“Our data science team is bigger than our investment team.”
Many VC firms claim they want to be “data-driven” or “AI-native.” But when Melody Koh at NextView shifted from full-time investing to spending 75-80% of her time on data and AI initiatives, she made a prediction: most firms trying to follow suit will give up within twelve months.
We sat down with Gopi Sundaramurthy (Managing Partner at Ensemble VC), Melody Koh (Partner and Chief Product Officer at NextView Ventures), and Jason Miller (CEO of Foresight Data) to unpack what “doing AI” at a VC firm actually looks like.
TL;DR: Everyone’s talking about being “data-driven” and “AI-native,” but most firms won’t follow through. Success requires top-down buy-in, a senior champion who understands both investing and product, and a willingness to fundamentally change how your team operates. The firms that get this right will compound their advantage over time.
This virtual event was presented by Foresight
Introducing Foresight AI, the first platform that lets you query all of your pre- and post-investment data on a desktop, on mobile, and in the enterprise AI service of your choice. If you’d rather just type or say, “what’s the IRR of fund III” than build an Excel model, Foresight is for you.
It Won’t Work Unless…
Jason made the point clearly: You need top-down buy-in for this to work. If AI is just someone’s pet project off in the corner, you won’t get the adoption you’re looking for.
He pointed to Melody’s approach at NextView as a model: a partner-level champion driving adoption across the organization. He’s seen the alternative play out too many times.
“They go hire some data scientists with, like, three years out of their PhD program. They’re brilliant, but the adoption across the organization is where they stumble without more buy-in from the organization.”
At Ensemble, Gopi has taken it even further. Their data science team is now bigger than their investment team. As he put it, “more than half our investment dollars actually go into managing the data function,” because they’re making a bet:
“The bet that we’re making is that this is going to be something that’s gonna be compounding over the years.”
The Behavioral Change Nobody Talks About
Gopi shared a framework that cuts through the noise around “data-driven VC.” He hears this question constantly: How do I get into data-driven VC?
His answer: It’s not a zero or one thing. It’s a journey with distinct levels.
“A lot of VC firms don’t have workflows. And when you don’t have workflows, no matter how much data you have, it’s not able to provide meaningful augmentation inside.”
The real shift happens when it changes daily behavior. When an investor wakes up in the morning, are they going to traditional sources, or are they starting with how the data platform can help them?
“That is a fundamental shift in terms of how people behave and how people are working within a given firm. And that’s not gonna happen on day one.”
This changes everything else too: how you hire, what skills you look for, even how associates spend their time. His team runs on a sprint cycle with monthly model releases. Team members spend less time discovering companies and more time actually networking and talking to them.
(Check out Gopi’s manifesto on the structural implications of data and AI in investment firms)
Kill Products Without Senior Ownership
Gopi offered a brutally honest take on why internal AI initiatives fail:
“I am a true believer that if a product doesn’t have an owner, just kill it. Don’t let it run for long, especially if it doesn’t have senior-level ownership.”
Even more importantly, the people building data products need to be on investment calls. The VC team is their only customer. If the data team isn’t hearing what’s happening in deal discussions, they’re not getting feedback.
“I’ve seen a lot of data teams trying to do this on the side where they would have them, hey, send me the list, and I’ll go to the partner meeting with that particular list. No. You should be on the calls when you’re talking to the founders.”
Vibe Coding Only Gets You So Far
Jason shared a story from a customer that captures where many firms get stuck: “I was vibe coding on a plane flight from Seattle to New York, and I came up with an agent.” Great start. But making it actually work in production? “That’s where you actually need some serious engineering work.”
“I think people sometimes are thinking there’s this holy grail of this one big AI thing, like EQT’s Motherbrain or something. It has a great term and a great name to it, but you’re gonna have, like, twenty different little AI things going on. It’s not gonna be one monolithic AI that does everything.”
Melody reinforced this distinction between data and AI. People jumble them together, but they’re different. And she shared a practical framework for thinking about ROI:
Time leverage (Type 1): “I spend a lot of time doing X, and X is extremely manual and annoying. Wouldn’t it be nice to automate the manual and non-creative part of X?” Because if you can shave off two hours a week per partner, “the two hours can be four new pitches.”
Time leverage (Type 2): This would take too much time, so we’d never do it manually. As Melody put it: “This pile of five hundred I otherwise would not be able to look at because it would be very costly time-wise.”
Better decision-making: Are we going to make better decisions by leveraging AI and data? Melody pointed to red-teaming as an example here. Jason added that he’s seen firms set up different investor personas and have the AI debate investment memos from those perspectives, revealing biases in the process.
The Real Trade-offs
Melody offered the most candid take on what it actually costs to have a partner lead AI initiatives.
After seven years as a full-time investing partner at NextView, she cut her pitch volume from five to ten per week down to three. But the upside surprised her:
“I’ll tell you six months ago, like, I’m a pretty technically aware person, like, you know, with a product background, and I hear these pitches. I have no intuition in terms of how these AI products actually work and are built. And then, now having spent a lot more time disproportionately over the past couple months with these models directly and indirectly, I feel like my intuition and feel is much better.”
The partner who leads this work becomes a sharper investor in AI companies, not a worse one.
But Melody’s also clear-eyed about why most firms won’t pull this off:
“For someone to do this type of thing well inside a firm, the person needs to have reasonable level of influence and seniority and trust. The person needs to actually understand investing in that particular part of the asset class and most ideally understand how that firm invests. And then the person has to, like, have some kind of ability to wrangle and prioritize and prototype and ship. So it’s a pretty hard role to hire slash find internally.”
Venture5’s Take
So your firm wants to “do AI.” Now what?
Start by being honest about whether you have the organizational willingness to actually change how people work every day. Not just add a tool. Change behavior.
If the answer is yes, find your champion. Not a junior hire. A partner-level person who has the credibility to drive adoption and the curiosity to understand what’s actually possible.
Then get ruthless about what you build. If a product doesn’t have senior ownership, kill it. If your data team isn’t sitting in on investment calls, they’re building in the dark.
Most firms won’t do this. Melody’s prediction—that many will give up within twelve months—isn’t pessimism. It’s pattern recognition. The firms that succeed will be the ones who treat this as a multi-year compounding bet, not a quick fix.
The full conversation went deeper on specific tools, time audits, and whether LLM providers will enter the VC space directly. Be sure to check it out above!
