Operationalizing Revenue Decisions with AI

Julia Nimchinski:
Always a pleasure. Next session, welcome to the show, Sahil Agarwal. Such a pleasure. Sahil is the co-founder and CEO of Bond, and we’re gonna be showcasing here operationalizing revenue decisions with AI. How are you doing, Sahil? Welcome, Sarah.

Sahil Aggarwal:
Hey, I’m good, how are you?

Julia Nimchinski:
Excited to dive in into this! What’s the latest and greatest?

Sahil Aggarwal:
Before we do that, what did you learn from the previous session?

Julia Nimchinski:
Well, we were really exploring the architecture of Agentic AI as introduced by Swan AI. So there’s three layers, context layer, orchestration layer, and the interface layer.

Sahil Aggarwal:
Okay, let’s… I think the session today will be to condense all of this into one single layer. Okay. Making it… making it super simple, easy, and digestible. I think, so for people who don’t know me, I’m Sahil Agrawal, I’m one of the co-founders and CEO of Vaughan. Before one, like, our company, main product was Rattle, and Rattle, like, more than 200 companies use Rattle, and it’s, it’s the most advanced Salesforce and Slack automation, and then evolved into adding AI, agents on top. But what I learned by adding AI agents on an infra that was legacy is that, you… you can’t really take full power of the AI. You have to build the whole infra from the ground up. But the second thing that I learned was that there are just too many agents, and everyone is building agents, and it would just be hard to be in this ecosystem where we have hundreds of agents running around, not sure what they are doing. And it’s almost like you have a digital workforce, and for every task, you’re spinning up a new agent, and agents have the challenge of becoming the new dashboards. They are there, nobody’s using them. So, what I want to talk about today is how revenue tech will evolve in the age of AI. So, it’s all AI, it’s all agentic AI, it’s all agents. But hopefully, I want to start with a problem, and then work our way into how I think we can solve it, for people who use one. So, Julia, I’m gonna share my screen, and I also have Sarah, who is my VP of Marketing in here, and then Sarah, please feel free to jump in, and I’ll also call upon you at a few places. 2… I do want to thank the people who are tuning in for a 6-hour marathon of listening to 8 people, or 8-hour marathon of listening to 6 or 7 speakers speak. So I do really appreciate the time that people take out of their day to be more forward-looking and understanding from my perspective. Where the world is going. Or from everyone’s perspective. I put out a LinkedIn post, that became viral, like, 8 months ago, where I said, Salesforce fields are dead in 12 months. what everyone was… is still doing is… is having AI update, Salesforce fields. And, what I saw, like, almost a year ago, is that having AI update Salesforce fields will not make them more useful, it actually makes them less useful, because it’s AI updating it. A sales manager does not want to go into the opportunity layout, and read MedPick. A sales manager wants to know, tell me which opportunities I should focus on and what to do. So… I have another prediction now, for where, this world will go in 12 to 24 months. I think I was definitely, like, I’m seeing this with our own customers, where they are not creating Salesforce fields anymore, they’re killing almost all subjective Salesforce fields, and they’re only keeping the objective fields. But I want to make another prediction for what’s happening in one to two years, but before we get there, very quickly, how RevTech will change? Two things. One, the era of point-and-click systems is over. that… the whole era of point-and-click systems was these previous systems were built on databases. And for a database, you have something to go and click and invoke that particular field or memory in the system. I don’t think point-and-click exists in the next few years. The second is the era of point solutions is over. We are seeing all these horizontal solutions come in, and by horizontal solutions, I mean, look at Claude Code. It’s such an amazingly horizontal solution. such an amazing solution that, A, it’s eating into cursor’s growth, which is insane. A year ago, nobody would have thought that, but then B, Cloud Code had to come out with a Claude Coburg for non-technical people. And then the whole CloudBot was built on the idea of Cloud Go Work, which is giving it even more autonomy. So, point-and-click is over, our engineers don’t point-and-click anymore, they all just talk to Cloud Code and get their work done, and then point solutions is over, because AI is just such a horizontal technology, where I think what will eventually happen is there will be an agent for every function, and that’s it. No more hundreds of agents running around, no more building agents. Julia, go build me this agent. Julia, go do this for marketing. Sarah, go build this for webinars. No, like, it’s just one agent. And the way that I think about this is… The previous era, we bought all these tools. In fact, we bought too many tools. But what is the challenge with sales managers? They are still trying to understand what is happening, and what do I do about it? Are my people sending the right emails in Gong, Engage, or Outreach? And if yes, then are people replying to it, and why aren’t people replying to it? I’m just trying to understand what is Sahil doing well on calls, and what is he not doing on calls. Do I go into Gong and listen to a bunch of his calls, and then do I go into my sales enablement playbook to see what he was supposed to do? Which is… which lives in high… part of Seismic. It’s just too many tools, and it’s not like we’re closer to the answers. So, from a problem perspective, what ends up happening? I, as a sales leader, I’m just waiting for information. Either I have to go into all these tools and scritch this information, or I have to wait on backlog for weeks. As a sales leader, as a sales manager, as an AE. As an AE, if I just want my territories prioritized and have a particular collateral for each segment, I’m waiting months on my content team, data team, and rev ops team to get together to give that to me. I want some analysis done, I’m going into a board meeting, and I’m hoping to God the information in Salesforce is correct, or I’m just pinging my revoz person last minute, asking them for help. And many times, I don’t even ask questions, because I know it will take 3 weeks, and it’s not worth the effort of what it will, what will go. And… forecasting. You can’t even get me started on forecasting, because hundreds of hours are spent every week A talks to the manager, manager talks to the VP, VP talks to the CRO, RevOps stays with everyone, and that’s how forecasting meeting happens, on a week-to-week basis. Hey, Jared, looks like you’re the next speaker. But on the other side, it’s not that the situation is much better This is RevOps. RevOps is not sitting on those items, because they like to sit on things. In fact, RevOps is probably working 12-hour days, and RevOps is thinking, I need more headcount. I have too many tickets to do. RevOps spends more than 50% time, being reactive, There is not a single function in the company, HR, finance, sales, marketing, that is this much reactive. And then reverse will go to CFO asking for headcount, and CFO will say. Maybe next quarter. Maybe next year. Because the sad reality is that RevOps is a cost center. Now, we can do… we can say whatever we want. They are not building the product, they are not selling the product. So RevOps ends up in this place where they’re supporting all the revenue. If you have seen that meme on the internet, like, they are the tiny piece that holds the whole thing together. But it’s not like they have another piece coming anytime soon. And on top of this, now we have this agentic AI wave, where we get to buy even more point solutions. So fun! I get to buy a point solution for my sales team, then I, as a CEO, have to approve a point solution for my sales engineering team, for my rev ops team, for my SDR team, for my enablement team, product marketing, BI, go-to-market strategy, Salesforce admins. I can go on and on, because even revenue function now has exploded. Revenue function has 10 individual functions within it. At a decent sized size company. I think this is… this… this is buying all these point… agentic point solutions, and then… Clicking through and building agents in all of the systems is not the way to go. So the way that we built one is we built one as a headcount on the team. It’s not software. It’s… it’s literally a headcount. We built one as an agent specifically for every single revenue function. it’s not customer success versus sales versus SDR versus enablement versus product marketing versus marketing. It’s just one co-worker. It’s like, I have a RevOps person sitting next to me 24-7. I don’t have to think about, hey, go build this agent. I just tell my reverse person, hey, this is what I want to do, and it, goes and does this for me. And the way we built it, we spent, last 12 months on building, their technology. I’ll talk about, in a few minutes, on how this is… it’s not just built on top of… it’s not a thin wrapper on top of LLM, this is the most advanced agent, on the planet. And it can do every single task that revenue team deals with on a day-to-day basis. Every single task. Anything that you’re working on today, this agent or one, is willing to do it. Before I talk about the examples of tasks, I just want to say that the reason why it’s able to do it is it is the only agent that can reason through both structured and unstructured data at scale. So, there are agents that can do it on structured data, which is your CRM data warehouses or contact databases. But then there is 10 times more unstructured data, which is what’s happening in calls. Like, even this webinar that we are doing right now, it’s unstructured data. There is nothing… there is no structure to it. It can’t be fed into a database. It’s all just a stream of text and video. And Vaughn is able to do this at scale. The engineering part was, how do you handle so much unstructured data? Because today’s AI context limit is only a million tokens, and 20 calls will fill up a million tokens. You know, 20 calls are how many deals? 3 deals. And a mid-sized company has a thousand plus deals in the pipeline. And one can handle a thousand plus deals. So, some of the things that we wanted it to do, again, it’s every single task. It’s ADOP reporting, it’s Salesforce admin work, it’s territories and commissions, it’s pipeline analysis, it’s forecasting. meeting prep and QBRs, and model building. And you don’t have to build a unique agent for everything. You just talk to this one system. It’s… it’s like Claude Code. For revenue teams. So… the 7 categories of requests, that one handles, and I think this is important from an architecture standpoint, which is. Like, when we were building this, we didn’t want it to be just for… glean for sales, which is, just tell me, like, hey, I have a privacy objection, can you help me answer that? That’s a search on unstructured data. Or we didn’t want to build it like a cortex for Snowflake, which is, like, analytics on structured data, which is Category 1. what one does is it’s able to do those two things, but then the complexity keeps increasing. It’s able to then go through individual deals. Hey, tell me what’s happening with Acme Opportunity. You can do that with Gong, by the way, as well. Like, Gong has, Ask Anything. But then, multiple deals. Tell me which deals don’t have a next step. Tell me which deals don’t have a decision maker. And you have a thousand deals. Vaughn is able to reason through that. You don’t have to build an agent that runs every night, to get the answer. Like, you just go and ask that. then Vaughan can do analysis, because if I have a coworker standing or sitting right next to me, I want him to do analysis. Hey, tell me, do we actually win deals when we don’t have a decision maker? Because some companies do, some companies don’t. Like, by what stage do I need a decision maker? Like, hey, what happens when my next step is not there? And why am I losing deals for this product in EMEA? And then… Every single person that we speak to, they say, like, hey, analysis is great, but help me take action. like, the same system can take action across every other system. You find an insight, you take that insight, and you put this into an action. I’m launching a new AI product, tell me every single person who has asked for AI features in the last 12 months. And write me a personalized outreach for each one of them, and put them into a sequence. And finally, data science, it can build chiven prediction models. So, these are some of the examples. I want to show you the product, briefly, so you understand, like, how it works, but the… Idea is very simple. It looks exactly like a chat GPT, or a cloud code. It’s… it’s… you don’t have to learn complex agent builder systems, you can just ask it what you want it to do, and it will do it for you. Now, let me show you, the system in action. So let’s say if I ask it a question, I think I put in the name for Amos. Let me copy the bush in here. Okay, so I am a CRO, I’m going to my forecast meeting, I forgot to prep, because I had people over for dinner last night, and this meeting is at 8am, and I have 50 sellers joining this meeting. And I want to look informed, and having done my homework, which I usually always do. So I go into one, and I simply ask it, I have my forecast meeting coming up for new biz deals. Can you help me prep? I usually call out deals that are more than midway in my funnel, and, like, call out, and we talk about the significant risks in those deals, and give me deals that are dead, so I can call those sellers to move them to close lost, for the love of God. And then, I asked it to mask name, context, and number, because it’s connected to a live database. Now, Vaughn is not software, it’s not point-and-click, it’s a human or AI human sitting next to me. This is what I would have told an AI human, and I don’t care how it gets the answer. It should just go into structure data, unstructure it, and get me the answer. So see how one reasons through, so much information, thousands of calls, hundreds of deals? To get this answered. Now, it has to spin up 50 agents, 100 agents, like, it’s making all those queries. I… I don’t have to think about it. That’s all, one. And while it runs, I’ll show you the answer, in a similar question that I ran right before this. So… It usually takes, for a question like this, less than 3 minutes for it to answer, which is, like, imagine spending 30 minutes, 60 minutes, painstakingly going through all of this information on a Sunday evening, but getting this so quickly. And then, I can see, like, it told me, like, there are 6 deals that I should be worried about, told me everything about CRM hygiene alert, what’s the status, what are the key risks, and all of those deals. And then it told me that, hey, there is one deal that is likely dead. So you know what I will… I can do? I can just simply go in and say. Can you move that dead deal to close lost? Keep masking the name. And Vaughn now has a human in loop that lets me manipulate other systems. And, it would automatically load up the scale. In this case, it’s loading up the CRM write scale. One will spin up agents, and will let me do that. And, like, just like that, that’s it. I can click on approve, and it will move the opportunity to close lost. That’s it. I can update Salesforce. You know what I’ve seen people do with it? And I will also show you an example. So, I went into one before this, and I asked it which accounts are assigned to me in the last 3 months that haven’t been touched. It ran for 3 minutes 35 seconds. Came back with, these 600 accounts out of the 1,100 accounts created under my ownership in the last 3 months. And gave me the top ones. And, I then asked it, hey, can you then assign this 600 to Jesse? I think we just went from a sales CRO use case to a RevOps use case. Same interface. No different agents, no different UI, it’s just a co-worker sitting next to me who is extremely smart, and can reason through information. Let’s look at a use case for Salesforce admins. the same Salesforce admin. will… this Salesforce admin will go into the same interface and say, hey, can you tell me the purpose of this flow in Salesforce? and then it comes back a minute later, with the exact purpose and the flow, and you know what? I can just sit here, and I can just tell it, create this new flow, create a validation rule, like. Again, one general purpose tool for every single function in the revenue world. And it’s not software, it’s a headcount sitting right beside you. So I think… I very clearly, hope that in 24 months, we don’t have an army of agents who are running around and doing things that are unsustainable. Agents should not become the new dashboard. We have this amazing bazooka in our hands, which is AI, which are these LLMs. In fact, LLMs can do more things that we know what to do with them. They’re definitely smarter than you, me, and everyone else here. And what we are asking them to do, when a lead comes in, can you route it to the right person and write a follow-up email? I think one is that agent which is… which will feel like super intelligence on your revenue team, and it is trained on revenue data to be able to do that. Now, how we build one, I think, everyone thinks that What you… what you do is you take an LLM, you take a chatbot like Claude or ChatGPT, you take an MCP, and you connect it to Salesforce, and boom, done. And that’s how something like One works. But in reality, like, just for the Category 1 use case, you need to build a semantics layer. Because if I’m asking the system, like, hey, tell me what is my win rate, for last quarter, it needs to know Is your win rate from Stage 2 onwards, or is this stage 1 onwards? Are we looking at only new business deals, or new business plus strategic deals for a win rate? What is last quarter? Is last quarter February through April, or is last quarter Jan through March? So one has a semantics layer. I won’t bore you with all the other categories, but essentially. Like, eventually, the system, to be able to get to Category 7, ends up being a system which has its own execution environment, like its own sandbox, its own machine learning code. It has pre-processed information on every single deal and account. It has a semantic layer, and then, obviously, it has a vector database, and regrettable mechanism. And if you see, this is… these are where the AI agents are. Like, that’s… that’s the role of Claude and ChatGPT in this… in this infra for one. We have a few minutes. I want to talk about the ROI very quickly, because I think we can make all the investments that we want in one or any other agentic AI solution, but what’s the ROI? How do I see this happening? I am… very confident that the ROI will show up in seller quotas. You know, seller quotas haven’t moved in 15 years. It’s 2% compounded annually. So… And, I thought maybe it’s because sellers don’t have enough time in their day. So we looked at more than 1,200 sellers, and, you know, what we found? On average, external meetings take up 1.2 hours per day. then why aren’t we giving these sellers more pipeline to go close? Why do we keep hiring 10 more sellers, and distribute pipeline and let everyone fail? We should be able to give them more pipeline, because they clearly have time in their day. We don’t, because they can’t hold that context. So, AI can hold that context. One can hold that context. One can sit right beside the seller. So a seller never has to log into any other software ever again. not ZoomInfo, not Apollo, not Gong, not Outreach, not Salesforce, definitely not Salesforce, not Seismic, like, it’s all available in one single system. So… I teased at the beginning, like, of what will happen. I think in… by the end of this year, we’ll start seeing $2 million quotas, and that is where the impact of AI will show up, where quotas will go up, and it won’t just go up by 10 or 20%, they will literally double. I was… I went to Perplexity’s office two days ago, listening to Erwin speak, And, You know, Perplexity has 5 sellers? And 9-figure in enterprise revenue? enterprise, I’m not saying people who are paying with their credit cards, I’m saying 5 sellers have handled 9-figure of enterprise revenue. So, I told him about this prediction, I said, like, your prediction is not bold enough. So I think that’s where ROI will show up, so… That’s what Juan is, hopefully this was helpful.

Julia Nimchinski:
Incredible session, thank you so much, Sahil. So many questions, but one of them… I love how you reinvented all revenue workflows, and it’s essentially an all-encompassing platform for AI native, GTM, revenue operations, you name it. I’m curious, how do you see the evolution of a sales team, a GTM team, in terms of function, since one becomes this one super agent, and it can be in RevOps, and, you know, a sales manager, and then an AE, how do you see, like, I mean, perplexity as 5 sellers or, you know, whatever the structure is, but how does it look in, in… In terms of structure in your team currently. And how do you envision it for others?

Sahil Aggarwal:
I think, the first… The unfortunate reality is that, what will… what happened to engineering will happen to sales. Junior engineers are finding it very hard to find jobs in this market, because I would rather hire a senior engineer and give them clawed code and let them run with it. My engineers are doing stuff in a month that used to take a year. And I’m not exaggerating. If your team is not seeing that level of productivity improvement, there is something missing. So I think for salespeople, the same thing, I think, will happen. If I can give a $5 million quota to someone that I know is super capable and will make the most use of that pipeline, then that is who I’ll give that, pipeline to, and not spread it over 5 people or 10 people. In perplexity’s case, like, every seller is closing more than 8 figures, or 8 figures every year. So, I think that’s how sales evolves. So, based on that, like, your… your supporting structure will also change. Like, it’s 15 sellers per RevOps, but now if 50 sellers reduce to 3, or 10 sellers, then I think that ratio will also change, because that same RevOS person can do a lot more.

Julia Nimchinski:
So in terms of role itself, the seller becomes an orchestrator while they’re selling, or how do you see it? So it’s a technical and a customer-facing role, essentially.

Sahil Aggarwal:
And that is what sales was supposed to be, not, learning High Spot, and Seismic, and Gong, and Gainsight, and all these platforms. Like, it was just meant to be, like, this one, seller just knows how to sell, and we will get their meetings per day to 4 hours, and not 1.2.

Julia Nimchinski:
Awesome. Sarah, what are we missing? Where should our people go? How can we test ride it?

Sara Kinsey:
Yeah, thanks for asking that. So, please feel free to follow Sahil on LinkedIn. He’s sharing all kinds of stuff there. Julie, I’m not sure if you’re sharing that link afterwards, or Sahil, maybe in the chat, if there’s that possibility, you can put it there. But you can also go to our website at vonlabs.ai, and also watch, we’re adding a lot to our website as we go, and there’ll be some big changes coming to the website shortly as well. as well.

Julia Nimchinski:
Awesome. Thank you so much again.