From Pilot to Revenue: How Sales Leaders Scale AI Beyond the Demo

Good morning, Jim. Good morning Gerhard and everybody. Jim, who needs coffee with that introduction? So you got me pumped up. I'm quite excited and maybe I can carry that thread forward. I'll do my best. So I'm going to pull up some slide. Any good consultant? I got a couple of slides. I'm going to try to pull it up here and as I get it up, I will just start with a quick introduction. So I'm Abe Avasti. I lead our sales transformation business at Deloitte. I have the great pleasure of having known Gerhardt for some years now and getting to participate in this and other forums similar to it. And I want to show you something specific today, which is not just what you can do theoretically with AI, but how do you actually build it into your revenue motion starting this quarter. And you know the thing, the thing that always comes to mind for me that I think about a lot is what, what does this mean for businesses? What does it mean for us as humans? And I end up landing. Jim, you know, pulling on what you just talked about in that we're going to create new frontiers of humanity. I think there are, there's a lot of folks concerned about the role of, of the machine taking over. I, I call me an optimist. I'm the AI optimist who's excited about what's to come. So maybe I'll talk it through what that might look like and, and then leave some time at the end for us to talk about it. So let me, let me start by showing you something. You know, you're, you're a rep and you just finished your meeting Tuesday morning with your prospect. You hang up, you walk to get coffee, and by the time you sit down, your call got transcribed. You've got an account record that was created in your CRM. Your contact record was created, or at least it was found and matched so that you don't have duplicates. We've created an opportunity for you. The AI agent has then checked your zoom info or other data source and it's found out who the buying group likely is for the solution that you sell and who the right buyers are in that organization. It's mapped your relationships within your organization. It's built a slack channel with the right people that you need to work with to go work this deal. And then it's drafted an email to your colleague Sarah for an introduction to the CFO who you know is going to be an influencer in this deal and who they've interacted before. It's filled out your Med pick forms based on what was actually said in the conversation. You know, what do you have, what's missing? And then it's identified the next two specific next action, not 20, but like what's the next one or two things you need to do. And here's the important part, the rep's not out of the loop. They have to review everything. They override the stage. If they don't agree, they rewrite the email to Sarah in their own voice. The agent has built the foundation, but the rep has to make it theirs. And that's what an agentic first process design means. And so you know, the thing about this is that it's, it's not a, it's not, this is not a minor change. This is going from on the left things that often don't happen at all. And if they do take a long time to. Now an AI agent is giving that sales rep so much more information to have so much better, to be able to deliver a much better experience to that buyer. And every single piece of this is buildable. Today I've deployed pieces of it across multiple clients. The transcription, the CRM automation, the buying group intelligence. It's commercially available right now. So the question is why isn't anybody or very few running the full system? And it's because we keep buying AI as features. A copilot here, a summarizer there, a chatbot in the corner. But they're not connected. None of them are operating as a workflow or a system. And so we're going to spend some time from here talking about how do we get system level results as we work through this. When AI pilots stall, what we're seeing across industry is it's usually not the technology, it's actually one of these five things. It's either that we're picking point solutions instead of end to end workflows, we're measuring time saved instead of revenue impact. And then we're surprised when the pipeline doesn't move. We're bolting AI onto broken processes. If the workflow was bad before, AI is just going to make it bad faster. We're skipping change management and wondering why we have resistance in the field. And then we're building data, we're building on data foundations that aren't ready. And so agents, AI agents will often amplify whatever's in your data, good or bad. And it's usually not one of these, it's often three or four at once. But all five of these are solvable. And I want to talk about really how so Maybe we start with the shift to be made. And I would propose that every sales initiative that I've seen is starting with the wrong question. It's often starting with how do we make the reps faster and more efficient? And that's not a bad question. But I think the right question to ask is how do we make the buying experience so good that the deals accelerate on their own? Can the buying experience, can the buying group, the company that I'm working with, reach consensus faster because we gave them better materials? Do they have personalized value articulation? Are we proactive in the ways that we just couldn't be before? And that's the inversion. And I'd encourage everyone to use that as a filter as you go through this process of integrating AI into your organization. So before you build a requirement stock or talk to a vendor, think about does this help buy the buyer reach consensus faster? Is it giving them better information faster? Is it removing friction from their process and ours faster? And if I was a customer, what I want this experience and if we can't answer yes, we should go back to the buying journey. And everything I'll show you today sort of flows from this idea. Maybe start with a quick case study here and then we'll get into how this organization, you know, large kind of modern tech platform, they had an SDR team that was spending 60% of the time not even being on the phone with customers or not emailing with customers. They were doing research. They had a very specific playbook. So they spent a lot of time ensuring they're following the steps of the playbook. They had a, you know, 20% response rate, which is not necessarily terrible depending on what industry you're in and what you're selling. But they were, they were targeting generally the procurement person, person within their customer base or prospect base. So we built a multi agent system where the agent does account research. It identifies the buying committee, like who are the different people? Not just procurement, but business buyers and IT buyers that they might be interested in. Finance teams as an example that we use zoom info for that they drafted personalized outreach that included projected savings. So an agent actually looked at know for this company what is the personalized value proposition? I could do. And the results were phenomenal. Response rates went to 50% pipeline improved, quotas were up 20% over the next period because they were engaging the right people and they were building a value proposition. Now the next phase, as they scale this is to actually have direct outbound from any AI agent, voice and text, both to lower Value outbound. And, and that's sort of the autonomy spectrum in action, if you will. The agent starts handling the high volume, lower sales, the lower stakes, outreach. And the reps are really focused on those that require a human touch. And you know, it's because they started with what's this end to end workflow and how do we actually change it? So let's spend some time here talking about the playbook. These are the five steps and I'm gonna walk through each one of these on subsequent pages. What does it look like in practice when you actually do this? And so maybe we'll start with number one, you know, the buying journey. And you know, before you even pick a use case, before you evaluate a vendor, I would suggest map your customer's buying journey. Not just sales stages, but their buying experience. Do research on this. Recognize that the journey is also changing as we speak. The idea that the front end of the buying process has become more self service and out of the hands of the account executive is not new. That's been happening for years and Jim brought it up as we opened here. But it is being accelerated. Buyers are talking to their LLMs now to decide what solutions are most appropriate for them. They're researching solutions, comparing vendors, building shortlist, the account, the account executives aren't even in the room for that. And then the, so the full front end of the funnel shifted. But then on the other side, we're also seeing in later stages buying processes getting stuck harder to build consensus, buying groups that can't align. Because there's so much change in the world right now, it's very difficult to land some of these. And that's a friction point we can start to address. You know, personalized business cases for different stakeholders, automated ROI models tied to different business buyer priorities, faster delivery of the right materials. You know, we think here about the process friction that happens with contracting and red lines. How do we move that along faster? So map the journey, find the friction, think about that. You'll find where AI has the highest leverage. So it's not just your internal task list, but the moments that matter for your customers. Step two, we start to think about the actual workflows and you can't do everything, so where do you start? One reasonable way to start is start with deal intelligence and progression. It's the Tuesday morning workflow that we started at the beginning. And why? I'd suggest because bad opportunity data is the root cause of all kinds of other problems. If you have bad forecasts, you miss follow ups. Deals get stuck in Pipeline, your data foundation's bad, so you can't use it for other AI agents. Next we might think about outbound and pipeline generation. It's increasingly important in this difficult world to have a sufficient size pipeline because of what we talked about in terms of buyer stagnation in later stages. And then we might look at pursuit, intelligence and proposals. How do we get the right proposals done faster to the right customer? Each of these is a multi agent workflow. It's not a single tool. And the principle is the same research that end to end process, build the agent system, keep humans in the loop at the right moments. Think of these as systems and it doesn't have to be disorder, but you know, think about which ones matter for your organization. From there you're going to then go to what is the workflow of the future. So what will the agent own, what will the human own? And the lens I use here is if it's data retrieval, pattern matching, templated actions, coming up with a plan, analyzing information. There's a lot of work that AI agents can do very effectively. Nobody needs to be manually creating accounts or duplicating contacts. You might have fallouts, but AI should take a lot of that on. If the output is customer facing requires judgment, the agent could draft it, the human might approve it. This is having humans in the loop and sort of what we call semi autonomous flows. And then if the moment, it involves trust, negotiation, internal politics, the human's got to own it. But an agent, an AI agent can support that human. So they can, you know, they, they might be sending the, an AI agent might send a follow up email to a, you know, small business or mid market lead depending on the size of your company. But it's probably not crafting and sending autonomously the message to a Fortune 100 CEO on a $20 million deals. So the principle is simple. The higher the stakes, the, you know, more the human, the more human the moment needs to be. And you know, don't take the thinking away from your, your field. You know, Jim talked about humanity and you know, what does this mean for us? And one of the things I would say with AI is that the, the our sales teams or us as professionals, as we walk into a meeting, if we haven't done any preparation because our AI agent has done it for us, we won't be credible in front of customers. Agent built, our AI builds foundations for us, but the rep has to make it their own. And this boundary model also can apply to task allocation. So if you think about trust and risk you know, when you run autonomous agents and you start to run that at scale, you can introduce biases and, you know, biases in targeting language or in outreach decisions. And when you have agents starting to operate at scale, that trust and bias can, can, can create problems, you know, in a, at a thousand, not just one email because of the, of the way that it can be used. I mean, you have a problem with AI, that problem might hit a thousand customers very quickly. So you got a design for this. You know, you can use AI similarly to this spectrum here, AI to monitor other agents. You know, AI agents monitoring agents for policy compliance, flagging anomalies for human review. That's human in the loop, and then ensuring that humans have a role in actually managing the agents you deploy. Those are new roles of tomorrow. And AI agents will have to earn their autonomy over time. That's the other thing I would say in this spectrum, during your prototype, humans are reviewing everything. During a pilot, you're expanding the autonomy envelopes based on what the data is actually showing. And then when you get to productions where it works and it's proven AI can, can run with those steps, humans in loop are in the loop on all the rest of it. So you don't flip a switch. It doesn't go supervised autonomous automatically. It's earned phase by phase. So let's sort of get, try to get specific on this use case now. So we, we talked about our Tuesday morning workflow for our sales rep and everybody starts to ask, okay, well, so what do I buy? And let me kind of break, break this down. If you, if you break down what's happening in that flow, there's capabilities, right? You've got call capture and transcription. You know, you can do that through your, your telephony platforms. And gong, you've got real time call intelligence, potentially with your rep. There's a couple different solutions out in the market to that. You've got CRM automation, you've got intelligence and data, orchestration data. All the things that you have to manage. And you know, I, I think what, what there's a principle here and that principle is don't buy seven tools for seven capabilities. You know, anchor on the one or two major platforms. You're going to have a cloud in your environment, at least one. You're, you know, whichever cloud you're on, you're going to have an LLM, at least one that you work with. You're going to have an orchestration layer. Whether you're using like a crude AI or using LangChain, you're probably going to have a CRM and an erp. You already have these major platforms. Start there, anchor on those major platforms and then be really selective what other best of breed vendors we want to add to that because it adds integration complexity, license cost, data fragmentation. That isn't to say there aren't really important vendors out there. I'm saying be selective on how you do that. Also think about scalability and cost. You know, think about the trajectory in three years. Will this point solution I'm buying justify a standalone use case? Will or will this capability be native on one of my anchor platforms? Am I buying this for a period of time or for long term? What's the cost to do of managing that change and then who's managing all the risk associated with this? So do I need new roles? You know, do I need somebody who owns AI governance across my sales, motion monitoring, agent output, managing data security, evaluating risk with you know, anthropic or open AI APIs? Some people call that AIOps, other people embedded in rev ops. But there are new roles to be had and, and then you know, I guess last thing I would say is that you know, in the buy versus build calculus here you, you have to think about the governance and your capabilities associated with that. Maybe that's a good segue here to the next slide. There's kind of three decisions you make, right? So you know, really when you, when you're thinking about it, you're looking, you're thinking about your engagement surface, like where do my reps live? You're thinking about the orchestration layer, what are my agents rent, live and run, like where do I build agents and run them? And then where's my intelligence layer? And there's a couple archetypes we start to see that are coming together. We're seeing that you know, most of our clients are either the platform play. So I'm a software company that you know, has salesforce or you know, and I run that end to end in my stack. So I've got conversational insights and agent force and data cloud and maybe marketing cloud and commerce. And I'm a platform I built on this platform works well for single platform shops, regulated industries. But you know, any customization you do on it could create tech debt. You know, you, you're tied to the roadmaps, the best of breed. You know, there are a lot of companies that are saying well what's the best capability for the different things I need to do if I want the best conversational intelligence, you know, for an enterprise is that gong as an example, if I need third party data, maybe I need zoom info. Or if I need the best data management, I might be looking at snowflake or databricks. I might be building custom agents with that and using some in other platforms. So this can work very well. You have to manage license fees, you have to manage integration complexity and think about the workflows to make sure it's seamless. And then lastly an AI first build so you don't have to go far these days and find SaaS. Apocalypse and death of SaaS and everything's going to be custom and you can build everything in five minutes. Articles and look, if you're building in a custom engagement layer, it might be because you have many engagement surfaces that your reps are in. So building one thing on top could be helpful. You're trying to consolidate and coordinate between many different applications you have. You might be building directly on LLM APIs and platforms because you have the engineering capabilities to do that, saving you license costs, giving you flexibility. But you also need to have a sustained engineering commitment and governance. Even if the software engineering is getting easier and easier, managing it and governing it is still task and work to do. And so the reality is that most, most everyone is going to be a hybrid of these. So the question is, where do you anchor, you know, as you go forward. And then finally step five, the ten week clock. You know what I would say here is the way that we've seen this work really well. And what's interesting about this page is that you're spending more of your time thinking about the workflow than you are building the prototype. Because the prototype is not hard to build anymore. You can do it in hours, minutes sometimes. But the research, like understanding the workflow in depth, understanding the humans involved in the experience they have with it in detail, it's not optional and you can't shortcut it. You have to go talk to the people that do it and who are impacted by it. You got to map the buyer journey and the seller processes. And then you need to think about the role of AI, how it will get deployed over time and how that impacts your workflow so that you can build appropriately. If you skip this phase, you might build the wrong thing. It might look beautiful, but it might be the wrong thing. Once you've done that, you can prototype it. And that's not just the front end, which you could do even during weeks one and six, one to six, but it's integrating with real data, testing it against real scenarios, you know, looking at real pipeline, like actually built testing your prototype, then you go pilot with defined success gates, a B test where you can, with your current process, define what it means to be successful with a pilot, that you can scale it. And when you get to your pilot, you start working on the next workflow. And so this is that continuous innovation loop that just allows you to keep running. As you do that, you have to think about what you measure. And so the metric trap that I've seen here is that we often deploy AI and then measure like how much time have we saved per rep? And the thing on the left there is none of these actually tell you whether the AI is working. You know, if someone logs in every single day and they can lose deals. So what we're trying to do is make sure we're measuring. Are we measuring business metrics? You know, because I, I may save some time in my Tuesday morning example, but I might use that to really think about how I approach the customer and be better prepared. It may not be more time with, I spend with the customer, but I may have a much more effective meeting that moves my deal along. So think about the business metrics that we should measure the conversion rate. But in our deals moving forward faster, are buyers responding more engaging, more moving through the pipeline, do we have more revenue per qualified opportunity than we did before? Do we have AI, a B test that we can really prove our hypotheses out. And then, you know, if I think about other challenges you're going to face, the human impact of this is not insignificant. You know, your top performers will say, I don't need AI. I've got relationships. I know how to do this. And if they opt out, your system's not working for everybody. You got to spend time demonstrating to that group how it can help the best do even better. And then the middle performers, you have, you know, really risk out there because they might over rely on it. They might walk into a meeting not having done prep because the agent briefed them and they'll get caught. Customers can tell when there's somebody who hasn't done their own preparation. You got to coach them in how to use the AI. And so there's some intentional change. Management leaders have to model it and also gamify it. Have people share success stories through the organization so you can learn from each other. And then before I get to where to start, what I would say is, you know, agents amplify the data that you have. So this is the last kind of guardrail to talk through whatever's in Your data, it'll get amplified. If you have data that's reasonable, we'll take the right actions, we'll identify the right buying group, draft the right outreach, surface the right relationships. Deals can go forward. Messy data can do the other thing. So, and it can do it very confidently. We can draft an email to the wrong contact, we can populate fields with stale information, we can recommend the wrong next best action. And we can do it at scale. And so damage at scale is something really to think about here. And what I would say is no one will ever have perfect data. It's not needed. You can even create synthetic data to fill gaps these days with AI, but to start with the data that's relevant to the workflow you start with. So that's another reason we start with one workflow. It allows us to identify what data do we need, what quality is it and what risk do we have. So in order to get to that Tuesday morning example in the beginning, what do we got to do Monday morning? And so what I would say here on Monday morning is there's five things, you know, back, back to what we talked about. Map, map out your buyer journey. Understand your biggest friction points. Pick the top two friction points where a deal's getting stuck, where do we have not enough information with buyers? And you'll have to do some research to really know what those are. Pick the end to end workflow you're going to start with. Deal, deal, intelligence and progression is our Tuesday morning story. But pick the one that you know, tackles your friction points. Look at your data foundation within that, you know, what do I need? Accounts and contacts, opportunity stages to be aligned, what do I need to have, what quality is it in? And then make some architectural decisions. You know, where the, where are the reps going to live, where do the agents run, where does the data get stitched? Am I going to buy? Am I going to bill what's already available with the anchor platforms that I already have and I already bought, I need to consum licenses last week, start that, that clock. You know, think of this as, as research first, you know, six, four to seven weeks depending on your organization and you know, two to two to four weeks of prototyping and then get, get out and pilot this, you know, with, with real defined outcomes that you're trying to do. And if, if nothing else, you know, just start with number one, your learnings from there will, will lead you to what, what you need to do next, whether it's with AI or, or otherwise. So with that, maybe I'll I'll just. I'll say thank you. I appreciate the time. There's a. There's a QR code here if you'd like to reach me. I'm happy to share this document. I have a slightly more detailed one I can send. And there's a QR code for our latest State of AI and Enterprise Research report as well.

Outstanding. Thank you, Abe. That is a mind stretcher, but it's also an organizer because it made sense of a lot of things. And the way you presented the contrast just fit totally with how my mind is built. So thank you for that. One of the points that occurred to me as you were getting toward the close was people will say to me, and I'm sure they say to you, oh, yeah, people resist change. People hate change. And my answer to that is, people love change. When's the last time you went to the movie theater because you wanted to see the movie that you'd seen previously over and over and over again? No, you always want a different movie. When you go out for dinner, you may want a familiar dish, but you want a new experience. People love change. Here's what they don't like. Discomfort and confusion, fear that they might make a mistake. And what we've got to do is find a way to make the transition comfortable to them or at least less threatening. And that way, you know, we're not fighting against people fearing change, resisting change, whatever. We're simply looking for a way to get them more naturally accustomed to the new way of thinking and doing work. Good job. Gerhard, what were some of your observations from this?

Well, I think that over the last two decades, we have been preoccupied with three things. How can we make people better? How can we make processes better? And how can we import better technology? And when I look at the big picture, there are compared to 20 years ago, where there were maybe 2,000 different sales technologies. Now we have maybe 15,000. And every year, about 800 to a thousand disappear. So to me, the big piece of the puzzle is what is the right alignment and orchestration between people, process and technology? And what Abe has done is created sort of a roadmap of what is possible. And the things that you said in the. In the beginning, what should we expect? And. And, you know, you talked about the bus. And if the bus doesn't stop, we don't go there. So what we need to imagine based on Abe's presentation is an optimal flow. You know, build the right and orchestrate the right pieces of technology with the right processes and prepare people and I think that you of all people, Jim, you know how reluctant people are to change. So that whole challenge of implementing the ideal infrastructure to have AI in the enterprise with AI at the center, it needs the preparation of people's minds. So it takes a different mindset now. And Abe lives all this, and he has seen before we started, he talked about a company that actually is doing things still with pencil and paper, and then they take pictures and they send those pictures to their home office and they put it in salesforce.com so there is a crazy world out there, and it needs a mindset, a flexible mindset, a positive mindset, and an adaptive mindset to make this all happen. So we need to rethink everything that the world, the way it is right now, it's becoming more fluid. And 20 years ago, there were so many gaps and so many stops and. Which means that our mindset needs to be more adapted to the possibilities that all those technology vendors are offering.

Yeah. One of the things I liked a lot, Abe, was in your early slides where you showed the contrast between the old workflow and the new workflow, and you pointed out how much of that workflow just became complete at the point of the, you know, at the click of a button at the end of the sales call or sales dialogue. Wow. I see. The biggest challenge for most organizations is going to be getting the, the C suite comfortable with the new way of thinking. Because as long as we have previous generations in the C suite, then a lot of the previous generation thinking and habit patterns are going to be superimposed on everything else and they're going to be using AI like new upholstery in a car that's going to the wrong destination. That's right. Yeah. So well said and, and very well explained. Thank you.

So now, Abe, I want to ask you a couple of questions. What are your three favorite AI tools that you're using every day?

You know, I, I actually love some of the tools that we've built on top of anthropics. Claude. One of the things that I really like about the Claude LLM is that is perhaps the most human or it's. It's tuned to try to make you the best human, if you will. So it tends to ask lots of questions and I find that to be one of the best. So when we build our custom agents on the cloud, on the anthropic framework, we get just phenomenal results. And so there's some custom things we built, but that it's on top of it. That's one of my favorites. I'm also really interested in the power of connection between them. So in other words, if you take a ZoomInfo or Sixth Sense or your data provider of choice, we often work with ZoomInfo, but it could be another one. If you take data and you then apply that against your, whether it's anthropic or your salesforce agent force or whatever tool you might use, that's when you get the power of using this intelligence with your other data in a workflow. And I think that's really powerful. And lastly, Notebook LM is still phenomenal for me just as an ability to dump data and to make sense of information. And maybe one thing I'd say about this, and Jim, you said it at the beginning. What does this mean for humanity? And one of the things is, and I said it in the, in, in my talk is that, you know, don't take the thinking away from the rep. But I think it applies to all of us in our use of AI. Like I watched a movie and I was thinking about the, what the message behind the movie was and I asked, you know, chat GPT or something and it gave me an answer and I had more knowledge, but I didn't feel smarter because I didn't think about it. I didn't give it my, I didn't give it any thought. And I thought that was the most let down. You know, I have more information, I feel nothing because of it. And I had to step back and go like, what did I take from it? And I think how we use AI to better us ourselves and humanity is the promised land. And like, you know, either it's going to happen to us or we can be part of it. So I'm excited to be part of it.

Thank you Abe for sharing all your deep insights. You've been very helpful.

That.