SaaStr 2025 Summit Stage - Fiesta A

Alright, I guess I'm supposed to be sitting here. Welcome to Late Night with Lobo. Give it up.

Ooh.

All right, we have a great panel here. Your finance department is still cutlery coding spreadsheets. Meanwhile, our folks are using AI agents to close books, run forecasts, and replace entire teams. Give it up for our wonderful panel, The CFO is Dead, Long Live the Chief Automation Officer, with Soumya, ex-controller of OpenAI and Rippling, and Asen with SnapLogic and former VP of Finance at BlueJeans, and Kunal, the gorgeous man at the end from Gorgeous. Come on, that's a little dull. Let's like give it up for them. All right. So let's dive right in. Let's take it with Soumya. What is one traditional CFO responsibility you see completely taken away with AI agents in the next 12 months?

You know, that's an interesting question. I don't think responsibilities are really going away. Maybe them doing the work might go away. But in my experience with finance teams being able to do so much more, I see more responsibilities land on the plate of the CFO. But one thing Something that the CFO at OpenAI used to do is she would take our monthly finance close deck and run it through her custom ChatGPT template to summarize her notes to the leadership team. So something that might have taken a lot longer is now like it knows her voice, it knows what to highlight, and it just ships.

How long would that take previously?

Probably about like, you know, few days of back and forth with the team. And yeah.

So when you, when you started at OpenAI, how many people did you have in your finance team? And when you left recently?

So when I started, so I started at OpenAI in the controller role in March of 2023. So that is the month that ChatGPT Plus launched. And we had about 10 people in the entire finance and accounting org. When I left, which was March 2025, we had about, I'd say, maybe 30 people in accounting and 15 or so in finance.

Why did you need more?

Although the AI models got better, the company went from, you know, like, I don't know if this is like a trick question. OpenAI is the maker of ChatGPT, and we, you know, we scaled to like half a billion weekly active users. It still takes humans, unfortunately.

Awesome. So with, uh, Essin, SnapLogix is an AI agent orchestration platform.

Yeah.

What have you done with AI agents to automate your finance department? But moreover, how big is your finance department?

The finance team is not as big. We're also not as big as OpenAI, but the finance team is 4 people. Accounting team is 8, give or take. So we're about a $100 million business.

$100 million ARR business. And a lot of the AI agent companies that I talk to, they don't implement AI agents in their finance department. But you're a little different.

Yeah, yeah. I think we actually had the first live agent deployed across the company. And we were talking earlier, when we present our use case, which was essentially an order form reconciliation. We're an enterprise business. We have unstructured data coming out of order forms, Salesforce, PDFs, customer-initiated documents, DocuSign everywhere. And going through that, reconciling that, getting that into the revenue systems was an immense effort, monthly, part of close, et cetera. So someone on my finance team, Jason M. Lemkin basically heard us pitching internally our own agent creation platform and said we could try to use this on our own finance processes. And we ended up cutting essentially a day and a half out of close, and more importantly, finding revenue that was essentially leakage, things that were entitled, that we should have been billing for. So it was, and we've done that and then built on top of it to do legal, essentially legal discovery, were there termination clauses instead of— so instead of throwing bodies at it, the agent and again, I'm, I'm kind of on the cusp. I'm a little bit— so I said, why don't we get people together, have everyone look at it. And the person who built that agent said, we can just throw the, throw the agent at it, retrain it and have it go service all that information.

Awesome.

Kunal, how big is your finance team at Gorgeous and what have you done with the AI agents? In the organization?

Yeah, so our FP&A team's about 6, accounting team's about 8. I also have the data analytics and the data engineering teams reporting into me, and that's a pretty big team. That's 16 people in total. So I think the reason is like, I think it's really important for the core data team. And so our data team supports both product analytics, go-to-market analytics, but also the engineering teams. Engineering team, which runs the underlying models. I think it's actually really critical for that to live within the office of the CFO because we can architect the models in the right way. And I think for the finance team to be able to be kind of the accountability officer, right, to hold other people accountable, we need to have source of truth and metrics that people rely on and can trust. And so if my team can kind of create the standard data models that pipe into our billing that pipe into the different places, it's really important that the integrity of the underlying information in the database is really good because that's where all the analysis is being done. That's where the agents are going in and importing information, running analysis. So if that is foundationally not correct, like it's not going to matter what's coming out of that. So I'd say we're— that was kind of the first step for me was to kind of invest in that layer, build that team out, make sure the foundational models are pretty good. I'd say we're pretty early on in the automation journey. I think we've started with areas that we can extract value the quickest. So kind of low-hanging fruit to just prove ROI. Things like kind of manual reconciliations on the accounting team, journal entries, stuff that can help time to close go down, just eliminate a lot of the manual work. And we're starting to get to more, I think, advanced use cases now where the FP&A team is is using it not as like a replacement, but more as like an augmentation to get smarter with forecasts. If we look at a lot of usage-based data, what do we think that the GPT is going to say is a reasonable forecast? It's just as a pressure test for what we're doing. So yeah, I think we're pretty early on, but we see a pretty aggressive roadmap to kind of—

Could you walk us through one use case maybe in your team, sort of end-to-end, that's adding a lot of value?

Yeah, so, um, our— Gorgeous is a business, we do customer support software for e-commerce merchants. A lot of that is usage-based, and so there's a lot of seasonality involved. So a lot of customers, you know, during Black Friday, Cyber Monday, will have high peak usage, and that will drop during certain months. One of the biggest business problems we have is figuring out, like, visibility and forecasting for a longer time horizon. How are customers going to behave? What's their usage patterns going to be like? And so that's, that's been a really big challenge for us. And so what we've done is actually used AI and some like kind of machine learning to help us really have more predictive patterns of user behavior and individual customer level behavior, which has been a real game changer for us for like across sales, for our post-sales team to be able to identify even potential risk and churn candidates through that. So it's been a big help. And, you know, nothing is perfect. I don't, I don't think this is a replacement to a forecast, but it's a good pressure test to help us understand what other inputs are.

What tools did you use to do that, build that, or did you do it all in-house?

We used a couple of tools. So like we are basically, we use Google Cloud and then we have BigQuery as the underlying database. And then what we did was we did, I mean, the data scientists on my team are basically trying to build some of the underlying logic. And then we use Periscope as kind of our visualization tool for that.

Soumya, with yourself before you left OpenAI, what was one process that you guys completely automated?

We actually took two very seriously early on. We automated revenue and we automated our GPU cost reporting. And the reason for that is because in a time of hypergrowth, you can't have people doing this work because then you need to scale your people linearly if your, you know, your growth is demanding that. So we said, okay, We're going to take the hardest problems for our team. So for revenue, we use Stripe, so a lot of the source transactional data was in Stripe. We were really close when I left. Yesterday, our data person pinged me and said 99% of GAAP revenue is going, touch less, from Stripe all the way to NetSuite, which means revenue close basically happens real time. I think for GPU cost, it's slightly less. Slightly more manual right now. A lot of it has been automated and dashboarded, but I think the business logic is just more complex. But really, really happy about what we did for revenue.

And how did you guys do that? Was it mostly leveraging OpenAI or?

We leveraged OpenAI in not the way that you might expect, right? We're not feeding Stripe data through LLM models having it do that. We— the place we leverage AI is we told our revenue accounting team, you basically need to teach yourself SQL and Python. And these are people straight out of like, you know, CPAs, Big Four accounting background. They have not necessarily done any coding in their lives, but with a tool like ChatGPT to help them, we essentially put it very similar to what you described, Kunal. I also we also really insisted on having a data org inside our team. And I think having the finance data team being embedded in our org was a huge thing because they, they would actually see the pain points of the downstream people that needed that data. So we essentially had Stripe data flow into our data warehouse and then the accounting team would use AI to write SQL, to write Python scripts, whatever you needed to do. We eventually We actually built plumbing to send that transformed data into NetSuite for RevRec.

As a result of that implementation, what were some roles that just didn't need to exist anymore?

I think for OpenAI, it wasn't so much that we had to, you know, say, okay, these roles are no longer going to exist. We just didn't hire them to begin with because we were 10 people when we started and suddenly business grew. And I'd say if we didn't invest in these tools, we would have probably been like a 200, 300 person finance team for, you know, if you look at our comparable like kind of numbers and complexity, that's the peer size. So to operate a team like that with like less than 50 people, we just didn't need to hire more than 2 or 3 revenue accounting people because we said your job is to kind of look at the alerts that are firing off when things go wrong. But if things are great, then you're not actually like booking anything or doing anything. Everything's automated and dashboarded. But of course, new products get launched, new things need to be set up, and like the work is more so like, you know, only the exceptions need to be handled.

Definitely.

Just to build off of what they both said, I think the important thing to recognize is our transaction volumes are very low. We're enterprise. Obviously you have a little bit of everything. I'm sure you have, you know, a variety of sizes. We're all looking for what is the thing off the— I need to use AI. What is the thing I can use? Like, what do I go get? And our CTO made this point at our sales kickoff. One of our reps was like, I need the use cases. And the CTO's like, the use cases are in the people and the processes. And there's not something you just take off the shelf. It's gonna come from what you do at your company, what you you do at your company. This is so early in what we're doing. And you had asked, why aren't agentic companies using it in finance? Finance and accountants, we're not— we didn't choose this lifestyle to take risk, right?

Yeah.

There's a cohort of people. Well, I mean, many of us have self-selected, backed into it. But the point being, like, those aren't the departments where people are like on the bleeding edge. Like, how can I be like 98.3% right? And oh, that 1.7% doesn't matter.

That's not it.

There's a right answer and a wrong answer.

Right answer, wrong answer. And I think finding those use cases, pulling them back, and what we've done actually is we've had the same struggle. We're constantly out there pushing all these use cases across all parts of the company. We actually established a center of excellence/enablement where the people who built these initial use cases are now going back and change management, mapping what are the necessary conditions inside of your team and how do you like get people comfortable with it? I mean, it's been a lot is happening so quickly and it's exciting, but we're in early, early days. And so bringing people along for that journey and getting them to be comfortable with it is critical. It's like a people, people management thing as much as a technology, as much as a technology effort.

As you look into the future over the next like 12, 24 months, How does your finance department look? What changes? What roles do you hire for?

What roles you don't hire for?

I think Samaya made the great point. It's not that roles will go away over the next 12, 24 months, but I hate to say it because this is where a lot of us started, like junior analyst roles are the ones that are ripe for like, like we don't need those, right? Those are the agents that we can build. And again, maybe because we are a leaner team and the people role that I have are of that mindset. They're looking for things to go automate and just kind of, you know, not have to hire those junior roles or not have to hire more people to take those on. So that's kind of it. Now, I think, again, probably less hiring, more scalability inside of the existing organization is what the next year or two look like.

When you looked at your finance department, how How do you determine what, what was like your rubric for deciding what I automate first?

That's a great question. I was actually just talking to our COE lead. He said it's essentially, it's risk, effort, and business value. Like that's the matrix to think through these, think through these use cases. And I literally sent a note to my team like 2 weeks ago saying, you're about to go go through close, all the things that suck that you have to deal with, just write them down. Like take a note, take a sticky, and then we're gonna go through with the COE and assess it in that matrix. There are things that like, it's not gonna be a perfect application for everything, but where you can, where it's, you know, the risk is appropriate, the effort is appropriate, and the business value is as high as can be, those are the places where you start and then you start to knock them down from there.

Kunal, as you look into the future, how does your team team change? Because you've got like these 3 teams working with you, right? The data, accounting, finance.

Yeah, I think I'm a big believer that I don't necessarily— while there might be some slight kind of consolidation, I don't actually believe that this is going to be a purely agentic kind of workforce, right? I think what you're seeing is an evolution of what the team looks like. So I think a couple of things that I'd say. One is I think the team structure gets a little flatter, right? I think in the traditional hierarchy, you have a lot of junior people doing kind of like more of the routine data-intensive stuff, and then you have layers of mid-management, and then you have kind of more senior folks in the team. I think now that model changes a little bit where, you know, a lot of those ground-level tasks don't need to be done by humans or not all the way done. And I think what we're seeing is there's not a lot of things that are 100% getting replaced by agents. I think what we're trying to get people to do is less of the preparation and more of the reviewing. So I think what I see is people— what I want to do is hire more people who are going to really push the boundaries of what it means to be kind of a world-class finance team. And to me, that's connecting, you know, finance with operations, helping to influence decisions and doing the storytelling. Those are not things that I think are going to be replaced anytime soon by an agent. But what enables them to do is like, hey, a lot of the routine things that take their time, a lot of the ground floor analysis, the pumping manual reconciliations or doing a budget versus actual that could be 85% automated by AI. That's going to allow them to focus time on getting more embedded in the business, spending more time with their business partners, and now bringing operational insights to me and the rest of the team about how we can do things better or more connectedly.

You know, what I've seen is finance is this very political department, right? And when I'm on boards and everything, finance sort of comes together at the end and maybe a few weeks after the end of the month, unlike OpenAI, which is where the close is real time. And this makes the conversation in boardrooms very political. Kind of, right? Like competing interests, the data is sitting everywhere.

You want to switch spots?

Yeah.

So I think with everything you guys are saying, you're trying to break that, make the conversation more real-time, bring the data to the front so you can have a more collaborative discussion on finance to move the business forward. How is that happening in your organization? That gorgeous.

Yeah, I mean, I aspire to be Switzerland, so I try to be kind of a neutral party. But, you know, maybe we'll have some political leanings. But I think, yeah, what your point is, like, I think our role— I view a lot of my role as the chief accountability officer. So holding the rest of the team accountable for delivering, right? Where are we veering? On track, off track? And help kind of bring that connective tissue to bring there. And now the best way to do that is having core foundational data that the entire team can look at. Right. So I think everybody knows here, probably like the worst thing can happen is if multiple people have sources of truth and a competing source of information, then the discussion never goes anywhere. So I think what we are trying to do is instrument the team to really think about how to drive those discussions. And I think where AI and agents can really help is giving them time to do that.

Right.

And so So the more that we can bring the data to the discussion, but then have a point of view, like I think my, again, my controversial opinion is like data by itself is kind of useless. Like you need to be able to wrap it with a story and a point of view around what that means. And then you can debate that point of view. Like data is not the story. Data is the kind of the canvas that you can write your story on. So I think that's going to be really important. But to get that and to have high quality is something that I think Agentic Finance can help with.

So Soumya, with, with you carrying on that theme of story, how have you done that in the past, conveying the story?

I think, uh, so, you know, my role is more narrowly focused on getting numbers that make sense and that are right, right? Um, I'll give you guys an example to make this more tangible. So when I, when I started at OpenAI, our GPUs were melting. Like, we, we just had a lot of demand for GPUs. Everybody had a lot of, you know, they wanted to know what's happening. Are the deliveries on track? How much, you know, what is our cost of revenue, which is primarily driven by GPU costs? And we couldn't answer any of these questions until 15 days after the end of the month because like, I kid you not, that's how long it took to handle the accounting for all of this in spreadsheets. And I'm not exaggerating. So basically what happened is, you know, there's like you get this Microsoft Azure report with all your GPUs and there's different subscriptions with different costs and usage data and like what it's being used for. And then every time somebody deploys an engine or like starts something and it's being used, it's a line in some spreadsheet. Post ChatGPT, this file just exploded on us. It became like 9 million rows for a month. And it's not like a complex spreadsheet, but when you have that many rows, you realize you can't even open it in Excel or anything because it has a million row maximum, right? And people had to do all sorts of shenanigans to split that big 9 million row file into multiple files, do different things, and stitch it back together. And invariably somebody messed some date cutoff off and the total doesn't agree.

Disagree.

Now you have to go audit like, oh my gosh, which one got messed up? And it was this whole like just brutal exercise to just get GPU cost numbers and, you know, to figure out how much is cost of revenue and how much is R&D. We couldn't tell you. And I think that like storytelling makes sense and you need to have, you know, the ability to put it in perspective. But in my opinion, like a lot of teams are stuck because you can't build that story until you have the data, and sometimes this data comes from vendors and random forms that you can't neatly put into anything. And so in terms of the first big use case I tackled, it was actually this, because what I realized is by the time you tell people these numbers, if it's like 15 business days after the end of the month, it's like the 20th of the next month, nobody cares. They're like, oh cool, so we did that last month, month, you know, they kind of shrug it off and they're like telling a ship you should have turned back there. Yeah, it like does not matter. And, you know, people just stop caring about the results if you give it to them late was my big kind of eye-opening moment. And, and so we did a lot of work. We didn't have engineering help because engineers are like, don't talk to me, we're trying to keep the site alive. And so we turned to ChatGPT. GPT. We took our big spreadsheet. Turned out you could describe what you're doing in Excel to ChatGPT and it spit out a Python script. And Python is really fast, as you know, at like running CSV processing. And what would take us, you know, basically 10, 15 days was done in like 10 seconds. It was almost— it was a slap in our face by the team because it's like, what do you mean it got it done? You know, this would have taken us so long. But the story naturally followed once you got your data right, because now we could publish a dashboard real time to say this is what the deliveries are as of today and this is where it's going. Like these engines are cost of revenue, these engines are R&D. And like we realized we were becoming not just like the storytellers after the fact, but really keeping— I love Chief Accountability Officer, but like that's really what was happening, right? We could tell our engineers parenting teams, this is what's happening real time. If you're a vendor or if you're not getting what you want, go hold those people accountable. And I think tightening your process is, I personally, like the first thing you got to do because everything you do after that stems from that.

It's the time and process compression that allow— like it used to be accounting would close the books and FP&A would be waiting. Forklift over the financials and FP&A would go do the close packs and all of that. But to your point, now it's like those linkages are tighter and tighter and the business wants that information, should have that information sooner and sooner. So it's all the time compression that's made possible that allows you to steer the business or course correct or double down where things are working. And that's where accounting and finance will start, should finally are starting to blur.

right?

And that's where, then that's a great, great example.

What have, what is one thing you guys did to streamline that storytelling, that process, the data?

That's, I mean, not GPUs for us, but AWS costs, right? Our biggest costs are people and cloud, like stand up if that's the case for you, right? So, so for us it was the same thing. You get this 70-page PDF from Amazon on, excuse me, you get some other dashboards, et cetera. But it was like, how do you make sense of it? What is production? What's not production? What's for this product?

What's for that?

So running that process through an optimization is another example that I'd love to say we're done with, but we're in the middle of doing that as well.

Soumya, tactically now, I mean, can we use OpenAI? Can everyone use OpenAI the way you described it?

Yes, you don't need anything more than, you know, access to chatgpt.com. I think you can get a lot of what I'm describing done even with a free account. I obviously— the, the higher tiers of subscription give you, you know, deep research and some of these like more advanced features. But I, I honestly think the big open secret for finance teams is that tools like ChatGPT and Gemini and Claude make code accessible to everybody, and you don't need to be a programmer to be able to do that, to accelerate your own work. And the barrier to entry is really low, right? Anybody can just start talking to these tools. I think the challenge comes because it's— you kind of need to see someone doing it, in my opinion. I demoed this to my own team at an offsite offsite. Until then, everybody said ChatGPT is so cool, but it doesn't help my job. And this was people in OpenAI's own finance team feeling that way.

When you say demo this, what was it that you did?

I basically took our GPU process, manual process, and I real time wrote this Python script and ran it. And, and, you know, the team offsite was like 10 people. It was very low-key, but It was enough to say, oh my God, I can't believe you just did that in front of us right now. And it gave people something to like go try out at home.

It demystifies it.

Yeah, totally.

So I'm a, say, Series A finance person. What are like 2 or 3 things you absolutely recommend? I don't have a massive budget.

Yeah.

That you recommend I do tactically.

Like do in ChatGPT, for example. Yeah. I mean, like coding your bank activity, coding your expenses. If you know you're spending a lot of time combing through various accounts to find anomalies, you can just feed that stuff into ChatGPT and say, find me T&E that looks off. Then it's just going to ignore any like, you know, Uber expenses under $20. It's not going to care about if it sees an Uber cost for $2,500. $100, it can surface that to you. It's like all the little things that you would otherwise have to, you know, do in a little spreadsheet by yourself, you can have conversations about and get it done much faster.

Asin, how about yourself? What is your recommendation for a CFO just starting out in early stage— they're not early stage, but like a Series A company?

Um, find out and then come tell me. I mean, I, I think I think those— I mean, I've inherited a lot of legacy systems and processes and that sort of thing. I think if you're building from the ground up, it is more of a blank slate. And so, short answer, I don't have any advice other than listen to my peers who have done this inside of shops in that way. But I guess the other one is to think about— to just bias yourself first towards how can, how can the AI or the agent help me versus I need to go hire for someone. And I think more and more of these recipes, these tools, this sharing will happen that will, I think, accrue to the benefit of companies that are building from the ground up now.

Kunal, how about yourself? What's, what's your advice at the Series A stage?

I would say one is invest early in the integrity of your underlying data. Data, right? Because again, if the data is corrupt, nothing you're going to do is just— it's going to be worthwhile. So I think investing in the right underlying data architecture that you can get out. I think the second thing is, I get it, you know, at a Series A, you got to just really define like what the scope is of like what the finance team really needs to do, right? Because you're going to be constrained, you know, your team is focused probably on finding product-market fit, on scaling. And so I think the challenge is Be really clear about what the mandate of the finance team is and also what it's not going to do. Because I think the trouble you're going to have is if you're trying to be like a world-class strategic finance team with one person and a CFO or a head of finance and they're chasing around all these things, it's just never going to work. So I think you've got to constrain it and maybe put some barriers around like, hey, we've got to close the books, have high integrity in our underlying underlying accounting numbers and where you're not going to drive a ton of like sophisticated forecasting or underlying, you know, maybe the rigor of what you're doing. So I think it's about applying the constraints of what you can do and then figuring out which of those processes can you automate or at least again try to get 70, 80% of the way there instead of replacing all the way.

In your department now, in your function, Kunal, what's one tool you guys can't live without?

We built—

so there's a tool that I built all the time. Again, I think we, we have the advantage of having this kind of bigger data team, but we built like a semantic layer on our database. So it's great for me because it just— I could interact with it in like real English, and I can ask them like, hey, how much— how many churn customers did we have in the last 3 months in our SMB segment? How does that compare with last year? How many segments churn customers did we have last year? How many new customers did we add? So I can talk to it like I'm talking to someone on my team, but I don't actually need my team to respond to me. They're just giving me all the answers from that. And I use that tool multiple, multiple times a day. So that to me is one of my most valuable tools.

So, I mean, I guess it's OpenAI. Way through.

I think once you have a team that knows how to use ChatGPT in their work, you can't go back. I think if we told the team, hey, you can't have this anymore, everyone would like throw a riot. You just, you get used to a very, uh, fun and easier way of getting your work done that you just can't go back.

And let me guess, Asan, it's going to be SnapLogic.

For us, it's our own. Yeah, because we, again, with the with our customer base, we use a lot of proprietary information that's not going to be fed into, you know, into one of the, one of the larger consumer-facing LLMs. So we started using it for ourselves that way, and our customers are obviously very sensitive to that too.

And those are like your own small language models that you create on top?

Yeah, they're specialized models, and again, we're— they're standing them up for pharma, for manufacturing. But actually, even within companies that are big enough, we're gonna have their own patterns, their own use cases. So regardless, that's our customers, but we're doing it the same way in terms of building agents for ourselves and building on top of them and evolving them has been critical for us.

Awesome. I'd like to take some questions from the audience. The mic's right there. So whoever wants to, let's take like the next 10, 15 minutes to get any questions. Yeah, the mic's right there. So you can just walk up to the mic and we'll take some questions.

Great. Thank you.

I'm curious, how, how can a company like mine use ChatGPT and input confidential data and keep it secure?

ChatGPT has an enterprise subscription tier where you get your own cloud environment. At that point, it's not any different from trusting your data with any other cloud service provider. Don't do it on personal or free accounts. Yes.

Any other?

Yeah, please. Great talk, guys.

Do you have any examples of how you've enabled product go-to-market to move faster with automation? It's something I'm thinking about a lot. Like, how do we not hire a deal desk, but rather automate that? I mean, any examples or perspectives on like outside of finance actually like driving growth velocity more efficiently or anything like that?

Customer 360. I mean, like, that's a— you're stitching to the data point and clean data, like stitching together all that information, a cube that people can go interact with, ask questions. Like, that's the biggest one, and that's multi— I mean, it's product usage, it's finance, it's accounting, it's everything in there. [Speaker] [INAUDIBLE] [Speaker] We have our own visualization service, but a lot of it ends up in Looker. Monker? That's just because we use that, but yeah.

We, um, there's two things. One is we built like our own internal like churn risk score. So for the go-to-market teams, customer success teams can identify which customers we think are the highest risk of churn. That's been somewhat helpful for them to kind of go and attack those customers. Um, the other thing we did is kind of look at inbound leads to be able to score inbound leads based on like market data, being able to enrich those inbound leads with more information, decision makers, past history, it's just given the reps a little bit more information to deal with and how to route things a little bit better. So the more information you can arm them with, the better.

Hey, Lloyd.

Hi, guys.

Thanks for being here.

Quick question.

So I'm a founder of a company called Vasul.ai. AI. We automate collections and accounts receivable. So my question is, as a CFO, other than security and compliance, what other risks or things that scare you in terms of automating your businesses?

Maybe we're part of the same, like, risk and compliance and data information Information leakage are the ones that I worry about. Otherwise, like, have at it. It's my, it's my fault.

Yeah, I think from, I think that's right. I mean, the biggest risk is just data, customer data leaking and doing those things. I think the other part is that's scary and like we sell AI agents to our merchants and I think where it's a similar risk that I feel is the risk of like automation you're taking taking a certain set of logic, and now you're taking that as like the truth, and there's not edge cases. And so if something, especially that's something that's touching the customer, you have to be overly sensitive about. And so things like automation, like collections, AR, those are like really actually helpful things and manual things to go after, right? But now what can be a trigger point though is like maybe there's a customer that you've arranged a relationship with of like, hey, we know you're paying us at 60 days instead of 30 days, and somehow that logic gets lost and you're reaching out for collections, that's like now a damaged human relationship. And so like I think those things are the tricky points. They're all like solvable and you have to deal with some kind of band of error, but I think that customer-facing relationship, making sure that's high integrity, that's really important.

Hey guys, hey guys, thank you for everything you're saying.

I haven't heard this yet, and I don't mean this as a negative at all, but I haven't heard you say the CFO is dead.

And so I want to understand that a little more because to me a CFO runs strategy and thinks forward for the company and uses those things to decide decision-making. Can you explain a little more why that's the title?

Well, that was just—

It was to get you here.

That was just to get this audience. You gotta be a little Conor McGregor to draw an audience. That's what I did. But you were—

I think it would be fair to say that the CFO is not dead, it's evolving. The legacy CFO, yeah, the legacy CFO has changed. I think among this group, and if you extended it out, what we're responsible for has shifted, and where we get involved is shifting a lot. A lot faster than it has over the course of my career.

Any other questions? I think Ray has a question. All right, let's give it up for this panel. We're gonna take a couple minutes and then Bring on the next panels.

This was great.

Okay.

Hello.

And I know both of you guys.

All right.

Could we get some of the audience back? Awesome. It's a little loud. All right, excited for this second panel here, and we're going to be talking about how to turn finance from cost killers into growth dealers. We have a great panel here with Brian from Demandbase. Rodrigo recently left Poshmark and joined Joby Aviation as CEO. CFO, Sima with a16z. I've been a longstanding fan of Sima. She has this wonderful finance pyramid. And if you don't— haven't seen it, just Google search a16z Sima finance pyramid, and it'll give you this sort of this pyramid where you can see where you start automating your entire finance stack, starting with like transactional stuff all the way to Strategic. And then Ron, I guess the youngest AI person in finance I've met at least, recently raised $8-point-something million from General Catalyst to automate and streamline finance. So thank you all for joining us. Give our panel a huge round of applause, please. Come on, we need something louder. So, um, to kickstart, I want to start with Rodrigo since he recently left Poshmark and joined Joby, and he's coming in from another AI CFO event. So let's, let's start with you. I want to unpack how you used AI in the most recent company that you left, Poshmark, to streamline finance and turn it from a spreadsheet driver to a growth dealer?

Yeah. So we use in multiple aspects from accounting to finance. In fact, I met Somnia here again. She used to be the OpenAI controller when I met her a few years ago, and then I saw her in a conference like that. Telling us how her team was closing the books. That was 18 months ago, and I think I was one of the first CFOs to sign up for the enterprise ChatGPT, and very quickly we embraced it. So number one, I found at least generative AI to be very good for descriptive analytics. So we used to have a process that goes through the business reporting. What happens last week. And again, Poshmark's a pretty big business, $2 billion of revenue, 150,000 transactions. So every month I used to go talk to the exec team and say, hey, this is what happened last week. Then I started writing about it. And then as of a sudden, there were like 2 analysts going through the weekend, putting a lot of hours of work. So number 1, we fully automated it. So today ChatGPT looks at what happens last week and explains in a 6-pager on what happens, what, what was the drive, what were the drivers, so what happened with the marketing spend and active users and all that. So that's FP&A. So accounting, we used to close the books in 21 days, now we close in 3 days. Not all done by AI, but AI helping, especially banking reconciliations given the size of our the complexity of the transactions. So that helps us a lot along with the implementation of other systems. And very interesting, in 18 months, I'm seeing some of those systems actually now offering AI capabilities on top of what they offering were. And then funny enough, even on the legal side, we are starting to look at ways of using generative AI generative AI into the flows. And very recently I have training of ChatGPT in Python for my entire organization, which was very controversial. Some of the accountants say, hey, why I'm learning? I'm not a programmer. And then my way to reply was there will be accountants that know AI and accountants that do not know AI in Python. You do not want to be on the second group. So So those are some of the examples I'm happy to unpack as we go through the chat.

Brian, how about with Demandbase? What are you guys doing?

Yeah, just to maybe build on the accounting side and then I'll go to a couple other functions. Similarly, you know, enterprise ChatGPT is really a standard for a couple of the leaders there. Kind of the basics. I say we're in kind of the crawl walks. Stage, right? Without like full automation, there's a lot of more Copilot-type operations going on, drafting emails, memos. Reconciliations are a big deal in terms of saving time. So I would say our controller is kind of quickly making that the default on anything that's highly repetitive, transactional, or in the case of more exploratory work, using ChatGPT to really get to the bottom of prioritization. So a good example is on the collections side where, you know, when you have 1,000+ customers and you're looking to kind of get to the accounts that need the most attention and then leveraging that analysis to really prioritize the week for the collections team, that's been really helpful lately. You know, switching to more like the legal front, I've actually been surprised that that team's been a little bit out ahead of everyone else. It sort of starts simple, right? Automating the NDA process, right? Customer, partner, investor engagements all require NDAs often. And that is one of the simplest things to automate out there with some of the tools that are out there. So that's been really nice. I mean, anytime you can get your sales team your legal team coming to you and saying, thank you, you've automated 80 to 90% of, you know, that workflow. That's a big deal. And so that's been, I would say, an early kind of easy win. We're now going more into depth around MSAs. So, you know, training the model on key clauses that are important and then using that as an escalation process. So one of the things we've been able to do is really take kind of a contract manager plus, you know, essentially a model to kind of elevate them to more of like a senior corporate counsel type role. So I think what we're seeing is we're able to take maybe a more junior role with the right tools and level them up with the right exceptions.

What is the one tool— like you said, ChatGPT a lot, both of you, but is there anything else that you guys can't live without?

This is for everyone or just for these guys?

Just, just for Brian.

Yeah.

So I would— I'll mention a couple of tools that, that we leverage. And, you know, what I would say is there are some great native AI tools that we're evaluating right now. So I don't want to necessarily mention all those, but we can talk a little bit about the exploratory nature of what we're doing. And right now on the, what I'd call more AI feature front, there's tools like Ironclad and Flowcast, Numerik, even Zip that we're leveraging some of the kind of native AI workflows within.

Awesome.

Yeah.

Sima, you have this, you know, I give you huge praise on the pyramid. Could you talk a little bit about that and what you're seeing out there? At different parts of the finance pyramid and where AI comes in and is doing the job very successfully.

Yeah, no. So just to refresh folks, or if you haven't seen it, there's a— I created a pyramid a few years ago which essentially said, you know, to Lloyd's point earlier, at the bottom there's the most transactional tasks that the CFO's office has to do, which which is capturing literally transaction data, whether it's coming from bank accounts or PDFs. And then sort of as you move up, it's like, okay, we've got the transactions, let's reconcile them, let's run our analysis, and then do sort of the reporting and more strategic work on top of it. But the reason it's a pyramid is because the bottom level traditionally has taken the most amount of time to the point around the 21-day close. A lot of time is spent spent on that manual work, which nobody enjoys doing. And most people in the, on the CFO's team are overqualified, frankly, to be sitting and inputting data from PDFs. And so AI actually, you know, when I wrote that piece, you know, pre this latest wave of LLMs in the last year or two, but it actually applies quite well because that sort of transactional data can now be ingested in a much more efficient way. Things like the close period can be brought down as a result. And so where we're probably seeing the most tools, they are addressing that bottom part, which is around data retrieval more so than it is reasoning. And that's where I think LLMs are really best suited today. And I think they're getting better to go up this pyramid. But that's, you know, like I said, extracting invoice data, contract data, Putting this all in one, one place, we've seen a lot of software that's creating journal entries as a result, helping do audit checklists. And then the reconciliation is another thing that can be flagged and move the accounting professional on your team from being a doer to more of a reviewer. So we can say, okay, these are the two numbers that are getting pulled up between Salesforce and what we've, you know, the invoice that went out and, and what's actually come in the bank account. What's, you know, where, here's where the discrepancy is. And then, you know, and Ron can talk more about this, but I think as we're going up, you can, with that core set of data, you can actually make more strategic decisions as a team. And I think we'll see more software sort of at the top of the pyramid, but I'm excited and I think hopefully all of you guys are excited that your team is spending less time over, you know, each day on the bottom of the pyramid.

Ron, tell us what Sapien is. You guys raised a lot of money, dropped out of Ivy League schools, computer engineering programs, to build this AI agent that runs your finance department apparently. So tell us what it is, what are some use cases you're seeing, and what success you're seeing out there.

Totally, yeah. First of all, thank everybody. I think the biggest thing The first thing that we saw to kind of start working on this was that all the tools in the world of FP&A and BI and strategic finance were one of two types. They were either incredibly cookie cutter and they expected every business to use the exact same thing, or they were very vertical specific and broke whenever you tried to kind of go outside of that vertical specific mold. We kind of saw that AI would have leverage in the most primary way of giving you a system that actually understands your company. If you want to do forecasting, if you want to do analysis, if you want to generate commentary, the only way that's possible is if the AI actually knows how your business functions. If you're, you know, a subscription software versus usage-based software, or if you're a manufacturing company versus a restaurant, there's no way the same software could possibly serve all those businesses. So that's kind of the first starting point. And then from there, we wanted to see, could you build a system that could sit on top of ERPs, data warehouses, CRMs, other systems that have financial and operational data and pull them all together and then let you automate analyses on top. And that's where we've kind of seen 3 other main areas for AI being useful, one of which is just pace of iteration here. Processes that companies used to do monthly or even quarterly when it comes to looking at variance analysis and drilling down, looking at attribution analyses, looking at cohort analyses, can now be done weekly or daily or live with the commentary on what actually matters. And so now that you're doing them more consistently, the other two sides that we've seen with companies is one is, can you drill deeper? The most common thing we hear from CFOs that we work with is, I know that this part of the business is down, this part is up, but I have no idea why. I don't have the bandwidth to go and check why. I don't have the ability to go have my team go look into why. But if you have an AI that understands all the calculations that went into it, it can also tell you, hey, you're up in these areas and here's all the drivers and the chain that leads leads to that happening. Now, the last part, and this part that we're most excited about, is— I think I actually mentioned this earlier— but AI as a prioritization layer, right? If you have this AI that understands your business, it's automating a lot of your post-close financial analyses. It's letting you drill a level deeper. Now you have, frankly, a lot of data that you don't necessarily know what to do with. And so a lot of the companies we work with, we consider data-rich, analysis-poor. And this is where prioritization lets you say, hey, my AI knows these I have 1,000 numbers about my business right now, but what are the 3 or 4 that actually matter? What's the narrative that needs to be built around them, and what decisions or actions can we take as a result? And so that's kind of the package that we've seen tied together where you can automate processes, but get more into what is actually the proactive analysis that can be done before people even are asking the questions that they wanna be asking. All this done in natural language and specified to each business that we're working with.

Fantastic. Rodrigo, um, Can you share an example where like faster forecasting materially changed a business decision? Maybe it's a budget shift or improved GTM or hiring, things like that.

Yeah, interesting enough, what I'm finding, and again, you asked the question about which tool we use, mostly ChatGPT for pure AI and then we have a lot of homegrown on top of it, like a lot of GPTs that we created. And I must say that for forecasting the business, I haven't found yet, at least the way that we use the tool, to be fairly either accurate or meaningful. So from predictive analytics, from forecasting the business, I have not. However, from a marketing resource allocation by channel, it's amazing. So there was a certain occurrence that we spend a lot of money think about like $200 million a year in marketing, and then we are looking at what— how can I link the investment to the return in terms of users, and then we use the tool just to dissect, hey, help us understand the correlation. When I'm buying TV on tier 1, tier 2, and tier 3, how does that manifest in terms of active users after that? So the tool was interesting because running we did the analysis and asked those questions, feeding the data, what we realized is the marketing team was investing a lot looking at impressions. So essentially they say, well, tier 3 TV is cheaper, so I can get more impressions for my money, but not necessarily would those impressions turn into active users, because active users can do a purchase. That's kind of how— that's the top of the funnel, the way that we found. So not necessarily for the forecasting, but looking at things that way, we started finding our R2 correlation, like statistics, looking at what happens after I make this investment down to the funnel in terms of converting the active users. And that dramatically changed how we look at the channel allocation for a big dough of money. And then today we improve the return on investment for their marketing investment investments for the company.

Awesome.

Brian, how about yourself and your organization?

Yeah, I, I would say kind of back to Sima's point about like the foundational layer and kind of what you can do at that like base transactional layer to kind of level it up a bit. We've been able to really ingest some of that like phone call and email data, not so much to make completely predictive forecast. We're not there yet, but really to help sales leaders kind of collate, you know, if you've got 50 to 100 sales reps, that, that information layer in a way that's easily digestible so that sales leaders can then kind of do their thing around the forecast. So I would say, again, it's more like, if you will, Copiloting or helping you get to that basic kind of ingestion layer that's, that's normalized. That's where it's been most helpful. On the go-to-market side. I'd say the other thing that has been interesting is more around customer telemetry. We've been able to use GPT to take time series data.

Yeah.

And analyze, you know, customer touchpoints in a way where it's kind of informed our pricing model. So we've taken, you know, customer data across maybe 100 accounts on a specific product. And use that to kind of inform our pricing model going forward. So in one case, we migrated more to us from a seat basis to more of a data usage basis, and that's been received really well.

Who does that in your organization? Is there like a specific role you've created for an AI orchestrator that handles all of this?

Yeah.

Everyone has the capability like Rodrigo talked about.

It's a good question because I think that's one of the things struggling with is like who's kind of the ultimate like point person for AI in certain regards. I think in the case of product data turning into pricing changes, that was a combination of product management and RevOps sort of partnering together to do that. But ideally, product is very engaged around what's happening with telemetry and then helping us decide where we go from there.

Awesome.

Sima, from what you're seeing across your portfolio and other vendors, vendors and what you're investing in, how are you seeing companies use AI to improve the strategic side of things like, like GTM or capital allocation, fundraising from a, from a CFO standpoint?

Yeah. So I'd say every functional area right within a company is seeing AI tools. So whether it's go-to-market, it's finance, it's HR and recruiting, recruiting. The CFO, the interesting thing about where we are in the buying process is the CFO is actually being brought into a lot of these deals, either to sort of think about automation or also think about cost and procurement, right, too. So the CFO is actually seeing a really— has a really interesting vantage point around what tools are being brought in. I think within, you know, across our portfolio, I think there are things that feel lower stakes in the sense that you're like, okay, I'll give it a try. Things like pipeline generation, it's like, okay, I'll pay some money for a tool. We'll see what leads come in. It's— and if we don't like it, we'll churn, or we'll like— it'll get better. And you're willing to take that risk. As you guys all know, CFOs can't be like, you know, let me just try this thing. It doesn't work. It's okay. You know, the quarter's numbers are wrong. We'll just reforecast it. No one has ever said that, right, before. For it. And so the stakes are just a little bit higher. But I think this goes back to the point earlier, which is like there is enough work to be done that's, I don't wanna call it lower stakes, but easier for an LLM to process. One thing we hear a lot from CFOs is using ChatGPT to generate technical tax memos. And that's like all language-based. It's not something that's necessarily like I'm putting this in front of, like, my board for the numbers, but it's doing the work and I can go back and review it. Or asking for specific questions on, for like an audit purpose. It's like retrieval of information, right? But it's, again, it's not like generating the report or generating the numbers or a critical decision being made. But I think what we're seeing is a shift going from, okay, really let's, retrieve the data towards like, how do we start making suggestions and equipping you guys with more data and more options about like, okay, should we— is this going to be the more prudent path to go down or this? And then so that you guys are actually— you're having a richer set of data to back up the decisions you need to advise on.

How has it improved board hygiene? Because you probably see how many boards you sit on.

7 or 8.

7 or 8. And you know, what I've noticed is like finance, there's always this political thing going on, right? What are you recommending to the finance leaders in your boards, and how is this changing sort of that conversation in board meetings?

Yeah, look, by the way, we advise things to our finance or the leadership teams. We don't make them do things, right? They have to feel comfortable comfortable with it and what their teams like to use. But we have seen, I've even seen in the 15 years I've been investing, the amount of time the team has to take creating a board deck and then the richness of the board deck has really changed. Especially as a startup or a growth stage company invests more into their finance team and progresses. But part of this is equipped by tools. Like, you know, if you can, um, you don't need a team doing 21 days of close. They can be doing other things and you can get extra cuts of the same data. Um, there's like, you know, automated presentation and software updating. Um, things like AI can generate, you know, a lot of the, uh, the commentary that comes off of— again, you have to review it, but I think a lot of that stuff can be better. I've seen with one of my own companies, um, there was, you know, when we did the Series A, we weren't really sure what was going on with churn in the background. Um, we hired a VP of Finance, uh, at that time, and this is actually pre-AI, but now could be done with AI, had set up a set of macros to actually go through and pull all the data out of Excel and figure out, okay, here's how we segment the churn by customer segment. Um, and it was huge. This— so I think just like the board, board packs, I've just generally seen a much deeper set of operational data. And I think a lot of that is, is going to even get better as, as AI tools proliferate.

Ron, what are you, what are you seeing from a sort of how do you drive the GTM, the strategic side of, of the business with finance?

Yeah, I think, I think what Sima said is actually very relevant. A big part of it is that teams are fundamentally bandwidth constrained and there's only so many analyses that companies can go into, right? There's a limit. If something takes 10 hours, there's only so many analysts that can be dedicated to it. But if you can use an AI system to get that down to 5 minutes, 10 minutes, run a bunch of them in parallel, now you're getting the data back and you're analyzing it. It shifts the role of the analyst a lot as well. So I think that's kind of the, the starting point that we're seeing. I think then it depends on the types of companies we work with. So we work with companies that are more like industrials in the manufacturing vein, we work with companies that are more like multi-unit retail, we work with companies that are software, and I think each one ends up seeing different things that matter. If you're a big manufacturing business, the most important thing is just understanding gross margin analysis, and that's something that you can usually only do on a monthly basis of how are certain customers and products and these sorts of things performing. But to the point we mentioned earlier about increased visibility, right, if you're seeing that on a daily on a daily basis, these companies can actually move how many people they're staffing at different factory lines. They can move how much they're mixing certain products and certain customers. And so that's definitely not necessarily on the GTM side, but more on the operational side. And that's a lot of what we think about is that right now, FP&A and BI and operational decision-making are kind of considered three separate functions within companies. And so a lot of times early on when I'd be pitching the software to CFOs, they'd be like, this isn't really FP&A. 'What's FPA?' And oftentimes my response would be, 'It's what FPA probably should be.' If you're making the analysis just to put it in a board deck versus to get to the decision, it's not gonna be as valuable. I think for software companies, you've seen a lot of understanding customer cohorts being a huge one. And oftentimes it's as simple as, there's 1,000 ways you could slice them. You're picking cohorts by the start date, the day they first signed up. Maybe there's a better way to segment your cohorts. Maybe there's a better way for AI to try 100 different clusterings of what all your cohorts might look like, model each one of those out, and now your forecast is 5% more accurate than it was before. And that's a pretty big driver, because frankly, all these forecasts I've seen companies use is 3% year-over-year growth, and they pick some hyperparameters, and they hope it works out. So we definitely see a lot there in terms of where you could do one thing, you can now do 100 or 1,000. And the role of the team is now, I don't have to go do the manual work. I can just think about, which one of these options is the best. And that usually ends up being much faster, much more effective to drive decision-making.

Fantastic. Give us one use case of how a finance person can sign up for Sapient tomorrow. And I didn't want it to be a pitch for you, but at least, at least you can.

I appreciate it.

You can give us a use case that is very popular across your customer base.

Yeah, I think the cohort analysis one is huge. Companies like plugging into their wherever their customer data lives, whether it's a Snowflake or a Salesforce or those things, just pulling that information in. It's a big one for software. I think the other one that we've seen a lot with is variance analysis plugging into NetSuite or whatever ERP system people are on and just really doing that budgets versus actuals quickly and not just giving you, hey, you're, you know, you, you overspent in these areas, but what are actually all the drivers versus, you know, going down 3 or 4 levels to understand those. Those have been 2 very big use cases for us. I think the other big thing that I'm not even touching on is just the ability to do ad hoc analysis. I think this is the big part with AI is that you don't have to structure out everything you're going to do. And so a lot of our customers just plug into NetSuite, plug into Salesforce, plug into Snowflake, and basically have a semantic layer on top of these systems. But they can actually ask financially oriented questions of like, hey, what would happen If we spent $1 million more on marketing for this segment? And it can actually go and reason, figure out, pulling in the relevant data that actually helps answer that question, and give you back something that an analyst might take, you know, 4, 6, 8 hours to do. It can get it back to you within under 10 minutes. And so I think that ad hoc side is really one of the big things we've seen, particularly with software businesses. A lot of the financials aren't the most complex, right? It's relative. There's usually like only a couple of key revenue streams. There's a lot of systems around those. but then it's like, all right, how do we then take all the things that are outside our consistent cadence and add those in? I think that's where we've seen a lot of power with AI as well.

Fantastic. Rodrigo, without AI, what's one thing you would've done probably differently and cost you a lot in the last, last year or two?

So without AI?

Yeah.

Yeah.

Hmm.

It's interesting because we have more data more like decoupled from things and using AI. But I'm going to say it's almost like coming back to the organization that I needed to do some of the analysis that we do again on the descriptive, telling the exact team on what happened. I must say what's interesting is I never had to lay off anyone. However, we had some churn in the organization. And when I look back, I haven't backfill any one of those roles. So the way that I think about is, have I not had AI, I would probably have backfilled almost like a pretty sizable part of the team. We used to be public. We got, you know, taken private. And with that, a lot of people left the company. But we essentially now we are faster, we are more accurate, and then we work with almost like half of the the size of my F&D organization. So in summary, I'm gonna say to undo what we have done, I would have to hire a lot of folks that left the company that I decided not to backfill.

How about yourself, Brian?

Yeah, I would build on that, some similarities. I've been thinking a lot about the Shopify CEO memo a few weeks ago. A month or so ago. Many of you have probably read that, and it's kind of really stuck with me a bit, which is he said a number of things in the memo, but one of the things that's really kind of lived with me is justifying the non-use of AI around any incremental resourcing decisions. And I think that's really been a rallying cry for, for many leaders where it's really kind of a gift, as CFOs because it, it forces you to take a point of view on AI, right, in different functional areas. And I think that's really healthy for, for all organizations. If something's innovative, you should understand it and have a point of view. Is it the right tool now? Is it the right tool maybe a year from now? And so I think that's shaping hiring going forward. I think one thing that I'd say I'm seeing you know, generically talking with other CFOs and even in our own business is there seems to be a bias, particularly on the engineering side, where more native AI tool experience and maybe a slightly junior profile is sometimes more interesting than the really senior engineer profile. Not saying across the board, but I'm just hearing kind of things where where there's some focus on that because the native AI tools are so powerful that having that is more important than maybe some of the requisite experience of yesteryear. So I think the, I think the hiring profile is changing in short because of some of the AI tools we have at our fingertips.

Awesome. Sima, from everything you're seeing out there, what is one strategic capability AI will help finance take over, and, and maybe one finance team role AI will completely eliminate?

Yeah, I think this is a great question. I'll speak from it from the perspective right now. I think we're seeing the most resources towards accounting just because I think there's a lot of this going back to the on the same theme of manual work. There's just a lot of, you know, as finance teams build out their personnel, there's a lot of people to do different roles within accounting and staff accounting, or it's being outsourced. And I think that's really where you're seeing the meat of the resources being put today. And again, these folks are getting repurposed again to like do more strategic or like review work versus having to do the more manual sort of bookkeeping type work. I do think we're, we're still not at the point where you can just like get a forecast spit out, even though I think every team would love to have that. But it's starting with like, let's get the analysis, see where the drivers are, make a more intelligent forecast and get to that point where at some point the forecast will be created for you and you can actually just think about like, oh, here are the drivers going into the forecast and let's see how it iterates versus like someone needing to sit there and like code up the forecast and then make each change to each driver individually.

Ron, what's the most strategic decision you help some of your clients drive with Sapien? Maybe one, yeah, a big use case. Well, they want to take that question with everyone, uh, sort of thing.

There's one that comes to mind. It was actually one of the first customers we were working we were automating their attribution analysis, basically looking at their raw transaction-level data over the course of each year. And it was similar to gross margin analysis, but slightly different flavor for this business, generating their waterfall charts that went into their board deck. And so we were generating this analysis for them. We kind of configured some of the modules, and the output that we got from our system was off by $10 million. And as an early, like a very early startup at the time, it was It was not where you want to be, telling the CFO like, "Hey, the numbers aren't lining up by $10 million." I remember we were sitting there and we're like, "Are we sure that it's wrong?" We actually went and we looked at all the numbers, all the analysis that Sapien had done, and we actually found out that their spreadsheet had a small mistake in it that they'd been literally compounding on for 4 years. They were off by $10 million, not just this month, but every month for 4 years straight. This is a multi-billion-dollar menu. Manufacturing company. And so I called their CFO that morning. It was the day of their board meeting, actually. And I was like, hey, you might want to take a look at this. And we were still in our pilot, so he kind of wasn't fully taking me seriously. And I was like, you really should take a look at this. And he was like, no, no, it's fine. And I was like, give me 15 minutes. Hop on a Zoom call. We showed it to him. He's like, that's pretty bad. And it was even more interesting because immediately after they found this mistake, they actually— it was basically they thought that a product was making $9 million and was actually losing $2 million, they immediately reallocated their funding from producing that product to a different one, and their margin increased by several hundred thousand dollars in that quarter alone. So that was, you know, pretty cool early on thing that we've seen. And I would say it sounds like a one-off, but every company we've worked with, we've found, maybe not quite at that scale, but something similar where, Humans weren't built to work with multimillion-row spreadsheets. It's just like not what we were built to do. And the fact that AI can look at all this stuff, it can do calculations in a way where you can actually verify everything. That's pretty exciting. And I think it leads to better decision-making downstream. But it's that early part that I think is particularly exciting.

Rodrigo, the same question with you is what's the most strategic thing AI helped you drive? Maybe it's too early for Joby, but—

Yeah, again, the two things I already mentioned about the marketing capital allocation, again, $200 million is not a small, you know, change. But the other one was interesting is we kind of recently, we made a pretty big fee change for the company. So imagine that, you know, The company is almost like 15 years old. We have a take rate or a commission base from the sellers that was essentially the same and unchanged. And then we decided to experiment with that. We use AI quite a lot to see the types of configurations and the potential outcome from that. And it drove a pretty big decision. So it turns out that wasn't the right decision. We had to undo it. But the point being is we used it to look at multiple things. I think we made the call that was not probably the right call, but we were able to do an informed decision. And most importantly is to understand when things will go wrong so we could potentially pull the plug. And because of the power of AI, so we look at many scenarios, we're able to undo the decision very quickly. So used to make a very strategic, massive decision, but also use it to find some control charts that say things are not going well, let's put a plug. And we did. Awesome.

Brian, yourself?

Yeah, so for us at Demandbase, we, uh, we sell into the office of the CMO and CRO largely. So from a product perspective, a lot of our energy is into, you know, how do we make AI come alive for our customers using their first part, first party data and any third party data we have. And one of the things that we recently launched is an AI pipeline summary for typically executives where essentially it'll take their existing pipeline and then using intent data, technographic data, other predictive data helps kind of give them a quick summary of, hey, here's what's not in pipeline, which might be interesting. Interesting, and then sort of the trail around why that is. So that's been really well received. I think, you know, making it easy for customers to consume massive amounts of data is really part of our core value prop. So that's probably been the biggest thing for us as a business in terms of what we've been driving for customer outcomes.

So after this question, we'll take audience questions. So please feel free to line up behind the mic. And we'll start taking audience questions. Sima, what are maybe two AI tools? And maybe they don't exist. Maybe those startups don't exist, or maybe they do, that you're most excited about?

Or—

fine. Obviously, Sapien is one of them. No, I'll— let me put it in categories and then not pick out favorites, just to be fair. But I think, again, I'm going back to this accounting theme. There are two sides of this, I'd say. That are really interesting. One is actually on the like accounting firm side. We've seen a lot of companies, a few companies that are helping do the extraction like from, I guess, bank statements, PDFs, invoices, etc., create the journal entries and then use that as like a base for the work that an accounting firm would do working with the CFO's team. And then similarly, we're seeing corporate accounting teams be able to You retrieve this data and then, and from whether it's contracts or whatever, and then start populating audit checklists, et cetera. And so I think that broad category is very interesting. I think the other thing that's interesting too, I think this is to Ron's side, there is this, the idea of like an agent making, helping you do things faster. And so whether it's on the financial analysis side or, you know, helping close, like how do we sort of create almost like Zapier-like links between processes internally and start, yeah, again, shifting the finance professional from being a doer of that to being the reviewer.

Rodrigo, how about yourself? Maybe something you use or something you're hoping should exist, somebody should make it. What are you most excited?

Yeah, again, I'll stay to my theme about we use a lot for the descriptive analytics, look at what happened yesterday. I hope that the predictive analytics will be something that we can use more, whether that, you know, generative AI or something that has a dose of artificial intelligence. And then my vision is prescriptive analytics. Let me unpack. It's essentially like instead of say, hey, look, "this is our business today, can you help me understand what's gonna happen tomorrow so I can make decision?" It's almost like, "this is how I want to build the business, what do I need to do to get there?" It's almost like to get to a point that I can say, "okay, I want to prescribe that we're gonna be a billion-dollar business by X, what are the decisions that I should do to drive that outcome?" I think that's where I wanted this to get. I'm not sure how far we are because I'm still in the second inning, which is predictive analytics. But my vision is to get to a point that we can prescribe how we want a business outcome to manifest.

What are you using to help you with predictive analytics?

Well, good question. It's essentially human beings with a lot of tools. So we use Pigment. Pigment now has a layer of AI. We use our homegrown predictive predictive analytics. We actually even use this very old school crystal ball in Excel, which is a Monte Carlo simulation. Yeah, we have a triangular distribution for outcomes. And then we, instead of saying like, this is what's gonna happen, I'm gonna say, I'm giving you this decision point with 90% of chance that it's gonna happen, that this is the outcome. So again, crystal ball, Pigment and then a lot of homegrown. But now we have the layer of GPTs using the enterprise ChatGPT.

90% of the public company CFOs talk about crystal ball. Brian, how about yourself?

Yeah, I'm going to reach out into left field a little bit. I would love to have something more on the quality of hire, quality of talent side. I think headcount is the biggest cost in any business. And I think oftentimes there's this kind of lag between hiring someone, understanding whether they've been onboarded effectively, and then what they're doing. And I'd love to bring together kind of the, the TA side of things with sort of, you know, engagement and quality of hire analysis in the first 6 months to really tell me, you know, where we're getting like lift in the org, both from a recruiting velocity and then an onboarding and productivity analysis. So I don't think that exists out there, but that would be really interesting to me.

We are seeing a lot of recruiting tools on the way in the door, and we're starting to see folks who also do change the way that performance metrics are doing too. So I think that's coming to you soon.

Good.

Good. Love it.

Could we take some audience questions? Go ahead.

Hi, great panel. My question is for you, Rodrigo. You mentioned that you had your team, you got an Enterprise ChatGPT subscription for your team and had all of them learn. You also mentioned that there was some hesitation maybe by some of them. How did you overcome that? And, you know, learning what you did, how would you approach that for someone like me with a small finance team?

I'd do the same thing with.

Yeah, it's interesting. I was in another CFO conference and we were talking about it. It's still, it's something, sir, that I have not cracked the code. But for example, it's an anomaly today that when I get the report for the usage of ChatGPT, I'm still the number one and I have 150 licenses running the company. What I've done is, again, I had a 3-year, 3-day training for ChatGPT in Python for the entire organization. And then they started doing like, hey, this is, you know, employee of the month, you know, like try to reinforce the good behavior. But for some reason there is some adoption, but not as fast as I understood. So a couple of thoughts. I think there is a lot of fear of the black box. I think especially finance professionals we tend to be way more pragmatic and risk-averse. Number 2, and I am starting to think about it, but people are, and especially the close-minded ones, are saying, "Oh my God, I'm going to lose my job. This thing is doing what I'm supposed to do." And what I'm trying to drive in my organization is this is the type of work that I don't want you to do because typically high complex that's manually driven, that requires very little judgment, those are tasks that artificial intelligence should be doing. I do not understand yet, but what I'm going to try right now is, I don't know, actually, this is, I haven't, it's not getting the traction that I hope to get. And by the way, I do believe that especially accounting, I'm still like 80% 3 months ago when I met Somnia in this conference to today, I still haven't seen like a fully blown offering that I am like, oh, now I'll need fewer accountants that I can have those accountants doing something else like you're mentioning. Business analytics more like some strategic value-added tasks.

Reduce their deliverable time by 75%.

Maybe they'll start using it.

Any further questions?

Hi, my question is to Sima.

Thank you for everything.

It made a lot of sense to me. But my question is, we are a SaaS automation platform, and as a CFO, we analyze a lot a lot of matrices, you know, quarter by quarter. But because of the adoption of AI, have you seen new matrices coming into the business? When you look at your investments, when you look at your portfolio, have you started analyzing those investments on new matrices, or have we redefined the same matrices or changed the benchmarks of those matrices because of the adoption of AI?

So if I heard you correct, you're asking if there are like new metrics that are being created or used now to measure things, to measure the business performance.

And yeah, for example, I'm just— let's say we look at revenue per employee and all that. Now because of adoption of AI, the benchmark of that metric itself has gone up. And how much have you— when you look at your portfolio, I mean, that's where I'm trying to Sure.

No, great question. So we are— so I don't want that right now. I don't want the metrics to be changing themselves, although unfortunately the funny thing is they have. I think a lot of— we've seen the new versions of ARR continue to pop up. I'm sure you guys have come across this where it's like, you know, multiplying. We saw someone who had multiplied a single day's usage revenue by 365, call that ARR. Anyways, but TL;DR, the metrics themselves are like, should be standard. And I think, you know, that's for good reason, right? So like, you know, the company can look from quarter to quarter and we as investors can help and all of that. That said, I think there are a few things that are popping up in terms of A, what the benchmark is. In terms of what's changing. And then I think even the way a lot of these companies are pricing is starting to change. And so there are additional metrics being created versus, I would say, changes in the metrics we're used to looking at. To give you— to make that more concrete, a lot of the AI companies we're seeing are just growing much faster than ever before with far fewer people. So what does that mean in terms in terms of metrics. Obviously, top-line growth is a lot higher. It may be monthly contracts versus ARR, and so you don't want to call that ARR. You need to call that MRR. And a lot of times it'll be, you know, there's some sort of fast-growing wedge, a voice agent, automation, et cetera. And then there's far fewer people. If you look at a lot of companies that have gotten, you know, the— without naming names, I mean, we've seen companies that have gone to, millions and millions of actual ARR with like 8 to 10 people, which is unheard of before because they are using a lot of AI tools and there's so much demand and a lot of budget, whether experimental or real, at companies to try these tools. And then as a result, obviously sales efficiency looks a lot better. At the end of the day though, a lot of— if you're selling an enterprise, like the same enterprise metrics matter and the benchmarks shift a bit, but I think everyone's trying to get to the same point, which is a viable business that'll stand on its own two feet, have an exit. And I think that general narrative arc is the same. Thank you for the panel. My name is Pian Pian. I have a startup called Emotions AI.

So the AI can listen in sales calls and guide salespeople to help them close deals faster. Curiously, as you're looking at like go-to-market tools, what are the top things you'll be looking at?

What are the desired results from leveraging a tool?

Thank you.

Sorry, the question is what are the top things? What are the top results you're looking at when you're evaluating a tool to help your sales or CS team?

Yeah, I think Yeah, depending on whether it's sales or CS or maybe growth marketing, I think a lot of times with new tools, it's velocity, right? How it's changing from whatever the baseline is today. So that could be time to engage with new prospects. It could be time to close. I think that's something important. So anything that sort of ultimately gets to whatever your leading metric of success processes, which may be SaaS bookings or such. I think on the CS side, to be honest, I haven't been as engaged on any AI tools.

Oh yeah, I'm going to say what we try to look at is twofold. One is a metric of efficiency and quality, so especially in customer service. So there are plenty of tools there, and I think that's an amazing use case, but what we, when we were evaluating, we were looking at, would I need fewer human beings but the outcome will be better quality? Can I have an NPS score higher? Can I have the time to respond faster? So again, it's a little bit like how much resource I'm putting in and what's the output, time to respond, and what's the quality? Those are the 3 things that we look at. Again, that's customer service, but I'm going to say even if we are looking at other tools, I'm going to look at efficiency, output, and then quality.

Awesome. We'll take that as the last audience question. Yeah, please. 2-part really fast.

If you could make one investment in your team to upskill, change their thinking, could you point me in the right direction? Then when you look to hire in finance, is there 1, 2, 3 attributes we should be looking for maybe that are different than we used to look for?

I mean, I'll be looking at ability to predict the future with decision-making, the predictive analytics team. And then it's interesting, we were talking earlier today, like, can this person be malleable? Meaning, like, can I hire an athlete, because you're going in a world that stiff thinking is not going to work. So flexible thinker and ability to look into the future.

Those are the two things. Brian, and maybe some of you.

Yeah, I would, I would say intellectual curiosity, right? Especially with the new tools. You've got to be open-minded and then great judgment. I think there are so many things you can empower the team to do. And if they have great judgment, they're going to be probably supported to make mistakes from time to time, but also like really get a lot done at the endpoints of the business. So those would be my two, intellectual curiosity and great judgment.

Ron or Sima, would you like to add something?

I think the interesting thing that's shifted right in the finance team is the leadership has gone more towards strategic finance than accounting. And as a result, result, especially when we're looking at the early stage, the finance team is often hiring— you need— there's like technical roles. They need to know their things. But then there's also a little bit more around like all-rounded complement who can kind of help build out the function over time. And honestly, I think some of that ends up being like, yeah, you have a baseline level of knowledge, but it complements, you know, the really technical accountant with somebody who can learn a lot of things very quickly and then is is very, I mean, at the end of the day, this job is detail-oriented. Like, can they, can they help or, you know, build out the systems to make sure that like the rest of the team has visibility into how it's doing? Yeah, it's interesting.

I'm, I'm not a CFO, but I spend a lot of my time talking to CFOs and I've been pretty consistently surprised that the hiring profile that the best CFOs look for is actually very similar to what early-stage startups look for. I spent most of my time looking for engineers engineers or salespeople, but I think it ends up being the same stuff. The manual work that most teams are doing isn't that valuable today. And so to Sima's point, you want to find people who are just going to be adaptable and be able to use the new tools really effectively. I think the thing that I've also heard from a lot of CFOs is just they want people who can take any set of information and just give them some deeper, some narrative around it, some understanding of what needs to be done next. And once you have those two skill sets, everything else can be learned. But I think those two are the ones that tend— I tend to see CFOs see they have to be a little bit more innate in the interview process as they're evaluating. Everything else can be learned on the job.

Awesome.

We'll take one last question. Yeah, please.

Thank you. So I have a question regarding using AI for forecasting. There's a lot of conviction in terms of using AI for accounting. But then when we turn to AI for forecasting, usually there are, you know, large volume of data or there's a high complexity. So how do you make sure which AI tool actually is giving you the level of comfort in terms of accuracy, you know, instead of just give you some kind of a prediction that will show something, but it's hard to validate.

By the way, I have the same question as you do, so I'm here to learn, guys.

I can speak a little bit to this. I think this is something that we think a lot about when building an AI system for CFOs is that the answers don't actually matter. And what I mean by that is like the numbers obviously have to be right. But from your vantage point, the story around the numbers, understanding how you got to the number matters the most. And so I think when you're looking at AI tools, especially for forecasting, the most important thing is do they have observability, right? Do they let you see all the assumptions and edit all the assumptions that went into that forecast? Do they do the assumptions in a human way? I think this is a big one. There's a lot of tools that try to do AI forecasting and it's let's train some neural network, big classifier to give you your numbers. You don't actually want that. You want the AI that can figure out what assumptions need to be made and then build build a human-esque model, the same model that your team might build, but just do it faster and do it more effectively. I think there's very few that are doing that. We're actually, we're starting to do this a little bit, but it's also a very difficult challenge on the AI side. I think that's the key thing that I would think about is does it offer you observability and does it do it in a human way as opposed to in a black box way? I think there's not a lot there right now, but I think there's going to be pretty soon.

Yeah, I think very tactically what I'd say is when we look at software, we often talk to customers, we talk to you guys, and they'll very tactically say, we compared it against what we were doing internally and what was the level of accuracy. And I think the reality that all these tools have to overcome is, yes, humans usually make mistakes, and the software is getting held to a higher standard. That's just the reality of the psychology there. And so I think to everyone's Jason M. Lemkin's point here. I don't think the forecasting tools are quite there yet, but I think this is all like, how are we equipping people to do the forecasting more easily? And also like, yeah, the audit and traceability is essential, essentially because you're selling into a customer that thinks that way and they need to be able to trust what you're doing, not just the outcome. And so being able to see where it's coming from is really important.

I'll also throw in really quick, I think benchmarking forecasting is really hard because a lot of times companies that we work with want their, like, want our forecasts to line up with their forecasts, but their forecasts are bad. And so it ends up leading to a weird incentive structure here where, to your point, like, people are benchmarking against some result. I think that's the other important thing is when you're evaluating a forecasting tool, are there more objective ways to evaluate it? Like, can you give it subsets of historical data and see how it would perform on future data that's hidden from it? There's those sorts of things that I think drive a lot of value because you don't actually necessarily want the forecasting model to give you the same output that your team would. Because your team might be making systemic mistakes, you don't want the model that you're setting up to do the same exact thing. So I think that's particularly tricky with forecasting, but we also see it across the board with every type of analysis that CFO and finance teams want to be doing.

And the last thing I'll add to that is if you're selling into this team, like I I think if you have the ability to sort of show the advantage that you have and also guide the team in like, here's how other teams are looking at it, or here's how you may change, that also helps in the sale. I think a big theme we're seeing is changing the way work is done. So instead of saying, OK, I'm offering you X software tool to replace Y software tool, it's like, oh, I'm offering you X software tool to replace Y hours doing bookkeeping, or I'm using X software tool to change the way you're doing forecasting. That's a really hard sell to CFOs that they've been doing things in a certain way and it's largely been working. But I think there is more of a window now to actually have that collaborative conversation than there has been in the past.

Brian, would you like to add something to that in terms of forecasting?

Yeah, I just think there's value in multiple points of view on forecasting, whether it's a tool plus a leader, they might have slightly different methodologies. And so I'm a big trust but verify guy. And so I think the time series analysis around accuracy of the forecast is ultimately going to win the day and allow maybe both to kind of tune going forward. So I think it's helpful to at least have two, two modes of forecasting if you can afford that. Awesome.

Give it up for this great panel. Alright, thank you. That was fantastic. Thank you.

That was great.

There's another death. We talked about death of CFO. We're gonna talk about the death of Deloitte. Before I do that, is anyone here employed by Deloitte currently? Raise your hand if you work for Deloitte. Okay, no one works for Deloitte. That's great news. Let me, let me start with some context. I have been an investor in venture— I'm a venture capital investor. I've been investing in B2B software companies for 11 years at Emergence. I wrote a piece last year called The Death of Deloitte about the rise of AI-enabled services. And it was the first time in my 11-year history of doing this job that I received a cease and desist letter. Our friends at Deloitte did not like the branding for this. They're apparently not fans of alliteration. Yeah. I don't have anything out for Deloitte in particular, but I do think that the Big Four and professional services organizations more broadly are in danger of being disrupted by AI-enabled services. And so that's the spirit of today's conversation. So first, some context. So this is a bit on our firm. We've been investing in B2B software companies for 11 years. We actually backed Jason M. Lemkin's company EchoSign at the Series A. So we've known Jason for a long time. Um, we've done a lot in G&A, obviously Build.com, Intact, Ironclad. Um, so we have a lot of perspective on this space. Um, but what's interesting is all these companies have only been able to capture software budget. None of them have been able to capture labor budget. And the labor budget is much larger than the software budget. And so, in some ways, this is the most exciting time to be investing. In the entire software stack, but particularly in the G&A software stack because there is so much labor spent that you can capture. So we think that LLMs can enable software to eat away at services, but it's a big question on where to start. There's a lot of potential areas to go into, but there's also a lot of potential pitfalls. So today I want to talk a bit about some of the areas that we think are exciting to eat away at the Big Four. There's $200 billion in revenue between the Big Four annually, and that's not just audit and tax. There's also a bunch of software consulting. A lot of money. Um, they— I believe Accenture has 50,000 software developers just in-house, not counting outhouse. So there's just a massive opportunity that has historically been off-limits for software and software investors that we think is now investable. So, um, like any good ex-consultant, I put together a 2x2 to help me understand where I think there's opportunity. In terms of how AI-enabled services can be disrupted. You guys can't see the x-axis because it's below the chairs, so I'll read it to you. In general, I think you could think about services that are provided on, on two axes. The bottom axis is there's a bunch of services that are what I call help you do it services. So these are things like advisory, like M&A advisory, investment banking, that type of thing. You want help doing that service. And there's a bunch of services that are do it for you. You don't want help, you want someone to do it for you. So this would be like auditing, certain types of software migrations, et cetera. So this is like one axis in which you can understand services. The other is repeatability. So there are certain services that are super repeatable where over and over again it's the same thing. And there's certain that are fairly customizable and every time it's bespoke. So when you think about where to build an AI-enabled software company, We think that the highly repeatable do-it-for-you tasks are the most exciting. So there's a lot on this chart, and let me explain it in a little more detail. So this sort of breaks down all of the tasks being performed by the major professional services companies. You've got legal services, transaction services, financial accounting auditing services, IT development support, and of course strategy management consulting. All the stuff that tends to be more advisory, bottom left, our perspective is that we think in the near term that's less likely to be disrupted by an AI-enabled services company because people still want, people still buy services in that category for advice. And software today, AI today is really good at doing analysis and providing you advice based upon that analysis, but it doesn't have the historical track record that you're paying for when you're hiring a consultant. On the top left, you've got the help you do it stuff that's repeatable, but we still think is unlikely to be disrupted in the near term because, again, people are buying not just the outcome, but they're also buying the advice they get along the way. On the right-hand side of the chart, you've got tasks where people just want you to do it for them. So they don't want your help, they just wanna outsource it and get it done. Some of those are less repeatable. Financial due diligence, you could argue where any of these go, but many of these we think tend to be a little bit more bespoke. Top right, you've got services that are do-it-for-you and are generally fairly repeatable. So tasks like bookkeeping, financial reporting we talked about in the last panel. There's a bunch of in the IT development stack that we think is likely to be do it for you, and highly repeatable. An example of one company that we've invested in that's in this category is called Mechanical Orchard. Mechanical Orchard is using AI to do cloud migration. So specifically, 75% of Fortune 500s are still running on mainframes, and those mainframes are aging, obviously. Deloitte, etc., have been trying to lift and shift those mainframes for a long time, but it's really expensive. It takes a long time when you're doing it by hand in a human way, and it's also unlikely to succeed. It's about a 50% success rate. This is a perfect use case for AI, because AI can analyze all of the COBOL code that's been written in, on these mainframes, and can rewrite it in a cloud-native format, and then host it in the cloud. So, this is an example of a company in this cloud migration category, that is disrupting a number of these, uh, legacy vendors. They're able to provide the, the service faster and, uh, at a higher success rate. We're trying not to charge anything less, but when you're thinking about an AI-enabled service, you either have to be some combination of cheaper than Deloitte, faster than Deloitte, or have a higher success rate than Deloitte. The most exciting startups are the ones that are doing it faster and with a higher success rate. But maintaining price. 'Cause they're also generally able to operate a higher gross margin because they don't have as many people doing the service. And that's actually what makes these things investable for venture capitalists. Historically, we haven't been able to invest in services businesses because they don't scale linearly. Costs scale— sorry, because services do scale linearly, meaning your costs scale linearly with your revenue and your gross margins generally can't. become, you know, the software gross margins, 80% plus that we're used to. We think in an AI-enabled services world where you actually are able to productize the service, you can have an 80% gross margin. So imagine if Deloitte was charging $5 million for a cloud migration service and they made $500,000 on that over time. If you're able to do it much faster and at a higher gross margin, you could conceivably take home 3 of that $5 million as your gross margin because you have way fewer people performing the task. And all of a sudden, this becomes an extremely attractive business opportunity for VCs and beyond. So that category, cloud migration, obviously we've already made a bet in, but there's pen testing, there's maintaining apps, and basically anything in this category, we think there will be an AI-enabled services company built, and we think it's likely to be an attractive opportunity. So let's talk about the CFO stack, 'cause you guys are all CFOs. When you think about the next-gen CFO stack, what's interesting is you can also bucket, these are all software companies that are doing FP&A, procurement, accounting, audit compliance. We gave Sapien a shout out up here because Ron's here. But when you think about the tasks these are performing, these tasks tend to correlate with help me do it versus do it for me. And what's been really cool to see is that naturally these startups have generally gravitated more towards a SaaS model in the help me do it land. And on the do it for me land, a lot of these companies on the right tend to have more of a services bent or maybe a hybrid SaaS services angle, AI-enabled services angle. I think in part because it's this do it for me mentality where you want to outsource it completely, audit compliance, being good examples of that. So, the, this, this is a very kind of new concept, this AI-enabled services concept. This wasn't really possible before LLMs, so there is no massive company. Obviously, Palantir has been built from a services business into a massively successful software services hybrid. Still trades, I think, at amongst the highest multiples of any software company. But that was all started before LLMs. So, the question is, now that we have LLMs, could you build an even better Palantir? And that's my, my job is to try to find those companies and obviously invest in them. So what I wanna share next are some lessons I've learned from investing in AI-enabled services so far, but it's not just necessarily lessons as a VC or a potential startup founder. Hopefully also for the CFOs in the room as you're considering buying services from some of these next-gen AI-enabled vendors. This checklist is a good thing to look at. Because it's, I'm sure it's really hard to do your job right now because there's a million of these fancy logos and like how do you sort through which one is special if they're all just sitting on top of ChatGPT? So the spirit of the next slide is to try to help you think through that, particularly as they're promising you a service, which is different than evaluating a software product. Software product, you can play with the product and see how it works. A service, you're really just trusting someone to say like, trust me, I'm gonna move your cloud your, your mainframe into the cloud or something else. So how do you actually build confidence to make the bet on an AI-enabled services startup? All right. So a few lessons from watching these companies in the first innings. And again, we're in like inning one of this, but these are the initial lessons. The first is to evaluate if the team has domain expertise and I think this is much more important in AI-enabled services than it is in SaaS land. In SaaS land, if you have like a couple Stanford CS dropouts who are super impressive, um, that might be enough to build a great company. We've seen that over and over again because you can build a product and then have the users use it. In the case of a service, they're not buying software, they're buying you. So it's harder for a CFO or anyone else spending money on this to trust that someone who hasn't actually performed the task themselves will do it well. And so when you're evaluating these teams, if you're thinking about starting one of these businesses, domain expertise is much more important, I think, in AI-enabled services than it is in historical SaaS. It also unlocks ideally access to high-quality talent channels because you are gonna have to hire certainly initially doers of the job. 'Cause generally speaking, when you launch the company, you don't have all the AI fully fleshed out. So you start as kind of an AI historical services hybrid. And if you have someone with domain expertise, they can hire their ex-colleagues. So in the case of Mechanical Orchard, the founder of that company started something called Pivotal Labs. And if folks remember, I got some nods in the audience, Pivotal Labs became a very large tech consulting firm based here in the Valley. And so Rob, the CEO of this company, is very credible with CIOs 'cause he's been selling to them for 30 years and has a bunch of people he's hired that can do the coding alongside the AI. So the domain expertise has been really helpful for this business. Okay, this second one is probably my favorite one. If you're building an AI-enabled service, product market fit looks different than if you're building a SaaS company. Historically, if you're building a SaaS company, the logic is if I'm growing super quickly and my customers are retaining, I have product market fit. There's demand for my thing, people continue to buy it, et cetera. In AI-enabled services, it's a different thing. Because if I come to you and say, hey, I'm gonna sell you a cloud migration product and it's gonna be faster, uh, than the incumbent and maybe even cheaper than the incumbent, you can sell that all day long because you know there's demand for cloud migration services. And if you're willing to do it cheaper than the incumbent, it doesn't mean you have product-market fit. It means you're able to sell you're, you're able to undercut another service provider. Product market fit for an AI-enabled services company is when the AI itself is able to perform the job, the service, at a high gross margin, and you're able to sell that and retain that, and it works. And I think a lot of what we're seeing in this early innings of AI-enabled services are companies that are selling really quickly and then discovering, oh no, I don't actually have enough AI here to really build an AI product, service. So I'm actually just building a services company with sort of lipstick on the pig. And so I've got this graph here that kind of illustrates it. This is like the early days of software, of revenue growth in some of these businesses. And then what you see is, um, often the best ones slow down after they get their first customers and stop getting new customers. So they have time to use this first cohort to develop the AI product. And then ideally the AI product gets developed, they start going to market again, and they're able to sell with a delta between the revenue and the cost, and obviously much better gross margin. This blue line is historical services businesses, which is the linear concept I illustrated before. In the nirvana state of AI-enabled services, you get to this, obviously, disconnect between revenue and cost. Unfortunately, what's happening for a lot of these businesses right now is you kind of here, and these lines continue up, and so you're not actually getting the gross margin that's promised in an AI-enabled service. 'cause you're kind of drunk on revenue and you're not actually building an AI product. So anyway, this one's sort of counterintuitive. You actually want to grow a little more slowly in the, like, in inning 2 of these businesses so that you can productize off of the initial customers. Okay, number 3 is partner early. And this one's a little controversial. We talked about, like, we call this is the death of Deloitte. Um, but I actually think a lot of these AI-enabled services businesses do well to consider partnering with incumbent services providers at the beginning. And the reason that's the case is because you need to trade off of someone's credibility, particularly if you don't have tremendous domain expertise yourself. But if you're selling a service, you need someone who is trusted by the buyer to say yes. And so what I'm seeing is some of the most successful AI-enabled services startups are partnering with services incumbents early on to do some sort of rev share or maybe even equity share and grow their company. This is most successful with smaller incumbents. Less the Deloits and Accentures and what have you, but if you go downmarket and look at the smaller SIs, generally speaking, startups are having more luck partnering with the smaller SIs because the smaller SIs care more. Whereas the Deloits and Accentures of the world, like, it's not, it's not mission critical for them in the same sense. So consider partnering early if you're building one of these companies. Okay, outcomes-based pricing is a fun one. It applies both in AI-enabled, AI-enabled SaaS and AI-enabled services. Generally speaking, when you're starting to sell an AI-enabled services product, you don't want to innovate on the pricing model because you're already innovating on the product and the customer's already probably freaked out about that. So my advice is start with whatever the pricing model was, which is often hours-based, labor-based in a services business. And then once you build trust with your customer and you know enough about your own cost structure and how good your AI is, you can move to outcomes-based pricing. And that's disruptive where I say, you know what, instead of charging you, you know, on the basis of this is going to cost, this audit is going to cost 300 hours and it's going to be 300 times $200, whatever, this is the price. I will guarantee you a fixed rate cost once I complete the audit. And no matter how long it takes me, you're just gonna pay me that. I'm obviously betting on myself at that point and the AI to do a really good job, 'cause if it doesn't, then I'm spending a ton of money and have a super low gross margin, sometimes negative gross margin. So I encourage startups to think about doing this, but doing it after you've proven yourself with the initial standard pricing model and you've learned enough about how good your AI is to be able to pull this off. Okay, lastly, um, It's the demo, stupid. And what I mean by that is this is probably the most common failure mode I've seen with AI-enabled services companies today. You build a service, the service has some AI behind it. At the end of that experience, you're like, hey, buy my thing. Like, I've got all this cool AI. Here's my presentation. I'm showing you all the ways I'm using AI. And the customer says, but where's the product? And you say, I'm the one using the product. Like, I built it and I'm just selling you a service. I'm gonna sell you an outcome. Don't worry, I'm using the product. There's AI behind the scenes. What I've seen over and over again is the buyer still wants to see the AI, even if the buyer themselves never touch the AI. And like, for whatever reason, AI-enabled services founders cannot get this in their heads. I think because it's logical to say like, don't worry about the AI because I am using it. You are never going to touch it. And often there's no pretty interface that can be go— that can go towards the outside because it's built for internal consumption. But what I've seen is once these companies build a demo, build a customer-facing UI, even though the customer will never touch it, sales cycles go much, much faster. Because people, buyers still want to see the, the AI and the cool whizbang thing behind the scenes, even if they themselves are not the ones that are going to be using it. So this is like the easiest possible thing you can do in building one of these things. And if you are evaluating an AI-enabled service, you should ask for the demo. Because it's hard for you to— if someone comes to you and says, "Hey, I promise I can do your accounting faster, and I'll take it all on." Before you transfer all of your accounting responsibilities to them, you probably do want to understand the technology they're using, even if you aren't the one that's going to be using it. So I would encourage you to push for a demo, even if there is not one. So anyway, that's it. That's— That's all I had for the presentation. I'm happy to answer any questions folks have on building these companies, failure modes I've seen, things I've seen work, et cetera. So feel free to ask questions if you got them. Yeah. You want to go to the mic just because it's easier?

Hello.

My name is Fernando Vitti. I'm the founder of NextForce. We are a group of companies and one of them is a professional services line of the business. And I'm wondering, what are the companies that you have met that are doing good in replacing humans in these professional services environments? Can you list a few company names and what they do? Yeah, sure, for sure. So as I said, it's, it's like work. In like month 12 to 18 of these businesses. So I can tell you businesses that have been funded, but like how well they're actually doing is still TBD. And I think the other challenge with this is back to this point, there are a number of companies that I think have grown really quickly, but we don't yet know how good the gross margin is, how good the product actually is. But obviously I mentioned Mechanical Orchard in the cloud migration, you know, And I'm on the board of that one, so I have visibility. The AI does seem to be working, but I'm obviously biased. On the customer support side of things, there's companies like Crescendo is one that is doing fully outsourced support. So instead of buying, I'm on the board of a company called Assembled, which will do the, it's a help you do it AI support company. There's also do it for you AI support companies that are more like services where it's a BPO. That you've kind of outsourced, if that makes sense. There's kind of a long stream, but like those are two that just come to my mind. Yeah, go ahead.

Hey, great session.

I had a question around like how do we think about who do we partner with?

And you said the larger companies are facing more of an innovator's dilemma and the smaller ones are more amenable to change.

So if someone is building on—

basically service as a software, right?

How do we think about, like, do you start bottom up?

Do you go top down?

I know you said bottom up, but can you throw some more light on it? Yeah, for sure. First of all, you used the term service as a software. I really don't like that term. And the reason I don't like that term is I think it's super confusing. It doesn't actually say anything. It was coined because it sounds like SaaS, because it has the same alliteration as SaaS. But what does it actually mean, service as software? It doesn't— if you— it doesn't actually— it kind of breaks your brain to think about what that means. Which is why I use the term AI-enabled service, 'cause that's super descriptive and very specific. So anyway, small bugaboo, but words matter. Why go to the smaller SIs versus the bigger SIs? The bigger SIs are gonna waste your time, is the long way to, like the short answer to it. Every big SI wants to talk to you right now because all of them realize, I mean, I got a cease and desist from Deloitte. So like all of them know this is coming and care, and they're all trying to figure out what to do. So they're talking to every startup. The odds that they are going to partner with you and introduce the customer to you and give you any sort of favorable economics in that process, if they do that, are super low. I just have not seen yet a company succeed in getting an Accenture or Deloitte to be a truly good partner if they're a services business. I've seen it on software, right? If you're like UiPath, yeah, that worked all day long. But this is weirder because you are a services company, so you are competitive. With them. So it's more of an awkward context, whereas if you go downmarket, these kind of midmarket— if you look at an SI that has like a billion dollars in revenue, they, they're more incentivized. They really need to do something, and it's a little more existential, and they have less time and all the rest of it. So they're more likely to move, is the, is the long and short of it. So that's kind of my, my take. There's a whole other approach you can consider here, which is buying an incumbent, and that we didn't talk about today, but That's a popular approach. Crescendo, the one I mentioned before, has bought call centers and then they're trying to inject tech into it. That's like an AI-enabled roll-up strategy, and that's a popular strategy. I think there's challenges to that strategy because you're trying to jam together an ops thing with tech. My own belief is that it's a little bit easier to start with pure startup and go from there and then partner with, you know, one of these more kind of mid-level Yes, Isaac. Thanks. Yeah. Hi Jake, it's Daniel. We traded some LinkedIn messages. Yeah, yes, yes. I work at Constellation Software, so there's a lot of questions I actually want to follow up on that around partnering early. Cool. One is, you said avoid the big companies for distribution. This is advice for startups. So one is, how do you identify what's big? Are we too big? I don't think so. Number one, we can help. And number two, what advice do you have for founders when you think about your portcos that have successfully partnered early for distribution with larger companies that already have that customer trust? Yeah, well, what you guys do is different. So an SI is different than a private equity firm. You used the word roll-up and I was like, ooh. Yeah, that's, yeah, exactly. But like a private equity firm, I have seen it be effective if you can get the private equity firm at the top to mandate that something should be adopted by other, for their full portfolio. That has worked. I've seen it promised more than I've seen it actually delivered, but I have seen it delivered in certain contexts when the portfolio companies, when the private equity firm is super involved in what's going on with the companies. The most successful channel partnerships I've seen in my time in venture were actually with like corporates. So like one that comes to mind is Bill.com, which is a company I'm sure everyone here knows given the CFO thing. So, Bill.com had a few years where it wasn't actually going very well. And the thing that helped Bill.com reaccelerate was a partnership with Bank of America. So, Bank of America agreed to integrate the product in a bunch of Bank of America's offerings, and that completely changed the trajectory of the business. And they eventually partnered with a bunch of other banks, and that took off. Intact, which is another one that sells into CFOs, which is an ERP company, That one took off when it cracked the accountant channel, when it got accountants to recommend Intact to their customers, and that business ended up taking off, was eventually bought by Sage, and now it's called Sage Intact. So to answer the question, Daniel, my, in my experience, the most successful partnerships I've seen for startups have been with these like larger corporate incumbents and or individuals that have incentivized to push. I mean, another great one is, is HubSpot. So HubSpot ended up— 40% of its revenue when it went public was through channel partners, and the channel partners were these tiny marketing agencies that were using HubSpot themselves and then selling it to their customers, which is sort of similar to the Intact model. So I guess there's a lot of different approaches one could take, but I'm sure Constellation has a bunch of cool ideas too. I guess my question was a little bit loaded. I used to run BD for Quicken, and one of the biggest concerns about giving exposure to startup. I'm also a former founder, so I'm trying to double-sided here, which is how do you get that distribution channel to bet on you? Now, I think what's interesting is QuickBooks, Salesforce, they have these existing platforms where you can plug in, where that's part of their ecosystem. But for, to do, for Bill to work with B of A, and Rene's awesome, he already did PayCycle before, so that helped. Like, that's, that's the type of insight I'm curious about that you would that you've seen? Well, there's the top-down. There's like the Rene who's like very credible in the space, and so it's gonna be easier for him. But at HubSpot, it was just like my wife helped run the program, so I know it fairly well. Yeah, talk about— no, but my wife and the people like Pete Caputa and the others that helped run that in the early days, the partnership program for HubSpot, there was no like special— there wasn't like a famous person that knew all the marketing agencies, 'cause like by definition, they're super distributed. It's like some random agency in Iowa. It was honestly with them. They created this inbound movement. If you guys remember this in the early days of HubSpot, where like the concept was instead of outbounding everyone, get people to come to you. And so we're gonna help you with content creation. Then they create a conference and it became this like, get all the marketing agencies in Iowa to come to Boston for this crazy conference with a cool concert and all the rest of it. They created a movement and then they built this like cool channel partnership. So the exciting thing about BD and channel partnerships is like, You can do them in all sorts of different ways. There is no one-size-fits-all approach. Any other questions? Yeah. I've got the negative countdown, I see. So I'll stop soon.

Thank you again, Andy. Work at Ontra, Ironclad adjacent. Great. Yeah. We— this chart is really interesting.

Pricing as you invest, you think about crossing the chasm with product market fit and you think about pricing. What are some of the themes or KPIs that you're seeing where you transition from the service-based pricing model to a true SaaS pricing model? Is it just truly the gross profit margin being realized from the AI or are there other things you're also seeing to kind of be on the cutting edge of that to make sure you're capturing all those gross profit dollars? Well, the, the gross margin thing is as much about cost as it is about revenue. The revenue side of things is also really important, which is how do you— I think one of the key things, think about transitioning from services to SaaS from a business model is what's recurring. Historically, services have not been recurring. So the question is, if you're selling a service, can you find a service you can sell that is recurring? Or you can, you can sell, you can cross-sell a software product, which is actually what Palantir did. Palantir's brilliance was like, they'll come in with a forward-deployed engineer. And say, I will build you a custom BI for your specific situation. And they do it. And then like a year later, they're like, hey, do you want me to run this BI for you? Like on an ongoing basis, it's going to cost you X dollars per year. And then all of a sudden they like land with these big consulting contracts, but they have this tail of software. Mechanical Orchard, the one I mentioned before that's moving mainframes in the cloud, we do the mainframe movement and then we host it for you. And that's just gives us recurring revenue. And so, like, I think some of the most successful businesses that do this find ways not just to get high gross margin from a lowering cost perspective, but they also figure out a recurring revenue model. Thank you very much. Yeah, you're welcome. And thank you all. And thank you to Lloyd, wherever— there's Lloyd. Thank you to Lloyd for organizing this whole thing. Let's give Lloyd a round of applause. Thank you.

All right.

I think there's an after-party somewhere. There's There's drinks in the back room there, and I think Jason M.

Lemkin—