the middle of, of, of that is, is I think the core role of ITs, particularly in a world where we want to enable lots of business units and supergroups across, across Canva. We learned this in the last few years in the more traditional automation, deterministic kind of core traditional automation platforms. You, you can move a lot quicker and build a lot more if you're decentralized and empowering other teams to build. IT will help you build, IT may build some solutions ourselves, but ultimately we have a very innovative culture where people are able to solve their problems largely for themselves. And so that's where more of that kind of platform governance, thinking about the right data so that we can be confident about the use cases that we're productionizing comes in.

Awesome. Nice one. Expanding on the production front. So on the security and data governance side, what are some of the key requirements for agents early on? And have these evolved as kind of adoption has grown internally?

Yeah, I mean, I think at its core, like, very basic things like role-based access control, right? We don't want one pool of users being able to touch everybody's use cases and go interact with sensitive API keys. We have monitoring and observability audit logs. We need to be able to see what our users are doing within the system. And then I think a really important one and one that was great to partner with RelevanceON is really thinking about how do we unlock certain integrations for certain use cases without exposing that to the whole company. So similar to the first. So I think, you know, it's those types of foundational security controls that enable us to move from one use case that IT is controlling everything to being able to onboard lots and lots of teams onto a platform.

Makes sense.

What did the testing and rollout process look like in practice? Yeah, so I mean, if you zoom out, I think we want to make sure that we start in a sandbox and we're able to really test and iterate before we roll something out to, uh, in this use case, it would be the, the go-to-market team. And, you know, it was really great to be able to use some of Relevance's functionality around human-in-the-loop or us being able to really validate some of the outputs for different use cases that we were having. And then over time, as we got more comfortable and confident with the output, we're able to kind of step out, move things into production, and allow our end users to kind of run, run with things independently.

Awesome. So we talked about go-to-market. Let's talk about just like in general, how do you, how does IT kind of partner with all these different parts of the organization? Because, you know, we're working with you guys, it's been very clear.

It's been an awesome partnership with all the different teams.

Yeah, I mean, I think, you know, Canva is an interesting place because you have such smart people and decentralized systems teams in some cases and every supergroup. And so it's not a one-size-fits-all approach. I think that We really want to find where is those kind of central players within each of the teams who are driving the AI roadmap, the systems roadmap. We want to partner with those folks to help them to understand these awesome platforms that we offer in a secure way or that they can feel confident building on. And so I think it's really about figuring out where our IT services and capabilities and different platforms that we offer kind of fit in in a way to accelerate each of the teams that we have internally.

Yeah. Awesome. Yeah.

Empowering the different teams.

Looking back, what has been the biggest learnings, both the wins and the challenges and the journey?

Yeah, I think that probably a couple of the biggest learnings, like we mentioned a little bit earlier, but it's very easy to get something into prototype. It's very hard to get something into production. And I think even though we in IT realize these things, like I think even when it comes to AI and agents, that's even more exacerbated. And so really being thoughtful and giving yourself a little bit more time to tinker, iterate, and get things into a place that you're really confident with was a really big one in the early days. I think being able to move with some level of urgency is also important though. Like you look at some of these more traditional projects or systems implementations, they're very regimented. They can take months and months. And so I think it's just finding that balance where you're moving with urgency, but still giving yourself enough time to get to a really confident output. And then I think the last learning would be a lot of the traditional automation approach where you, you go map your process end to end, and then let's think about how we can automate each of these steps. Like agents are creating a completely new way of architecting how we get to productionizing solutions. You know, you think about the context that you need to feed it, you think about just the patterns are completely different. And so I think there's a new muscle that we're having to learn internally about how, how do we build the right solution? How do we architect it in the right way? And I think that's been really helpful to have partners like Relevance to help us guide us along, along that journey because this is a new muscle, a new skill that you can't go find in market people with 10 years of experience in architecting, you know, agents. So I think that's been a really interesting part of the journey as well.

Awesome. Canva is continuing to invest in agents and AI more broadly to essentially free up teams for higher value work. To ultimately bring value to the customer much, much more quickly. So how does IT kind of think about this to accelerate it and democratize it even more?

Yeah, I think that probably my mental model for this is we want every single employee using AI assistants every single day. We want to accelerate their work. We want them experimenting. We want them finding patterns that are going to add value. But I think where we're moving now is more of that end-to-end business process agents being implemented. And so this requires a little bit more focus, a little bit more solutioning. And so really finding where are the areas of the business that are most ready for this? What are the highest opportunity areas for us to invest our limited resources? And how do we use these game-changing platforms that are coming out, uh, to really move the needle forward for Canva?

Makes sense.

What's some of the most exciting either initiatives or just like, you know, using AI that you're really excited about for the future?

Yeah, I'm really excited for us to really democratize these AI tools out into the company. Like, I think that we're at a really precipice point in terms of turning the corner from where some of these tools are a little bit more challenging to get to production, like we talked about earlier, but these tools are getting better, the LLMs are getting better. And so I think we're going to see a huge proliferation of internal productivity. And ability for each individual to kind of bring what they do well and kind of democratize that to the rest of the company.

So from a product perspective, what are some of the exciting things from your perspective when it comes to Canva and AI agents?

Yeah, there's a lot of exciting work being done by the product team, obviously, in our visual editor, but our ecosystem team that owns the Connect API or the API platform out for developers has been building some really incredible stuff. So you may have seen we're in the ChatGPT App Store, or you may have seen that we were in the ChatGPT applications launch and so we're well integrated there. And then there's a lot of cool use cases, whether it's with Relevance or other providers where we can programmatically create designs or insert data from other systems and have the agents kind of determine how to mock up notes or send presentations. And so I think we're going to see a lot of really cool use cases in the months and years to come internally and for our customers.

Further democratizing design. Absolutely.

Yes. Awesome. Thank you so much.

Thank you so much for taking the time to answer all of our questions. Absolutely. It's been a great partnership and appreciate all the support.

I'm joined here today by Rémy Pinto, Director of Performance Marketing at Lightspeed Commerce in the beautiful city of Montreal and an even more grand and historic building of the Lightspeed offices. Remy, do you mind just quickly sharing a little bit about the space that we're in today? Yeah, sure. Well, thank you for coming, Daniel. Welcome to Montreal. Welcome to Lightspeed. Yeah, this is a historical building. Used to be a train station, then got converted into a hotel, and now it is our office here, our HQ at Lightspeed. I'm very fortunate to be working from this beautiful historical office. Yeah, it's wonderful. What an HQ to have. Some of the folks on the team actually, they were like, have you seen their office? It's like a castle., and I was like, yeah, it was beautiful. So can you tell us a little bit about the business of Lightspeed? What do you guys do? How do you guys help power your customers? Yeah, absolutely. So Lightspeed is a commerce solution. So, uh, we're essentially a point of sale with a payment solution, and we do anything that a retailer and a restaurateur and a golf course would need to manage their operations. So from inventory management to selling online to managing your customers, we're trying to, yeah, give all the tools that small and medium businesses will need to, uh, compete in today's world. And so when you think about marketing, what are some of the key challenges or opportunities that your team has been pursuing over the last 12 months or so? Yeah, I mean, without talking about the elephant in the room, obviously, you know, in the last 2 years, I'd say we've seen a big rise in LLMs and AI technology. So I think we've seen a lot of ways now to, you know, improve the efficiency that we need throughout the organization, from content production to, you know, video editing, a lot to do with AI, obviously. And I think as marketers, we we owe to, you know, serve that trend and ride the wave. Yeah. So can you share a little bit about the broader strategy or business objectives that you guys set when it came to adopting agents and thinking about where you could use them in the team? Yeah, absolutely. Well, I was just, we're just chatting before about, you know, how LLMs and AI agents have been improving efficiency. So I think it's really with that mindset that we're trying to leverage this new technology. We've realized that what humans would need, you know, a couple of hours, to do what agents can do in a matter of minutes. So, yeah, I think for us it's really about saving time, doing things faster, and, you know, automating a lot of the most repetitive tasks so that our human can actually focus on the areas where they add the most value, which is essentially, you know, connecting with our prospects, connecting with our customers, building good connections and and serving them. And so how do you expect them to ultimately influence growth or efficiency in the go-to-market? Like what are the sort of things that you're looking for? Yeah, I mean, again, it's really about automating a lot of different tasks. I think going deeper, faster in terms of the research, in terms of the communication as well. One of the things that Lightspeed has done really well from a leadership perspective is thinking about kind of how to build a structure of the organization to safely adopt agents at scale. Can you share a little bit of that insight, kind of some of the steps that you took and your team took about thinking about the structure of team and how to really effectively integrate agents into that team? I think very quickly we realized that implementing such tools would require a lot of collaboration from a lot of different teams. So we in the marketing, in the go-to-market structure, we kind of identify different use cases across the org of things we could automate, things we could improve, things we could leverage this technology on. So there was kind of putting together the business requirements, the business objective, the KPIs. The business case lies behind those different use cases. And then once we had identified those different business, those use cases, sorry, it was really a matter of prioritizing them quite ruthlessly. And then once we figured out the different steps we needed to take, it was about rallying all the different teams behind us. So obviously, you know, there was a lot of tech work needed to connect the agents with our different systems that we use internally. We also had to go through security teams to make sure that all the connections were compliant. We needed to work with our content team to make sure that whatever would be produced by the AI agents would fit our brand guidelines and then had to review this with comms. So it was really a big, big team effort and to do all of this. So we kind of built within the marketing org an AI pod, which takes essentially resources from all those different teams. So we now have subject matter experts from comms, subject matter experts from content, subject matter experts from, or MAT, uh, automation team, uh, subject matter experts from IS, um, and all these teams are now kind of working, um, together as a, as a squad, um, to, uh, build our, our use cases and drive the value. That's amazing. And I guess someone who might be early on their journey now thinking about doing this within their business, how would you advise them to do that stakeholder alignment that you mentioned? How do you think they can best communicate kind of the vision? Or the impact that they want to create? It's interesting because I think at first the main challenge we had was to manage everybody's excitement and expectations in the sense that I think, you know, when all the LLMs came out like a couple of years ago, a lot of people quickly realized, you know, the power that they would have. And so at first it was like, as I mentioned, really about like ruthlessly prioritizing the different trials that we needed to to work on, we had to essentially prioritize. We were not going to be able to tackle them all at the same time. And then once we had that alignment done with the main stakeholders, then it was about, you know, going in full execution mode and, and yeah, building from the ground up. Obviously, as we built those agents, we learned a lot throughout the whole process. We've been supported very well by your teams as well. It was a learning process, you know, um, the, the, the person, Miles, who was, uh, hands-on into the, the, the, the tools, obviously building the tool for, for the first time. So I think it was a, a big learning curve for everybody, but a very exciting one to, um, to take. So you're performance marketing. Yeah. Performance marketing is all about measurement. When you think about agents and actually measuring what good looks like, what are some of the metrics that you typically advise people to think about in order to succeed? Yeah, well, I'll say it depends on your use case, right? For us, the first use case we've built was really about, you know, generating pipelines for, for a sales team. So based on that objective, you set your KPI. I think it's important that here we're not necessarily, we're just trying to essentially supercharge a process that were, that was already in place, right? So set that KPI, but also set the benchmark against which you're going to be measuring the effectiveness of the tool at the end of the day. Then you're going to need to set secondary metrics to essentially have a an understanding of how you're trending versus the overall KPI. There's a lot of talk about AI and trust and making sure it's doing the right thing. What are some of the practical tips that you might have to kind of build confidence and make sure that the team, you know, one, trust the process that you guys are having, but two, also trust the output? Yeah, so trusting the process requires a lot of, you know, again, stakeholders alignment, talking with security teams, talking with your systems team, uh, just making sure that, you know, you only give the right amount of privilege of rights to, to the tool to achieve what, what, what you want. And then in terms of output, I think it's really about, you know, rolling out a structured approach, right? It's kind of a, a crawl, walk, run approach where first you're going to roll out the agent at the very beginning, small volume. You're going to be essentially reviewing every output. And then from those early reviews, you're going to be able to tweak on the model, tweak on the process to improve it, optimize it., and then it's, and then it's, you're gonna just essentially iterate until you're able to, to run on autopilot. And even on autopilot, I'd say you'll still need a bit of monitoring from time to time to make sure that, you know, the output meets your quality standards. So one of the things I think you guys did also really well was you ran a phased rollout. You had approval gates, you had gradual scaling. What were some of the key learnings from that approach, good or bad? Yeah, I think the key learning is that, uh, things usually don't go as planned. We usually, well, we usually, we have run into some roadblocks. So, you know, from very early on, we were optimistic and we, you know, wanted to go through this like review mode, like relatively fast and then scale and finally go on autopilot. We have actually run into roadblocks where we realized that the output was not good quality enough. We've also realized that some of the data we were feeding into the agent was not the right one. So we had to tweak a bit that piece. So yeah, I'd say, you know, take your time. In many instances, we favored, you know, quality over rapidity, despite, you know, some stakeholders wanted to like hold these use cases as fast as possible. And I think it was the right choice. So yeah, in terms of learnings, take your time and don't expect that things will go smooth because it can be quite tumultuous, right? And maybe a little bit of a kind of questions peering into the future. How do you think agents are going to interact with go-to-market teams in the future? What role do you think they can play to help improve that efficiency, supercharge teams, and really help teams deliver the outcomes that they have and the priorities they have? Yeah, I think the agents will essentially be their assistants, you know, in terms of doing research about their prospect, in terms of writing up comms that then the salesperson can follow up on in terms of, yeah, giving all the tools to those go-to-market teams to do better. And again, to spend more time where they add the most value, which is on the phone, talking, connecting with the customers and the prospects. Well, Rémi, thanks so much for hosting us here today. It's been beautiful to visit the the Lightspeed offices, and thank you for the insightful discussion. My pleasure, Daniel.

Thanks for coming here.

Thanks for having me here at your wonderful office.

Well, could I get a quick intro from you, Levi? So I'm Levi Waters. I'm the partner in charge of AI solutions for KPMG. And my role is really twofold. One is to transform the way that we deliver our services through the UI, through AI. And that can be being product-led or the services we deliver in market, but then also re-engineering the way that we deliver our services today to be more efficient and higher quality through the use of AI.

Awesome. So at the leadership level, what AI themes are dominating Australian enterprises this year?

Is it efficiency, growth, risk, or all of the above? I think risk is always going to be a key concern. And I think there's been some issues that have probably highlighted that at the moment. But I definitely have seen the emergence of return on investment to be the most dominant theme. So organisations have probably been through the proof of concepts. How do they get from proof of concept to scale and then see that return on investment? I know that there's studies out there that have sort of claiming that the these proof of concepts and, you know, and a lot of failure. We are not seeing that and I don't believe in that sort of rhetoric. We are seeing a lot of benefit come through and seeing a lot of scale coming through and return on investment for what organizations are chasing after.

I understand and I totally agree.

When leaders evaluate AI and a Gentec solution, what are some of the biggest concerns you hear? The main concerns initially are really around security of data. And that's very balanced and the right thing to be focused on. And then it's about people and adoption. And I think the cultural story and the adoption and taking our people on the change management journey is a really important area to address. And I do acknowledge that there's a lot of concern around that with a lot of organizations. I think most people are getting to the point where they understand that in the future, all of our business is going to be based around AI and everything that we do will be in some part supported by that.

100%. From your experience, what separates companies that successfully implement AI projects from those that stall after the pilot phase or even the ideation phase?

It still is early days. And I think to acknowledge that, particularly at sort of enterprise scale, but the organizations that have done the pilots, possibly fail, but then got the successful ones and looking to scale, are starting to understand what are the foundations that you need to be successful, which I think will be that hockey stick in the future. So getting data into the right, obviously data is critical, getting that into the right format so that can be used, exposing systems via micro APIs so that we can actually connect agents and other things to it. That's the foundation work that always kind of existed in digital transformation, Those organizations that have started and then focus on that will really see the big unlock when they start to scale across the enterprise.

Yeah, it makes sense. And even just like putting down what is the current process of doing things so that you can give that same instruction and process to AI so that it can understand and do it.

And we've seen, I mean, to keep building on that, that's been probably the really important part is Yes, get started, learn. And then we talk about context now and then providing AI with context and having standardized patterns. We've seen massive improvement in that, in standardizing the way that we and what we use relevance for or what we use our other tools for. And so that we can, because we've done that and we've trained the AI in the way that we work, the onboarding process for new people is now hours rather than weeks.

So it's a big productivity shift in relation to that. Definitely. Very impressive. How important is vendor-agnostic AI tooling to exec buyers and yourself?

Look, I think that that's probably one of the key fears that a lot of organizations have is that either they get locked into one paradigm And because this is moving so quickly, one LLM will be ahead in this area, then next month, and it changes so quickly. There is a bit of, do I just wait before I get started and wait for this to settle down? I don't think it'll ever settle down. So there's that bit. I think the other part around the vendor lock-in is that there is a fear of having lots of niche AI tooling that services this one piece. And then you get this sort of proliferation of licensing and then the nightmare across that. So I think the ability to be vendor agnostic and the ability to scale across a breadth of function is actually a really key feature of what you've got. And that sort of, I don't want to be locked in and I don't want to have hundreds of these tools that do this, I've got to pay for this tool to do this bit and this tool to do that bit. That's definitely a blocker for an executive.

Executive mindset. Yeah, the sprawl of all these toolings. Yeah, that's why I guess Relevance AI, we decided to go horizontal and not vendor locked. What factors are most important to a successful AI implementation from your perspective?

For me, I think, yeah, my perspective now, my opinion is that 90% of what we want to do with agents, the technology is already there. And so The technical aspect is probably there. It's how do we actually get adoption within the business and the change management of that? And so we really need to think about how do we change the work, the way we work with agents embedded or augmented with agents, and then look at how do we capture those efficiencies through changing the process? And so we've been guilty of it where we ran out, did a pilot, the technology, the classic of the operation was successful, but the patient was dead. We had a really successful pilot, everyone think, but no one used it. And so how do you actually get that? And so we focus a lot now on embedding and changing the processes and the operating model so that we realise the benefit of adoption and start to really measure adoption and usage statistics, capturing that feedback of eating your own dog food and the other parts around how does that work to then feed that back and continue to improve. And I think once you get on that piece, that's the most important part about actually realizing the benefit is adoption.

Makes sense. So yeah, adoption is really the kind of key to making sure that these AI rollouts are successful.

Yeah.

You kind of mentioned a little bit of the different kind of tactics, like things like dogfooding, et cetera. But how does KPMG specifically like to drive and achieve this success with clients?

Our key element is to really understand what's the pain point and it's focus on that, but think about it from an end-to-end perspective. And I think again, related to the relevant story, but you only really point solutions or solving this pain point will only get you so far. You need to have the vision about how is this actually going to reinvent the way that we work. And when you do that, that's when you get real benefit. And so that's the way that we approach it. So yes, here's a pain point in an area. I can create an agent to do a task that makes it more efficiently. But what we really focus on is actually how do I reinvent this to be AI at the centre and drive that from an end-to-end perspective. And that will completely change the way we work.

And that's where you get the big unlock. Yeah, 100%, completely agree. Outside of the AI tool or platform, what are some of the critical things for companies to get right from a change management perspective to make sure that, you know, first, as we kind of talked about, the implementation is successful, but two, that their teams on the ground actually adopting, actually, you know, starting to trust Yeah, the first part is data.

And in our early— and we've been on this journey for a couple of years now, we were very lucky to have access to secure Azure OpenAI in what, February '24. So we've been doing this for a while, but some of our early forays where we were doing this with our clients, they didn't have the data for us to then be able to review and uplift. And so a lot of the effort, there was just not going to see the benefit because they didn't have the data ready or the documentation ready for us to review and do that. So that really is the most important part. And just to kick off, people then go, oh, we won't start until we've got the data ready. What we learn as well is that actually use the AI to get the data ready. And so we then started on the first bit that we do is use agents to do a data diagnostic to then, okay, these are the gaps that you've got. These are the 3 priority areas that you need to fill from a data perspective and then use the tools to actually create that or harvest that to get it in the right state. And so that really is the first part about getting started is understanding the current landscape of your data maturity and then being able to get going. Again, I come back to the adoption bit. It's It's about really understanding how people work and going through the adoption process with them so they understand how they should use it in making it in the flow of work, not something additional to what they do. And so that change is how we actually lead to adoption so that people can actually see the real benefit of augmenting their work with AI.

100%.

And I love that bit about essentially using AI to accelerate AI adoption. Which businesses, teams, or functions do you think have the most potential to leverage AI agents and why?

A lot of our focus is in what we would call back office and knowledge work. And so back office, it might be finance teams, risk and compliance. A lot of our work is in that space. Procurement, a lot of these sort of corporate functions. I know that a lot of organizations are really out there focused on call centres and other things. But I think that the big unlock is actually in operations. And so that's where we are focusing a lot of our effort into sort of pulling that through.

That makes sense. Yeah. What governance structures or operating models have you seen work best to manage AI responsibly at scale, especially across different business units or regions?

Obviously, KPMG, we're a big area of governance. So we've got a trusted AI approach to governance and I think that that obviously is important and having the right guardrails as well as responsible and ethical application of AI committees and governance forums. The bit that I and we're really focused on is actually the monitoring of the output. And I think a lot of organizations are really not doing that well. And so how do you actually monitor that the agent output is aligned with the intent of what you wanted it to do? And so that's where we are putting a lot of effort. So yes, we're KPMG, we do governance, we're pretty good at that. But our emphasis actually, the real difference is monitoring and the opportunity that AI itself unlocks in being able to do that.

Yeah, amazing. And that that monitoring and that governance bit is very, very crucial to essentially keep gaining the trust of these teams, these enterprises.

Yeah. And I think with that, just it's the transparency of that so that senior leaders can be confident that it's being used in the right way, that the output is that, and that comes from visibility and transparency around it.

You're right.

Yeah.

And I think you guys release a yearly report report on some of this as well, right? Yeah, we did a lot of work with the University of Melbourne around trust in AI and people's attitudes toward it. It's a really good paper, I recommend it. But, and Nicole Gillespie or Professor Nicole Gillespie sort of led that with us and there's a pretty good framework in that around what you need to be doing to make sure that you've got that trust right.

So it's good. Makes sense. AI is developing so quickly now. Yeah. Are you seeing a gap emerging between the companies that are investing in AI and those that are hesitant to do so?

Yeah, I actually had a conversation with a client this week who said they're 2 years away from Copilot. I almost fell over considering I was going to talk to them about AI. And so you've got a lot of organizations that are really, I guess, waiting for this to fall. They're risk adverse or whatever. I genuinely think that that's that's not the right approach. And that those organizations that are on the adoption curve are really going to accept the level of acceleration that we see with the change and the efficiencies and the rest of it. They're going to accelerate much faster because it's always just going to keep getting better. So they're getting better at an accelerated rate, whereas you're not adopting AI, you're staying the same. And so you are really the level of acceleration between you're going to get left behind by is going to be massive.

Yeah, it's a real competitive advantage. Yeah. The leaders should really take hold of. Which brings me to my next question. What would you say to any leaders that are watching this that haven't started leveraging AI in their teams?

Just get started. I think there's a lot of examples out there where the way that you get started, start using it yourself, right? Use it in your personal life, then think about how do you actually bring that in? As a leader, actually promote, I use AI to do this. So I am very overt and I learned this off Kim Crow Anderson at Telstra and I've had the conversation with him and I think some people still have a fear about it's cheating. It was sort of the thing and certainly see this in our business. So I'm now really overt. I do a lot of my presentations with HTML I'll just create that through chat and I'll tell people and I'm much more overt about when I'm using AI and so that people feel more comfortable and do it in a safe way, but so that people feel more comfortable about using it and then sharing about how they use it and then you start to get that adoption curve. Yeah, I love that. And that's what we're doing with relevance. So we've had a bunch of sessions where And the bit that I really wanted to sort of put into this is where we're using Relevance to actually help train leaders in what an agent is. And so by having this sort of capability where they can create agents themselves, that was a big unlock that they went, oh, now I know what we're talking about. And it was because you've got the building blocks behind it to actually show how this works and that workflow that people had the sort of aha moment around how this works. And so it's just getting started. It's about educating that. And then once you then— we created this prototype, which created that prototype. We then pulled that into production and we actually started to really accelerate.

You get that hockey stick effect that we're looking for. I understand. And I love the fact that you're showing this adoption via leading as an example and even just starting from the slides and PowerPoints. Side of things. Looking ahead, how important is it for leaders to leverage AI today? How big do you expect this gap to continue to grow? And what do you feel like the future of AI is going to look like in the next year or two?

Yeah, well, in the next year or two, that's probably maybe easier to predict. I think that, I don't know where we go to from there, but I think that there will be, you know, the emphasis on return on investment. And I think the proliferation of use cases will continue to grow. Obviously coding has been a big focus for AI in general now about using AI for code. But what we're seeing certainly is that that application also applies to everything. So context and business analysts and all the rest of it are really starting to use that more broadly. And so as I've said probably a few times, a few too many times now, It is that network effect of adoption is what we're really chasing here and obviously what we're trying to get going with our clients. And that will rapidly increase the gap between the haves and the have-nots.

Love that network effect of adoption.

Yeah.

Awesome.

Well, thank you so much for your time and thanks once again for having us at your lovely office.

Thank you so much.

No worries. Thank you. Well, Jess, thank you so much for taking the time. It's good to see you again. We've obviously worked together at Asana, so it's really good to reconnect now that Relevance and Canva are working very closely together. So I'll start from the beginning. Canva was an early company to embrace AI across the workforce. So when you all first started shaping the AI strategy for the go-to-market team, what was your initial approach and what did you want to achieve at a high level? Yeah, I mean, we were really lucky at a company level that Canva was very much at the forefront of empowering the workforce with AI. And so at a company level, in addition to making sure that everyone had the tool of their choice, we're very lucky. We had a lot of initiatives to try and drive AI fluency more generally. And so they were making sure that everyone had the tool, really showcasing kind of education and then shared use cases as well. So we kind of had that behind us, which was great. And then for go-to-market specifically, we really focused on intentionally building a roadmap to find ways that AI could drive efficiency within our team. And the goal there was really about looking for processes that could be manual processes that could be automated with the goal to free up time so people could spend more time with their customers rather than less. Sure, absolutely. And how, how has that journey been been for you all so far? It's been great. I mean, we definitely had an early burst like everyone else where every individual was kind of pioneering their own workflows and their own prompt-based platforms. But since then, I think we've done a really good job of picking out a few use cases and workflows that we really wanted to scale at high quality across the entire team and have been delivering on that roadmap. Amazing. That's fantastic. And then how did you go about identifying the most promising use cases to get started with, and were they driven more by your team's pain points or just by exploring new AI capabilities? I think we definitely did not start with tools first, right? I think that's a common pitfall or trap that you could fall in, especially with a bunch of really exciting new technologies out in the wild. But I do think that you run the risk of adopting tools for tools' sake there. And so what we actually started with was where does our business need lie And for us, that was okay from a business lens, which is definitely what my team brought. Where do we feel like we're— we've got gaps, right? Where do we feel like we could be doing more? And that was led through data, etc. The second thing that we did was also shadow our workforce and hear the kind of pain points and the feedback from folks on the ground, our account executives, our BDRs and our CSMs about parts of their day that they really felt like they could use help and automation as well.. And so the combination of both those lenses, we built a roadmap together and we really look for things that, you know, where AI could transform or improve today and not at some point in the distant future. And everything had to be backed by a really strong ROI. So is this something that could free up time for our reps so they could spend more time with customers? Is this something that could help us close business faster? Is this something that could help us onboard our customers faster? And so making sure we, you know, took those inputs and backed it up with really clear business cases helped us focus on the few things that we could do really, really well. That's fantastic. And as a follow-up, did you already have some ROI metrics in mind when you kickstarted the process? I think so. I mean, it's in the same way that, you know, we think about annual planning and kind of all that typical stuff. It's like, what are the input metrics that drive the things that we care about, right? Which is more value to customers. And so we use the same framework. We didn't invent a new one for AI, I think, to prioritize where we want to spend time. Yeah, it makes sense. It keeps it consistent right across everything. There's obviously so much potential impact that AI can have at the moment, and so it could be hard to know where to start. How do you think about balancing the quick wins that build momentum versus longer-term bets that might be transformative but really take some time to deliver value? Yeah, I think the approach we're taking is to do both. Quick wins initially to help us understand what it's like to work with vendors, what we can, we can kind of get from our own tech stack as well, and to build that credibility and impact within the field. In parallel, they're building a really intentional roadmap about the bigger, more transformative projects that are really going to move the needle for us, but may take more time and detail implementation, and treating those like almost like product launches in and of themselves with clear roadmaps and making sure we're bringing, you know, all the people involved in putting that together into a clear project plan. Yeah, makes sense. Okay. And then obviously we're seeing a shift from standalone AI tools towards integrated AI platforms that agents can, where agents can actually handle multi-stepped workflows. So how do you see the role of AI agents or platforms fitting into Canva's go-to-market AI strategy compared to more narrow point solutions? Yeah, I mean, we see a big role, right? And which is why we're so excited to be partnering with relevance. I think the thing that we're balancing in addition to AI experimentation is just tool proliferation in general, especially within B2B SaaS, but more so within like the workflows that our folks on the ground are really having to navigate every day. And so while we do want to drive experimentation on the AI front, velocity, like rapid iteration, at the same time, we're really, really loath to adding more tools, more surface area for our reps. We're already you know, in a situation where, you know, you yourself as a customer-facing person, like 7 to 8 different platforms. And so we're really excited about platforms like Relevance because we can think about how to bring AI automation into existing workflows, into existing surface area, whether that's surfacing them in Slack, surfacing them in our CRM in Salesforce as well. I think we can harness a lot of that power behind the scenes without adding them to the kind of rep interface, which, which I think is some of the failure states, especially with some of the UI-driven point solutions as well. It's going to be hard for us to create that space for them when it's already so noisy. Yeah, that makes sense. I mean, I'm curious, like how did you actually drum up excitement for these go-to-market teams that are exposed to just a whole host of platforms on a day-to-day basis? It's the quick wins that we talked about previously, right? Like how can we pick up a couple of these that maybe are not super complex, maybe are not going to have a huge ROI, but actually like help Help our reps save 15 minutes every day prepping for a call with a new prospect or write that follow-up email. They seem basic, but really high impact, and then it kind of builds that momentum and appetite for more. Yeah, I love that. That's fantastic. And then you recently created a dedicated go-to-market AI lead role within the team, which a lot of companies are actually following suit on that we're observing. What prompted you to create the role and what's their sort of central objective? Yeah, I love that. I think what what prompted it was really the kind of state we were in before, which was a lot of appetite, a lot of different, you know, the roadmap was kind of evolving, but really, really hard to make concerted impact without someone whose sole job was to get through all the blockers, block and tackling basically, and have this sole focus. Because super easy to prototype, super easy to deliver quick wins, low complexity. But if you're looking to scale something that applies across a workforce, force of folks in the go-to-market org and applies to a number of different situations. I think that takes a lot of, you know, lots of decision-making on the business front, but then also just lots of integrations and lots of problem-solving as well. It's really, really hard for people to do that without, you know, it being their sole job. And so that's what we kind of made a decision to do. And Robert's here behind the camera as well. And it's really paid off. Like, I think we've been able to not only ship faster, but also ship more complex workflows driven by AI. As well beyond just the quick wins that, you know, other companies are also going towards. Yeah, I love that. Can you dig a little bit deeper into what impact you've actually seen since you've had this role for a while now in having this dedicated resource? Any examples you can go into there? Yeah, I mean, one, one obvious one is we've done a lot of work on the enrichment front. I think with AI, it's just a new world of possibilities around not only basics, right, basic data hygiene, basic, you know, the things things that help with territory carving and territory scoring. But I think more importantly, how do you build account database in a CRM that really reflects your very specific go-to-market, right? Edge cases and data sources and way of identifying accounts that don't exist with off-the-shelf enrichment vendors. So that's been one big project that we've led through a set of new solutions. And I think it's really amazing. Like we've got new creative license, I think, to shape our go-to-market market in a very nuanced and specific to Canva way and to kind of focus our sales and success teams in that direction. That wasn't possible, you know, 12 to 24 months ago. So I think that's been one of our big projects. Yeah, you actually led me into the next question, which is Canva actually has a very strong culture of innovation and experimentation. So how much of that culture contributed to your success with AI, just relevance and beyond? And what can other go-to-market leaders essentially do foster a similar mindset on their teams? I mean, AI has infinite potential, right? Tools like Relevance, it's really up to your own imagination and creativity about how you want to harness this technology to achieve the outcomes that you want to for your team and your customers. I think like maybe that's why we've been so successful is like we have a strong bent for the creative and the art of the possible. You have to bring that because tools like Relevance are infinitely configurable and you you can infinitely integrateable as well. And so without a very clear set of inspiration and perspectives on what are you actually trying to do, I think you're only harnessing 1% of the power of a tool like Relevance and other vendors out there as well. Yeah, 100%. Yeah, 100%. Awesome. Well, thank you for taking the time and sharing your experience. Before we wrap, I'd love to finish and ask what advice similarly would you give to other leaders that are eager to leverage AI in their teams but just really don't know where to start? Yeah, great question. I think starting with building a roadmap like you would any other product development cycle, starting with your customer journey, what you're trying to solve for and fix, or where you're trying to accelerate, and working back from there, and then finding the tools and technologies that can apply, as opposed to starting with tools and technology and the vendors that come knocking. Because I think there's so much you can do. Implementing things in a really scalable and sustainable way is not easy as well. It takes a lot of intentionality. So picking the few things, and that's a combination of quick wins and more transformative projects, and building a roadmap and treating it with that kind of the respect that you'd give a full development cycle, I think is how you can actually get the impact going. And it should be a pretty, pretty straightforward process of starting with your business objectives and working back from there. Awesome. Thank you.

Appreciate it.

Yeah, great to spend time with you.

Hi, here with Miles from Lightspeed. Really excited to have a great conversation about your AI strategy and the work that you've been doing with Relevance. Nice to be here. Let's get into it. As Lightspeed Commerce has implemented its AI strategy with Relevance, you and I have had a number of great conversations, a number of great lunches here in Melbourne. It'd be great for you to kind of share some of that, you know, with the audience today and some of the learnings that you've you've had along the way.

Maybe to start off, can you tell us a little bit about Lightspeed as a business? So Lightspeed began in 2005 as a Montreal point-of-sale provider. Two decades later, it has evolved into a unified commerce point-of-sale and payments platform. More than likely, your local coffee shop or restaurant or retailer is using Lightspeed, and I've been with Lightspeed for about 3 years, and it's a great company to work for.

Awesome. As a senior AI marketing technologist, you've played a central role in deploying agents at Lightspeed. How did you first get involved with the Relevance project and initiative?

So I've been working in marketing for probably around 8 years. My previous role was senior growth marketing manager in APAC, handling our paid media strategy and implementation. Prior to ChatGPT, I think ChatGPT came out about 3 years ago. I was messing around with machine learning, sort of like predictive models and all that kind of stuff, and I remember when ChatGPT was released and you could kind of chat to a large language model, that was kind of a turning point for me, and I shifted a lot of my focus into AI and how I could apply AI to performance marketing essentially. So I spent a lot of time experimenting with using AI for paid media use cases, so landing page creation, ad creation, analytics, and I was, yeah, really sort of blown away by what it could do. Definitely a transformative moment, I think. Yeah, 100%. And as far as getting involved for Lightspeed, as a result of, I guess, sharing some of these sort of AI experiments I was doing, Remy and Richard, who I work closely with in North America, asked me to come on board to, I guess, co-lead the Relevance AI implementation at Lightspeed with Leslie and some other people from the marketing AI pod.

Fantastic. Well, great to have you involved in the project.

It's been a super fun journey so far.

Working in marketing, you know, you're a domain expert, but you've also developed a lot of kind of new technical skills, as you talked about there, you know, since this kind of ChatGPT moment. Can you tell us maybe a little bit about that learning curve, like where you started, how you progressed to the point you're at now?

Yeah, it's a good question. I think over the last 6 months that we've been working closely together, it's been a massive learning curve. I think I started off not really understanding what AI agents are. It's confusing in the space where a lot of SaaS companies and a lot of companies generally are saying they're deploying AI agents when often they're deploying AI workflows. There's been a lot of education and a lot of experimentation. I think the AIOps course has really helped and working closely with you, Michael, Ishika, Tony, and the team have helped us understand internally what the difference is and where to apply either AI agents or AI workflows. I guess workflows are deterministic and AI agents are autonomous systems is probably the easiest way to think of it. As far as learning goes, as I said, I got kind of obsessed with AI a few years ago, reading a lot, watching a lot of YouTube videos. There's a guy called Nate Hurk who's pretty good, he does a lot of videos. I did a course by a guy called Marcel from Growth X. It's really good working at Lightspeed, they kind of support stretch projects and development as well. So lucky to be in that position where I'm able to experiment and everyone's very encouraging and very open to new things, which is super cool. Generally, I think of learning in two phases: explore and exploit. I think As I've been learning over the last 6 months and over the last couple of years of learning more about AI and how to apply it in businesses, doing a lot of research, understanding the lay of the land, understanding what is actually maybe just marketing and what is actually a genuine different thing within this field. So pulling in a bunch of knowledge, kind of understanding how it all works together and then applying it is generally the way that I like to do things. I think if you're trying to do both at once, I find it overwhelming.

Yeah, fantastic, thanks Miles.

I think that's a, you know, really great, great answer. Definitely, you know, since we started working together, the space evolved so quickly, you know, so even from a kind of a relevance perspective, our platform's coming on, so that, that kind of, you know, ability to be both a self-starter and have kind of the buy-in from the business, you know, is fantastic and it's great to see. So our co-founder Daniel Recently went to the Lightspeed headquarters in Montreal to speak with Remy, the Director of Performance Marketing, and Remy explained a kind of a very specific approach that you guys are taking at Lightspeed, which is the crawl, walk, run approach. Could you share a little bit about how this works in practice and how you kind of applied that here at Lightspeed?

Yeah, so crawl, walk, run is a great way to describe the strategy. I think building agents within enterprise has a lot of unique challenges that we've successfully navigated over the last few months. So our main agent at the moment is a closed-lost research and outreach agent, which hands over to human reps to speak to the prospects. So Lightspeed has hundreds of sales reps globally and also an amazing and diverse tech stack. With this, I think, comes a lot of complexity. The agent has to fit in with human processes and also automated systems based on triggers. On top of that, it has to work through complex data where it doesn't always have full context. So in ramping up the agent slowly, we were able to identify, document, and ultimately fix issues both with internal processes and systems. We're also, by ramping up slowly, we were also able to identify the failure modes within the agent, which we then sort of adapted and fixed on the fly by building out tools and also realigning our processes internally.

Fantastic.

So quite an iterative approach to agent building, you would say?

Yeah, a lot of iterations, I think.

Yeah.

So, you know, one of the keys to success success that we've definitely seen across a lot of customers is the ability to kind of embed that domain expertise into the agent.

What processes did you use to make sure agents truly reflected the Lightspeed way of working? Yeah, so in rolling out Alex, which is the name of our main agent that we're running at the moment, we work closely with subject matter experts from the sales team, marketing team, technical teams, and also security, privacy, and legal. We worked collaboratively to design an agent that could fit really neatly within the Lightspeed ecosystem. We worked really closely with the sales team. We have a lot of amazing sales team members at Lightspeed. They've got deep product knowledge, industry expertise, and also understanding of sales strategies and what works at Lightspeed. So we identified a number of champions who were the experts within their particular niche and interviewed them to understand how they work, what processes they follow, and ultimately why they're the top performers within their sort of niche within Lightspeed Sales. After we went through that interview process, we worked with the relevance team, Michael and Ishika namely, to create a process map for both like what is the current human process and then what is what the agent should look like. Once we mapped that out, we got kind of sign-off from all the main stakeholders, and that served as the source of truth for the agent build, and aligning on that, I think, was super beneficial.

Fantastic, great to hear that you named your agent.

That's definitely something we encourage from our customers, you know, to humanize them a little bit, and fantastic to hear about how you're bringing together both that, you know, real domain expertise, but then also imagining kind of the art of the possible, you know, what can we do with agents maybe beyond what we could potentially do, you know, without agents, is fantastic to hear. So yeah, moving on, obviously security, data governance, we've talked a little bit about the role of IT and IS within the project, but this kind of ability to deploy agents safely, securely, is, you know, obviously a priority for Lightspeed. As it is a priority for kind of a lot of our enterprise customers. So how do you think about that?

How do you bake those safeguards into your agent workflows? Good question. So as we just spoke about before, we, we followed a crawl, walk, run approach. We relied really heavily on Relevance's human-in-the-loop approvals functionality, both through the platform and through Slack. It's very cool being able to read through through agent runs and approve them while on the move, so found that super useful. Given the fact that we were reaching out to customers with this agent, or reaching out to potential customers with this agent, we wanted to heavily approve what was going out. So initially we approved all the research reports, then we approved all the emails, and then we approved all the email replies until we were fully comfortable that the agent was performing exactly how we wanted to, we didn't put it on autopilot. So that process of continuous iteration and refinement was really important. In some cases, it was me in the platform refining the output if it was a small change. If it was a change related to maybe strategy or bigger changes, we'd open it up to a wider group who fed into this this iteration.

So it sounds like, you know, quite obviously an ambitious agent project, but also a high-risk one, right, where you're interacting directly with customers. You find that that process, that iterative Core Work 1 process, really helped build trust with the wider stakeholders inside Lightspeed?

Yeah, definitely. I think these projects are very new. I think everyone's aware that the opportunity is massive, but also the risk is massive of keeping everyone up to date internally is very important, and obviously getting the right people in the room initially with this kind of project. So with security, privacy, and legal concerns, we worked really closely with our internal teams on that at Lightspeed. Lightspeed's a very AI-forward company. Everyone's kind of excited about this technology, and it makes it a lot easier to work Everyone makes it easy to collaborate on these projects, which is great.

Yeah, fantastic. That's great to hear. But obviously, change management is always a big piece of these types of projects. So it's super exciting to hear those insights there, Miles. So let's fast forward to today, and, you know, I'd love to hear a little bit more about your kind of current focus.

Are you thinking about building new agents?

Are you refining, scaling Alex? You know, what's the strategy from here?

I'd love to hear a little bit more about that. So now that Alyx is sort of scaling and proving valuable, we're looking to both enhance what Alyx can do as far as the channels it can use, and also other use cases. For instance, we're decoupling, I guess pulling out the research component of Alyx, and going to provide that as a resource to the sales sales team, and we're also looking at expanding the use of the platform to other departments in a pilot sort of project.

Fantastic, great to hear. And you know, you're really at this very exciting point where there's reusable components of Alex which you can apply to other agents to kind of give you a head start. Plus you have a lot of integration now kind of in the platform, so kind of the opportunities ahead are, you know, are almost endless, so it's a very exciting time.

Maybe it would be—

I'd love to hear kind of how you measure success.

So you mentioned Alex is kind of, you know, now delivering results, like what does that mean?

How do you think about that at Lightspeed? Yeah, it's a very good question. I think we, the growth marketing team at Lightspeed, is performance-focused, so our goals are tied to revenue. I guess like learning, framing performance in a new way has been something that we've had to do over the last few months. For instance, the labor productivity kind of metrics are new to us as a performance team, so that's something that we're getting used to reporting on and monitoring. We're mixing in a combination of labor productivity and agent-based metrics, I guess, as well as the traditional performance marketing metrics. We've had to build out custom reporting as well, which has all been part of the agent iteration process, and now we're in a place where we can report on the value of these initiatives, which is obviously going to help us scale to other use cases as well, using the same sort of frameworks that we've built up over the last few months.

Yeah, fantastic. And I guess Finally, you know, looking to the future, what role do you think AI agents are going to play in performance marketing at Lightspeed and maybe even kind of more broadly?

Yeah, so I feel like the way that we're heading is, I imagine like a near future where we've got ambient agents working in the backgrounds. I think these agents will be researching compiling information on our prospects, doing outreach, and then ultimately handing over to the human sales team to do what they do best, which is speaking to the customers and really understanding their business. I see it as a collaboration between the AI agents and our amazing sales team, so we can essentially just automate some of the boring stuff and really help them focus focus on the higher-value human stuff. Our sales reps and also the growth marketing team will be interacting with various agents on Relevance through both Slack and also the Relevance chat interface. There are so many different use cases across SEO, I think paid media, CRO, just anything we can think of. I think kind of the sky's the limit. I think we've proven proven value, proven that we can integrate it into our human and tech processes and systems.

So now it's about sort of scaling horizontally. Yeah, fantastic.

Thank you so much, Myles. Really appreciate your insights. No worries. I'm joined here today with Lachlan, Director of the AI Lab at KPMG, here in your beautiful Melbourne office. Thank you for having us. No problems. KPMG is really uniquely positioned because it's not only implementing its own AI strategy internally, but it's also helping shape and roll out AI strategies for some of the largest companies in the world. So to start us off, Lachy, Can you please tell us a bit about who you are, your role, and the work your team drives across KPMG?

Yeah, great. So yeah, I'm Lachlan Hardesty. I lead the AI lab in a division called KPMG Futures, and where we focus is on the cutting edge and research and development around AI. And so for what KPMG, for what that means for us is that my team and myself essentially go into clients piloting and trying to prototype AI agents very quickly and rapidly to try to uncover where there are opportunities for us to then grow and move forward.

Amazing. KPMG started using Relevance to build agents for their internal teams over a year ago. What led you to try out AI agents internally? This was, you know, a little bit before the craze as well.

And what outcomes were you hoping to see? Yeah, so we started out trialing agents through some of the open-source frameworks. So Crew AI, and then moved on to things like Autogen when that was emerging. And really what we found was that whilst those had great capabilities, great depth of capability, they took too long to actually service the client need that we had in front of us, which was clients don't want to wait multiple weeks or months to work out whether something's going to have a commercial return. And so that's where Relevance came in. And we discovered that, you know, you could get really rapid essentially outcomes and solutions using Relevance that then really served that client need quickly, got to a point where we could assess it commercially and understand whether it had, I guess, relevance to a client. And that's where we really saw kind of that golden ticket for us with Relevance was when we started trialing it on clients, seeing how quickly we could get to those outcomes.

And actually saying, hey, this has commercial viability for you and then taking it forward from there. Awesome. Amazing.

What were some of the first agents the team built and what was the impact? Yeah, so the very first one that we built in relevance that was deployed. So part of what my team does as well is we spin in mini startups within KPMG and essentially try to foster those and try to see whether they can grow into businesses. Processes for us. And we had a business that was essentially ideated out of a hackathon that wanted to do healthcare, sorry, aged care scheduling. And so we used Relevance to essentially build that solution end to end and deliver the pilots into those home care clients. And because Relevance was very, you know, low code to be able to do that, we were able to deliver that solution really seamlessly and actually deliver 6 pilots using it.

Wow, awesome. That's, that's really impactful. As mentioned earlier, KPMG now helps leading organizations develop and implement their own AI strategy, with Relevance being one of the solutions that's being used. How did KPMG's own internal experience with AI agents influence the way you design and deliver AI strategies for clients?

Yeah, so I think when we started playing with agents, it actually changed the way that we, uh, started thinking about digital strategy. And so if you think about traditional digital strategy and kind of transformation strategy, there's a platform or a solution, and then you're changing the workforce around it. And so the change programs that you deliver are training up the workforce to use the digital parts of it. There are parts of that that stay the same with agents, but I would say the difference is that the AI strategy and the digital strategy is now integrated with the workforce strategy, and you're actually looking holistically at a workforce and thinking about which components of that become agents rather than, you know, retraining personnel. And it's a really interesting, I think, difference and kind of evolution of what digital strategies and AI strategies are becoming in the agentic world.

That's awesome. Any frameworks or learnings that you'd like to share as well from, you know, these kind of different activities?

Yeah, I think one of the big learnings that we had is it's very useful to think about agents as if they are you know, human, as if they're a part of your workforce. But it's important to make sure that you do understand that they do have different implications and they do have different support mechanisms required. And it's not just like hiring someone, right? It's useful to think about it that way as a framework, but it's not as straightforward as that. And there are new capabilities that are emerging, especially around the support of agents and things like accuracy of models and maintaining accuracy of the the agents that you build that are really important foundational capabilities that a lot of businesses don't have right now.

Yeah, 100%. What AI requests are you hearing the most from clients right now and where do they want to start?

Yeah, the biggest problem, especially in Australia, but I'm sure worldwide, is that there are POCs and prototypes everywhere and many, many, many clients are struggling to one, assess their commerciality, and then to take them from that, you know, POC or proof of concept through to actually delivering it and actually making a meaningful difference in their organisation. And that's the number one that we're hearing from our clients. And I think it comes down to an ease of experimentation that agents and AI has right now where you can so quickly get something that kind of does, does the cool thing. But it falls down on actually delivering a business outcome or it falls down in the integration to the business that you need it to have.

Yeah, it makes sense. And that's where you guys come in and make it all a reality.

Yeah.

Yeah.

Where do organizations hit obstacles in their experimentation with or implementation of AI tools?

Yeah, I'd say, and it's an interesting one because businesses are not struggling to get outcomes like as in, sorry, I shouldn't say outcomes, outputs. So it's like, you know, it's incredibly easy to say, oh, I want this type of report drafted, do some simple prompt engineering and get to that report, right? Quite easy. Where businesses are struggling is accuracy. And the main reason that they're struggling with accuracy is not the underlying models. It's the interpretation of what accuracy in this world of words, in this world of, you know, white-collar kind of work. What accuracy actually means. And they've never had to define it in the way that you need to define it for an agent. And that's where a lot of these experiments are falling down, not in necessarily not reaching the accuracy, but almost simply not knowing what accuracy actually ever meant for that experiment.

Makes sense. And how do you kind of suggest, you know, customers and clients to to help overcome this? But it sounds like, you know, it's like the domain expert. How do you kind of explain what is good looks like?

Yeah, yeah. So, um, it's kind of, uh, funny that it all heads back to basics, right? Part of automating a process is articulating your process well, right? And I think those fundamentals are still there, right? You still need to understand what your process is, what you're doing, and the actual kind of way that that comes together. The bit that agents, I think, have really uncovered for people, which was always important, but it's really exposed a lot of businesses for this, is why are you doing the process? What is the actual outcome and what's the acceptable level of that outcome that you're actually able to make, right? Businesses are quite used to setting up processes as a reaction to it not happening well. But I think businesses are now realizing that they have to do a lot better at actually articulating what well means. For that process. Why is that important to our stakeholders?

And actually coming down to first principles.

That makes sense. You guys are almost like first saying like, hey, you know, why are we doing this process? Let's first audit this process. And then the more refined it is, that's when AI agents can come in and really help make a difference.

Yeah.

And what was the outcome that you invented the process for? Because it's there inherently. You may have never articulated it, but there was a reason that you kind of you know, put this process in place that you're now trying to automate.

I understand. How have the attitudes of businesses towards AI changed over recent years?

Yeah.

And so in Australia and for, for those who are kind of looking, I would recommend looking at the KPMG Trust in AI report. We do it every year and there's one just recently from 2025. We have incredibly low trust in Australia around AI. And I think that's kind of twofold, really. One of the areas that really drives that distrust in AI is that people feel like it's going to take their job, right? In a market that's tight, in an economy that's uncertain, people feel like it's going to take their job. And so that's an almost unavoidable part of the sentiment around AI. The other bit is that we are so kind of going back to what we said earlier, we're so POC heavy and like predator heavy in Australia that almost everyone has seen AI hallucinate. And the unfortunate bit of kind of that sentiment is that you remember the hallucination, not the 30 times it got it right before then. And we don't judge humans by that same metric. And so It's an interesting one because I've certainly seen a lot of hesitation in not only business but just consumers in Australia of adopting AI and really taking these solutions forward. But the unfortunate bit of that is that we will inevitably fall behind if that happens and if that sentiment kind of stays because genuinely these solutions do have sound fundamentals that can impact businesses right now. And so I hope that that sentiment is kind of dying away and people are realising that it's not just a threat to their life, it's actually an opportunity. We're certainly seeing that in other countries much more. And I think, you know, Australia has a great, you know, history with AI and the algorithms that sit behind it in a lot of our research institutions, right? Like, and so we should be one of the positive, you know, sounders of AI. We should be one of the leaders.

And I hope the sentiment changes to that. Yeah, I hope so too. And some of the top researchers, AI researchers in the world, are actually Australian. Exactly. Yeah, yeah. But it's really interesting how you mentioned one of the bits, which is like, when the agents or AI does hallucinate, we tend to almost like over-index on it when realistically, you know, even I hallucinate sometimes. Like, sometimes when I'm too busy, I forget which day it is.

Exactly.

Have you seen the appetite, like, you know, in the market to leverage AI really, really increase over the past couple of months?

Yeah, definitely. It's definitely really hitting a point where, you know, almost so, so I guess if you take it back a year, AI strategies were a thing, but companies were kind of like, oh, is it part of our digital strategy? Is it, is it just, you know, a component of our overarching strategy. I think as we sit here right now, most companies understand that you need a separated AI strategy and you need to understand what you're doing with AI. And that, that's a genuine shift that has also shifted, I think, investment funds realistically towards it. And the other thing that's happened, which I think is great and it's happened very recently, is all these pox and prototypes, all the companies have gone through it, you know, maybe got limited outcomes from it and now realizing they need to refine their funding, right? Rather than going wide and, you know, dividing your funding in 50 different places, go after the 3 main use cases and do them well. And that focus of funding is really leading to some positive results, I think, in the AI market right now.

No, that's awesome. What are some common concerns you've heard from businesses looking to implement AI?

Yeah, so there's a couple of things that businesses consistently ask us when we're going to do AI work. One is around just simply domiciling of the information and where the information is being processed. A whole bunch of our clients have policies in their business that you cannot process any information outside of Australia and just being honest, we are often released later in a lot of these model releases. And so that is a limiting factor, but it's something that's really important and it's a concern that our clients often have. But, you know, there is enough in Australia, there's enough, you know, domiciling here that there's always ways around that. The second bit, which is a little bit more complex, is around IP and kind of the trust of, I guess, some of the underlying, you know, model owners. So the Geminis, the OpenAIs of the world, and around giving away IP and kind of business IP into these models that, you know, might then use it to retrain. And it's an interesting one because it kind of comes back to some of the maybe broad distrust that Australians have, is even if you have the kind of arrangement in place that they're not going to use it to retrain, there's almost like some people feel like, oh yeah, I don't believe it. So it's been an interesting one. It's a concern that always pops up. And we as kind of providers in AI, we have to have answers to those. And so we do have things like offline models and those type of things to basically say, okay, fine, you don't want it to be used for retraining, like we'll take it completely offline. We have things like enterprise agreements with some of these AI providers so that we can kind of, stand behind and go certainty on where it's domiciled and what it's being trained on. But it's an interesting market right now. And I think that's where certainly the biggest questions we get are, is like where is that underlying information going and how is it being stored?

Yeah, that makes sense. And yeah, like all those different kind of things you mentioned help increase that confidence, increase that trust. When you were evaluating different AI agents, platforms, what were the biggest differences you saw between them and which makes the biggest difference to your clients' success in adopting AI?

Yeah, and it comes down a little bit to the use cases that we are often brought in to do. And if you think about the AI market right now and kind of the ease that is coming into it, KPMG is very rarely brought in to do simple stock standard agent builds. It's just not. You know, it's not what companies kind of contact us for. And so the nature of that is that we end up being in the more complex kind of processes or the more complex agent systems that they haven't been able to work out how to build. And so for those, the kind of frameworks that we were looking into were things like CrewAI, things like AutoGen, and then obviously Relevance when that came on, where we could, you know, quite easily and seamlessly have multi-agent systems collaborating with each other because it was is actually a base requirement of a lot of the use cases we saw, you know, out of the one agent into the multiple working together. And so that was basically the basis of our assessment is, you know, how much control do we have over that collaboration between the agents? And how can we see the outcomes of that collaboration like being traceable and being explainable were the two main kind of assessment factors for us when we were going through this process.

That makes sense. Um, what are the other factors that, you know, uh, maybe you haven't mentioned that you feel like really drive successful AI rollout and adoption?

Yeah, I think having very specifically under your AI strategy at least, uh, an indicative vision of what you want agents to be for your organization. Um, And what I mean by that is, do you want to be an agent manufacturing house for your own organisation? Or do you want to leverage third parties and integrate third parties in to solve those problems? And broaden that, are you planning on actually building a hosting platform for those agents? Like, people call it a library or a tree, right? There's a lot of words out there, but Is that, is that your plan or is it going to be more use case by use case and you're going to kind of tick things off as you go? And that foundational kind of what you want to do with agents in your organization is something that when we don't have that direction, it becomes very hard for businesses to actually make decisions on where to put their money or where, you know, what use cases to back because it lacks an overarching what's this heading towards, what's this contributing to? And so that would be the main one I'd say is something that I'd say businesses should look to do is just be very clear on that. It doesn't mean putting the money towards it now, but it means having that north star, having that vision of what type of business you want to be in the agentic space so that you can then inform these investment decisions.

Yeah, that makes sense. Like having a clear vision and mission and where you want it to go. Which leads on to my last question pretty well.

Let's look forward.

How do you predict, you know, the market and things will change over the next year or two? And it's, you know, whether it's attitude towards AI, whether it's how they implement AI, etc.

Yeah.

How do you see everything change in the next year or two?

Yeah, I think the massive one that I would actually be surprised if it didn't happen is business-to-business communications will be agent-to-agent. And I don't see, you know, there'll be the exceptions to that, but the efficiency that that allows for and also the kind of certainty in that communication and the traceability of that communication, the benefits that that allows for means that there's really not a lot of upside of keeping a lot of these B2B comms, you know, invoices, you know, all these levels of communication that go from a business to a business being just agent to agent. We can even see it potentially going into the realm of, you know, contract negotiations and, you know, all these things that, that yes, yes, now of course are humans, but in an agent world, there's an element of fairness. There's an element of, you know, standing for what the business actually wants their position on those contracts to And so a lot of these type of comms that right now are human to human, I strongly believe will be agent to agent. And what we'll be left with are the exceptions that have the highest value to essentially businesses, the ones that matter, the ones that are actually complex, the ones that we have to deal with. Most of those, you know, ones on the ground will be agents. I think the other thing that's likely to happen, and I think it's going to happen over the next couple of years, is mass adoption of agents in the sales cycle. And so actually a commoditization of the sales cycle. And, and I don't know what that looks like, but you can imagine, and I'm sure, you know, a lot of people are already getting this LinkedIn messages, 20 a day from different bots that are articulating stuff in different ways for different companies to the point that the reach out means nothing anymore. And so I'm interested as to what the sales process now becomes over the next couple of years.

I'm fascinated by it. It will be very similar to what you mentioned before with the business-to-business communication. Yeah, not just agents selling, but agents procuring and, you know, assessing. Yeah, but no, I completely agree. Like, the way the world's heading to, it will only be good for everyone, the consumers, when business-to-business kind of communication and collaboration becomes much more streamlined, like logistics is going to be a lot faster, procurement is going to be a lot faster and everything. Well, thank you so much for your time. Really enjoyed a lot of the insights you've shared.

And yeah, thank you so much.

No problems. Thanks for having me. Rob, thank you so much for taking the time today to chat with us. I'm really excited about this conversation because you've, in your career, led some amazing companies. You've done a variety of roles across go-to-market and currently as Chief Customer Officer at Canva. I thought you have such an interesting perspective on what AI in the enterprise looks like, both internally and externally within your product. And I thought it'd be really helpful to to share some tips and tricks, I guess, of what kind of you've seen works well in the market, because a lot of companies are right now trying to transition into an AI-first world, either with their product or internally, and they're struggling with that process. So we thought no one better than you in Canva to kind of help share with us some of those stories of what works and what doesn't.

So thanks for making the time today.

Yeah, I'm happy to do it. So maybe if you can give us a bit of context, like what are some of the stuff that you guys at Canva have been doing recently kind of to embrace AI? Maybe first starting off from the product side, because there's been a bunch of product launches all the way from Magic Design to other features and capabilities. What's kind of been some of the top things that your customers really care about when it comes to AI?

Yeah, well, Canva's mission has always been to make complex things simple and at the same time to empower the world to design. So when you think about AI, AI is the perfect ingredient, if you will, or catalyst to facilitate that more. And so You know, I'm speaking on behalf of our product and engineering teams to, to sort of share with their vision and their vision along with our founders is to make it easier to do tasks that used to take multiple steps. So one of our most popular features is background remover. And in the past that would take many, many clicks and it was very hard to do because getting, you know, the outline of some of the items you might've wanted to remove a background from, it's tricky., but using AI, it's literally done in a click. And now Canva went one step further, which is to allow for bulk background remove. And so when you think about like the amount of time and energy that, that takes away from users so they can focus on doing like more valuable, high impact kind of steps, that's the, that I think encompasses the philosophy of our AI approach.

And how does that change when it comes to selling Canva and kind of working with your customers to get them to maximum success? Like, How has the shift in your product roadmap and vision, uh, and really embracing AI, uh, so much, like how has it helped impact your customers and kind of, are you doing anything different to sell that product?

Well, I'll tell you the, um, the demo process for our customers is a lot more complex. Um, but I think that's good because what it means is our sales teams have to do a lot more exploration and a lot more discovery to understand what kind of problems our customers are trying to solve. And then the great thing about a really broad product platform like we have, we can solve many problems. So the key is figure out what problems they face and then create a demo that shows how we can solve those problems. So I guess to wrap up your question, the product breadth helps us finding lots of customer problems that we can solve is a great way to get there.

And then getting really deep with customers in the buying process is how we solve it. And you guys have really embraced kind of an AI-first strategy externally and internally. And how does that change your thinking in terms of how you shape your customer team, your go-to-market team, your sales team?

Like, what are you doing differently in this kind of AI-first world to build out your team? Yeah. Well, I would say there's some things that change and some things that stay the same. So I think we're really clear on first principles around go-to-market. And one of those first principles is we focus intensely on what steps a customer has to take to move from being aware of our product all the way to being loyal to our product. And so we're clear on those sets of steps. And internally we call that the bow tie because if you looked at it on a little like graphic, you see customers moving from one side to the other. And in the middle of the bow tie is the buying process. So there's all the presale and all the post-sale. So that doesn't change. What I think has changed is our ability to use data to be more informed about where our customers are on this journey. And it's actually the perfect product name that you have, your company name, Relevance. When you know more about your customer, you can be more relevant in the ways you engage with them. So presale, we can be more relevant. Post-sale, we can be more relevant when we're using more data. Now increasingly, that means using more AI tools and agentic experiences.

So I think the base work doesn't change, but how we do the work is changing. And you've said previously that you view adopting AI for customers and internal teams to be part of the same process at Canva. So what does the kind of end-to-end motion look like, uh, for your team at Canva in terms of, of that adopting AI and then using it to push customers through your buying journey?

Yeah, well, I mean, I'd say we're still very early. In the journey on the go-to-market side. So the way we use it to help our customers find, buy, use, and then grow. But probably the best description of what we're doing is we're, we're just evaluating carefully the way we do work and we're looking for every place where AI and automation could make it more efficient. And in particular, allow sellers and customer success reps be more engaged with customers. So if there's administrative work we can remove or there's process steps that we can shorten, that's what we want to do.

If there's data that we can use to be more relevant for customers, that's what we want to do. So how would you advise someone who's thinking about finding those opportunities to prioritize? Like, where do you think is the most impactful work, that administrative work or that repetitive work, um, that leaders can think about tackling first in their sales motion?

Yeah, well, I'll jump to the answer, but then I'll tell you, I think a good way to get there.

Okay.

So for us, recapping and preparing for calls, so preparing for a customer engagement, like a direct, you know, live meeting or a Zoom meeting or a phone call, the preparation steps that are required to do that work, um, are numerous. Hmm., and then the post-call, like recapping and preparation for the next call is a really heavy burden because we're recapping a bunch of conversations and we're recapping next steps and we're putting account plans in place. And so I focus there as, as one place. Um, and we have a lot of data about where customers are in their journey and what problems they're trying to solve and, um, what features they're using in our current product that can inform well, how we do that. And so we're doing more, I think, on the preparation and automation side for that. How did we get there? Two things. One, pay very close attention to the way customers move through the journey. Mm-hmm. Like what steps did they need to take to become aware of a product, engage with a product, decide to buy that product, deploy that product, and then use it to a point of loyalty. And then map on top of that the jobs to be done for all the people who help those customers. So whether it's pre-sales or sales or marketing, customer success or customer support, being really clear on the jobs to be done through that journey, customer by customer, I think just sort of pops out the areas to think about automating and making more efficient with technology.

Yeah, absolutely.

Couldn't agree more. I think that perspective of what are the jobs to be done within a business and really truly understanding that helps you identify where can potentially agentic AI play a role and where maybe it shouldn't play a role. And that's one of the interesting things that I like to ask leaders as well is what are the bits that you think are distinctly human parts of your motion that maybe you want to really have people spend more time on, not less, and actually want to double down and invest even more, like maybe relationship building. What are the areas that you really encourage your team in the world of AI to actually use an advantage that maybe they couldn't in the past?

Yeah.

Well, I think there's a couple of examples. And they're all very human. One of them is, uh, working together as a team. I mean, I think that's something that's really overlooked, uh, early in your career. One of the best things you get from people who have been in the, the job or a job similar is coaching and training. And so that is a very human interaction. It's a way to practice your communication skills. Like it's a way to like brainstorm problem solving. It's a way to think about building account plans. So I don't think AI can replace that. I think that is, that kind of empathy coaching engagement model is just uniquely human. That's part one. I think anything that relates to engaging with customers, outreach steps, you know, whether they're emails or digital touches can be automated, but when it actually comes to building trust with a customer and truly understanding customer problems, that's a human engagement, and that's like, it's eye contact, it's verbal, it's empathy. That's really hard to replace. Uh, and then I think post-purchase, helping customers find ways to get the most value out of a product, whether it's Canva or anything else, I think is uniquely human and it's hard to replace with AI.

Yeah, absolutely. I've heard this from, you know, uh, junior folks on teams asking their leaders what should they do better in now. And one of the common remarks I hear is, well, now that you have a little bit more time, maybe freed up from this part, invest in your relationships, invest in those customers. Because I think if we all could, we want to spend more time with customers and actually just speak to them all day if we could, rather than having admin work.

Yeah. There's a certain intuition that you gain about, um, your customer only through customer conversations. I mean, like a lot of reps are using tools to do customer research. And I think that's really awesome because it creates a foundation and surround sound of like insight to that customer. But the very best insight you can get is a direct conversation with the person who's going to be using the product about the work that they're doing and how that product will help them.

So yeah, it's really almost impossible at this stage to replace that. Absolutely. And AI now is kind of like sitting across many parts of go-to-market motion. If we try to look a little bit into the future, not too far, but maybe the next couple of years. How do you think the go-to-market motion is currently changing and where do you think maybe it's going to end up for the next few years once agents play an even more critical role in that kind of toolbox for sellers and CSAMs and everybody else?

Yeah, that's a hard question. Well, I know that AI is going to make information more readily available.. And so what I would hope is that the decision process is shortened dramatically, that there's a lot more give and take, that, that right now we think a lot about AI experiences when we're talking about go-to-market. We think very much the software entity to the customer. So like if you're Canva, you're thinking about Canva to a customer. And I think we probably in the future will be seeing a lot more customer to Canva. That the customer is going to pick up more agency on their own to engage back with a tool that may help them. And, um, that's something we don't really have today. I mean, it's like most of these bigger transactions are really seller-led as opposed to buyer-led. And I think the balance is coming more in line between buyers and sellers. So that's probably the single biggest. If anything, I think we're still doing the same sets of steps. There's still a problem identification. There's still a, like, what would be my critical event to decide to buy now? And there's still a discovery and sort of like experiential phase. And then there's certainly going to be some negotiation phase. Those things will still happen. You'll still create, progress, and close deals. I think the timeline for which you do that and the efficiency that you do it with will increase dramatically.

Absolutely. I think that buyer-led comment is really interesting because One of the ways we actually as well internally think about our customer buying journey is where can we find opportunities to make that experience for the customer better, faster, easier, whether that's speed to lead, whether that's giving them more information upfront that they maybe had to wait a little bit longer before, before they could meet with us. So what are all the opportunities between touchpoints with our people that we can help enhance and engage and make sure that every call we go to, we're more prepared than ever. We're giving them more value than ever. And that's kind of the bar I think a lot of go-to-market teams can start thinking about in the future. If we kind of derail the conversation a little bit in security and governance, how do you think your teams are being enabled to adopt and embrace agents whilst kind of maintaining all the constraints you'd love to see within an enterprise organization and not let AI just run around and kind of really have it controlled and enhancing your team?

Yeah, well, I mean, this is the best example of a great partnership with our IT organization. Because they are uniquely tasked with, um, data security and data compliance. Um, I think in the best scenario, the IT organization is unleashing some data that's appropriate for use and controlling others, and that's exactly their role. Um, and then once the data has been released, the appropriate data has been released for use, then empower the sales marketing, customer success teams to use it to its fullest extent.

And I think that's a good example of how we're using it at Canva. Amazing. And an interesting thing about Canva is that you guys are, I mean, within your history as a business, you guys have always been innovating quickly, adopting quickly. AI is no, no different to that. And in particular, what we found impressive is how quickly, um, and how much appetite there is for you to enhance your growth. Find areas of opportunity with AI. What's, what's the unique thing of Canvas culture that enables adoption of whether it's agents or any other thing in the past that you guys have done? What enables that quick adoption? Because a lot of the times people think this is a technology problem, but oftentimes it actually ends up being a bit of a mindset problem. Like what's the right culture and attitude you need to have as a business to enable rapid innovation, whether it's with AI agents or anything else?

Yeah. I love that question. Thank you. And it's a very unique cultural dynamic that I think other businesses could aspire to, but Canva has done a fantastic job of accomplishing, and that's entirely thanks to our founders. A couple of key things. One of them is we are a mission-driven company and we aspire to help everyone in the world design. Like that is just where we're going. And we have a two-step plan that backs that up. And the two-step plan is one, to be as big as we possibly can, to be as most valuable as we possibly can, but not for selfish valuation reasons, but for the second step of the two-step plan. And the second step is to do as much good in the world as possible. And so when you're that mission-driven, you're not Like fear is not your motivator. Fear is definitely not our motivator. Finding ways to help people is your motivator. And we have another expression that we use regularly, which is give value first, get value second. So if you really are a company that wants to give value first, you're not so worried about like, well, what will happen if we don't get value second? Because your primary objective is to give as much value as you can. And so, um, we regularly talk about like, let's do whatever we think is the best thing. And if it doesn't work, we'll fix it. We'll come back and we'll, we'll change it. And so we're not, we're not motivated by controlling outcomes and we're not motivated by trying to like use fear as a motivator to constrain us. We're motivated to help customers design. And so that's a, I mean, that's a hard thing for companies to get their head around because usually it's a, you know, evaluation thing or, some economic driver. And I think economics are important to us, but they're not the primary driver.

And then so for leaders who maybe are just starting the AI journey now, what's some practical advice that you have in terms of where to begin, what pitfalls to avoid, and how to really embrace some of this attitude required to be really successful in this kind of new AI-first world?

Yeah, like be fearless, I think is probably the single biggest headline. And do something. I mean, I think that's the other thing is we, there's so many good case studies of, um, companies not knowing where to start and so doing nothing. And so they become laggards. And then there's examples of companies that don't know where to start and they just sort of throw a whole bunch of things against the wall and see what works. And I think that's an okay way to go. I think assuming the AI is used in a way that follows either your customer journey or your own internal workflows, you're going to find something great. And so I think the early, you know, early examples that lots of people use in the go-to-market world is reps were using ChatGPT to do account research before a call. And although today we've like leapfrogged that 10x, that's a pretty okay place to get started. And so if I'm a go-to-market leader and I'm wondering where to get going, I encourage all the people who are going to talk to customers to do a little bit of research before they have that customer conversation.

So they're really prepared to have relevant discussions. Thank you so much for your time, Rob, today. I think once we revisit this in the next few years and look back at how AI has shaped the world, I imagine go-to-market is going to be fundamentally different, but also the same in terms of like, at the end of the day, we're delivering value for customers. And what I'm really excited about is seeing how can we do more of that with AI better than we ever have before. And so I really appreciate you sharing some of those thoughts today. Yeah, thank you. Thanks so much.

Thanks.

And that's a wrap on Agents at Work. Thank you so much for tuning in, and a particular thank you to the people who share their stories with us to help you get started on your journey with an AI workforce. If you want to learn more, check out relevanceai.com and book a demo to speak to our team and start your journey with AI in 2026.