Agentic AI is emerging from the curiosity stage and transforming core deal origination processes. Similar to fictionalized robots, “Agentic AI is focused on learning, thinking, and reasoning all towards the outcome of the goal we’ve given it,” shared Tom Lesnick (Co-founder, Copilot/VASS). Joined by Chris Brown (President, VASS/Intelygenz) in this Inside the Source conversation, the two delved into where GenAI is today, the key dealmaking functions that agentic AI is impacting, and how teams can effectively take advantage of these use cases.
Going beyond basic automation and back-and-forth conversations with chatbots, Tom detailed key tasks and workflows that are being enhanced and accelerated by agentic AI. From company identification to data collection, he shared the next level of utilizing GenAI — to research, leverage signal data, customize and personalize emails, monitor portfolio companies, and more — across the industry. But there are still benefits to uncover, Chris revealed. While data analytics has been a part of the deal origination process, the information has been limited to human processing and reasoning. “The two superpowers [agentic] AI is bringing to the table are speeding up the interrogation of data and improving the accuracy of understanding what the data is telling us,” Chris stated.
Your data doesn’t need to be perfect to start, but what you give the agents will determine what you get. Chris went on to say, “The quality, volume, and correlation between your data are critical to the accuracy and quality of the answers you get out of [the AI agents].” This is where the input: available data, has to align with the AI agents’ output: the firm’s strategy. Tom emphasized, “Figure out the workflows you want to focus on and the ROI you’re trying to get out of the project. Then, figure out if you have the data to support it or how much work it needs to get there.” This is one of the several areas where the human-in-the-loop component can’t be overlooked.
Watch the full video for a complete rundown on AI tools, AI agents, data, and how dealmakers can use them for more efficient and powerful sourcing machines.
Transcription:
Sourcescrub: Tell us about yourselves.
Tom Lesnick (TL): Hey, everyone. I'm Tom Lesnick, senior director at VASS. I run our private equity practice here. I've been in the private equity space for about 15 years now, focusing on all things data, AI, and CRM, both in-house at private equity firms and at various consulting firms.
Chris Brown (CB): Hi everyone. My name is Chris Brown. I'm the head of the AI and Data practice at VASS.
Historically, we joined the VASS group through acquisition, through an organization that was specialized in the deployment of artificial intelligence and automation solutions in the enterprise.
Sourcescrub: How does GenAI relate to private market dealmaking?
TL: I like to liken it to a robot. Everyone you know has seen movies over the years understands what a robot can do. I grew up watching The Jetsons. So Rosie the robot is like the robot maid. But instead of cleaning the house, the AI these days, the agenetic AI these days is focused on learning, thinking, reasoning, all towards the outcome of whatever the goal is that we've given it.
And there's a lot of really interesting things you can do in private equity. For example, can do a lot of research from a large number of data sources and then generate some sort of output. Could be an investment memo or an opinion on all of the due diligence documents that have been fed to it. Think of it like a robot, but the robot is more autonomous and less on the rails, and can do more reasoning and generation than in previous iterations of AI that have come in the last couple of decades.
Sourcescrub: What are the most exciting use cases for AI agents in dealmaking?
TL: The tools are getting better and better at proactive company identification, especially within specific niches. Specific pieces of your firm. So market mapping and proactive company identification is getting better and better and more interesting. Once a company has been identified, the signal generation and prioritization aspect of some of these tools is very fascinating. AI is taking this, with a really interesting step forward where it's leveraging all of the research from that, from market mapping and from the signal generation.
And it's also now starting to leverage first-party information, like what are the types of successful emails that get sent, you know, when you're running an outbound campaign, and it's taking all of that information and making hyperpersonalized emails to the CEO of a company. And what's really cool about that is they can do it now leveraging agentic AI and generative AI in the exact voice of the analyst who's writing the email. So it doesn't sound like a canned one-off email at all.
We've also seen further down the line when you get into diligence, as you know, when you're doing due diligence in private equity, you're getting hundreds, if not thousands of documents of all different shapes and flavors, PDFs, Excel files, PowerPoints. And that takes a long time to comb through that. But with enough contacts, with enough good data, with enough automation and cleansing of that data from the data room, as well as other previous examples and investment memos and other pieces of information from the firm's history, AI is getting better and better at outputting a point of view for a specific company when you have all that information about it.
So that's exciting for me, and it also can help compare the information you're getting from the company with the overall market. So like, not only what are the KPIs for that company, but what are the benchmarks for those similar types of companies that are out there. And it can go out and do that research. So for me, those investment team use cases are interesting.
We've also seen on the investor relations side, there's always a lot of paperwork. And like DDQ and RFP responses to be done. So AI is pretty good at leveraging previous responses and what it knows about the firm to fill those out and make that process simpler and easier.
And lastly, on the value creation front, there's been a lot of ask and a lot of progress as well on how do we monitor the portfolio better? Public information that the portfolio companies are putting out or that are on social media? It's across the KPIs that are being delivered to the PE firm from the portfolio company. It's scanning of emails and other types of communications, or lack thereof, with the portfolio company to kind of understand, especially for firms that have hundreds or more portfolio companies, what really needs attention from that value creation team now.
And then, of course, like gathering all that data too, like it's a pain to gather KPIs from your portfolio companies and we've seen in terms of data gathering and ETL and cleansing, and normalization of that data, big strides there as well. So those are the things I think are actually happening today that are relevant, that are, that are impactful.
Sourcescrub: What are some use cases that other industries are using that private equity should pay attention to?
CB: So, if you take what Tom said and we broaden that out, really, what we're talking about is the ability to do two things. One is to speed up the interrogation of data, and the other thing is to improve our accuracy on understanding what that data is telling us that they’re the two superpowers that AI is bringing to the table.
So it's not always just about speed. It's also about what insights am I driving from my data that that I couldn't drive before? Because we've had data analytics for decades, right? I say for centuries. And what's happening in data analytics is as a human, you're writing the rule. So you're looking at the data and you're saying, hey, and let's stick with PE just for now, I'm looking for particular companies that I think would be great to invest in. And I'm looking at all of the data. I'm looking at my historic data. And I'm and I'm starting to build a pattern in my head and from my data analytics that I'm, that I'm driving rules that say I like I like to look at companies in this industry that have this profile, that have this kind of trajectory. I'm seeing all these different patterns, and I'm writing and I'm writing some basic and often not basic, some intricate rules that help me to apply that against my data, that I've already got to try and understand, what is it telling me?
What AI is allowing us to do is not only to do that with speed, which it does right? It gives you the speed, the speed element, which can be what can be huge benefit in its own right and lower effort. But it's also allowing you to now write the rules. It's also allowing you to say in history, these have been my successful investments, and I may or may not choose to tell you why I thought they were great investments.
That's a choice for going through the data science process. But maybe we don't say anything. Say, can you find me organizations that will fit this, but I'm now giving you the rules. So now we're automating the rule set creation. And we see that with hey, can you give me a confidence level?
And what I always say to people is that the technology can be specialized. It is specialized. Let's be honest. And it can be sometimes a little daunting, a little complex, but the majority of the solutions that we deploy for our clients are poetically simple to understand what the value is, and that can be anything from can you classify and understand from my data good, good investment strategies and good targets that I want to invest in from a PE. All the way through to, and we do work with many industries, so if this is not valid to the audience, I apologize. But all the way through to can you detect finite minuscule cracks in silicon that will electrically test correctly, but will fail in the field, to stop that failure happening in the field? Or can you tell me that there's a tarpaulin on the back of a truck that is correctly fitted before that truck is allowed to move, right, because that is a classification through image?
So, it's really we always say: if you have a lot of data within your organization and you believe that that data holds the answers to valuable questions, then you really now have a duty to apply artificial intelligence to that problem, right? And irrespective of what that data can tell you, because it can be as the random examples I just gave you there, it can be across the spectrum.
But the key is, is there a correlation within this data, and if we believe there is, then try and extract it. And artificial intelligence is absolutely the methodology to be doing that.
Sourcescrub: How can teams get their data ready for AI?
CB: So it's absolutely true the the quality of your data, the volume of your data, the correlation within that data are all going to be critical to answering the questions of what accuracy or what what level of quality can I get out in terms of answers, whether that's a prediction or classification, or from generating email or whatever it happens to be?
So I think having we all start at the end of what's the goal? And then we look back at, hey, have we got the data in place that will allow that to happen? So, if you want to do a daily prediction on something and your data is aggregated on a weekly basis, AI is amazing. But it's not a magician, right? It's going to make that challenge very, very difficult, if not almost impossible, to understand. So data storage, data aggregation, data cleanliness, having an architecture that desilos your data, and it makes it available to be used within the input source of answering questions for your business. Obviously, the first thing you need to do is to ensure that it's available, that it's that it can be cross fertilized from different silos.
There's no point having data that's valuable to this challenge that is stored in places where it cannot be. It cannot be cross-fertilized. And then you're you're already you're in a stop point, or you're certainly hindering your capability to achieve the best possible results. So, making sure it's clean, disaggregated to the level that you need it to be.
But, but overall available right to be used as a whole entity is super critical to the process. But I would also say don't start your journey with trying to get the cleanest data you can possibly get. What you want to get is the cleanest data or the data cleanliness and availability you need to solve the problem, to start with, right?
Otherwise, you go on this massive data journey trying to get this perfect data set and perfect cleanliness, and you will never start. And you're leaving a huge amount of value on the table by getting your data perfect, and you've missed the market, or things have changed. So, start with what's the challenge? What's the problem? What's the what's the issue? What's the opportunity? And then look back at, do I need to do what kind of data pre-processing do we do? Often, we do a lot of data synthesization. Maybe we've got to do that. And there is no real-world data. There's not enough real-world data. And we've got to synthesize.
Or it might be we want to create feature stores and do a lot of data pre-processing in order to make the model work easily. But start with start with the challenge opportunity first, then look back to see, have I got the right data structures in place to solve that problem? And if you've got a multitude of challenges, problems, and opportunities. Of course, have the biggest view possible. So you can start to build an architecture of data that can match your ambition.
TL: I think for private equity firms, the different categories of sort of raw data that exist that are, that are important, and almost every firm has like CRM data, whether it exists in the CRM or it's in a data warehouse. Some of that unstructured data, you might have a SharePoint folder for each company that you've looked at that has PDFs and board decks and call notes, and customer calls, and that kind of stuff.
Then you kind of have other first-party data that kind of can help train on user behavior and preferences, like emails, calendar, Slack messages, that kind of stuff. That's all really useful. Yeah, sometimes the data is not clean as Chris said, it's a process to clean and normalize the data, and then other times, with the unstructured data, in order to help make it more useful for AI, there's some work to be done. So, feature tagging of that data, like making sure that it's clear what type of files are there, what type of data is really being shown in some of these PDFs, and there's some work to be done there. But the first step is, do you have those core data sets, which are your first-party data, your CRM data, your file data, plus the third-party data that can help you make decisions and help again with top of funnel stuff.
So, do you have availability to that data? Is it in a place where the models can get access to it? That's the first step. Of course, there’s always cleansing, tagging, fine-tuning, and generating knowledge graphs from that data. And I would agree with Chris, figure out what are the workflows that you want to focus on? What's the ROI you're trying to get out of the project, and then figure out is the data supported or how much work does the data need on it to get there?
CB: You might not have the data, so there might be an element of data sourcing and data collection that you need to incorporate into your solution. So you might be looking to solve challenges, opportunities that require other people's data or data from the market or data that's in public, or so you might have a little bit of that work to do as well, just to round off that conversation.
TL: Some of the new tools in the market are good at researching, gathering some of that data. In addition to sort of more traditional third-party data sources.
Sourcescrub: What are the first steps firms should take to incorporate AI agents into their deal flow process?
TL: What are the 1 or 2 workflows that are happening now? They take time. They're complex. They happen relatively frequently, and the data exists or is or is accessible to purchase or gather for that particular task. What's the 1 or 2 use cases you want to focus on? Make sure you have access to that data and tools, and vendors like us can help figure out is that data in a good spot and does it need work.
You also have to choose the right tool or the right vendor as the next step. So there are, you know, the big platforms like the OpenAIs and the Anthropics out there. There are purpose-built tools that are tuned for very specific use cases within private equity and elsewhere. There are implementation firms and consulting firms like VASS, who can build you your own custom agent or custom application, so you have to figure out when is it appropriate to buy a tool, when it is appropriate to use a generic platform, and when is it appropriate to create something yourself.
But ultimately, once you make that decision, my suggestion is that you run a pilot. So figure out what that one use case is. Help understand what are the metrics you're really trying to change and measure, and make sure those are figured out upfront. And this is like any, you know, development or software project. You should figure out what are the success criteria, how can we measure, and then try and time-box it.
Projects like this can go on for a long time, but at least try and prove some ROI in a relatively short amount of time. You're talking about like weeks, not years here.
CB: AI is technology like technology. And I think we've got used to start in technology projects correctly in locking it to return on investment before we get excited.
The only difference with AI, or the core differences with AI, is one, it's new-ish, so everyone's super excited and they're desperate to go down the rabbit hole of talking about models and this model and that model and LLM and generative versus those machine learning like. So we get super excited by the technology, but it is just technology, so if we're not careful, we don't lock to ROI. You end up with creating phenomenal mathematical and science technical solutions that don't solve the problem that you set out to solve, right?
TL: Yeah.
CB: And so locking to return to investment, understanding what good looks like, what are the KPIs, how does that drive the target variables of the AI? And making sure that's super clear to the point of we often build monitoring for those KPIs before we build anything else. So we can see how we're tracking.
Sourcescrub: What’s your advice to people early in their careers who are worried about being replaced by AI?
CB: Your job is changing to use the tools that are existing today. This is tooling. So think of it as tooling. And these are tools that bring efficiency to your world. It isn't going to take your job. The amount of projects we've started where the idea is, hey, if I can automate this or I can apply artificial intelligence plus automation, I could I could reduce my costs over here with headcount.
We've started a lot of projects under that conversation that never happens because when you look at, hey, hold on, if I can keep these people and the productivity goes up by this amount, there's all of this market over here, and the growth, and this is not hearsay, it's not from the AI company talking a good cliche, right?
This is genuine where the growth opportunity becomes orders of magnitude bigger than the cost saving. Everybody switches in the project of what their goal is, from saving a little bit of cost over here to changing or addressing an opportunity in a market over here, it becomes because the challenge is not reduced cost, the challenge is increasing output or productivity per person.
TL: I believe that certain functions, perhaps things like certain roles within finance or certain entry-level jobs, or certain folks who do repeatable processes, could be recruiting. Like those industries or those functions will be more disrupted. I think it'll be more you'll have AI teammates with a smaller number of humans leading those functions. So I think there'll be some transition.
But again, as Chris said, there'll be more opportunities for roles that are around managing, leveraging, building on top of optimizing AI. So I think I think it's going to go both ways. And in Chris's advice, I think it is spot on. Figure out how to work with the tools, understand the tools, think about which are the hottest tools that are coming out that are going to help you do your job better, so we all have to figure out how to leverage it for the things that it's good at.
Sourcescrub: How should senior leadership approach familiarizing themselves with AI tools?
TL: You want to be experimenting, like monthly, quarterly, with different ideas. Not everything is going to stick. But I think there will be things that show ROI that'll be exciting and that you'll want to continue to roll out more broadly. So if you haven't started already, dive in. Obviously, build a network of other senior leaders who are thinking about AI, bounce ideas off of each other, and figure out what's working.
Make sure you're staying aware of what's going on in AI for your industry. Also, we have to start thinking about our roles a little bit differently, meaning think about our roles as really being helped by and aided by AI. Whereas I know private equity traditionally has been a very relationship-based type of business, I think it will still continue to be that, but aided and guided by AI. So we need to implement that in our daily lives as well, not just for the firm.
CB: If you're leading an organization or you're leading strategy, you're making significant bets pretty much day-to-day. So AI is a companion. It's a tool. Don't go asleep at the wheel. Understand the limitations. Understand the guardrails. Understand. Ask the questions of what might go wrong with this, and so on.
Really understand the environment of that tool in order that you're still putting consciousness into the decisions that you're making and using AI as a companion and not just ticking boxes and saying, do this and that and this and that. So I think that's a really dangerous place to go, alright? There is human in the loop here, and that isn't just a human checking a box, taking a copy paste and put it over here. That is about taking it as an input, right?
And we talked about agents today, and we didn't talk. Not today, but in general in the world today. They are companions, alright? They absolutely are companions. But then they're not replacing your own consciousness to go and make your own decisions, especially when you're making bigger bets in the world that you just talked about. Use the tools responsibly.
