Venture firms love to talk about what they’ve built. New dashboards, sourcing models, founder portals. Ask any LP and they’ll tell you about the sea of sameness coming out of any VC annual meeting. But when I look at the infrastructure supporting our own investing work, what stands out the most to me isn’t what we’ve built. Don’t get me wrong—I’m super proud. Instead, it’s what’s missing.
Adding AI simply as a tool or layer is limiting its power. What I’m talking about is rethinking foundational systems that change the way that we approach the game. Ones that:
- Help us understand where the market is going (not where it’s been)
- Surface the right company at the right time (not just the loudest in this cacophony of startups)
- Turn patterns of human behavior into actual investment signal
For me, this thinking is more of a map of the white space. A reflection on infra that’s on my wishlist. It’s not exhaustive, more a conscious turning of the wheel toward where we think data-powered investing is headed.
The messy middle: where infrastructure actually matters
Where we sit is in a hard to automate zone. We’re not pure seed where it’s founders and ideas. We’re not late where it’s traction and market dominance. We live in the messy middle, where discovery and prioritization is the name of the game.
You can’t outsource that to a single model. You need an ecosystem of tools—data pipelines, feedback loops, prioritization logic—that makes the ambiguity more navigable.
The signals we wish we had (and some we’re already working on):
We’re not short on data. But the signals that matter most don’t come through pitchbook exports or form fields. They show up in behavior and sentiment, which are notoriously hard to quantify (ask any marketer what number they use to measure brand and watch their head explode). So, how do we measure these subtle, nonlinear, distinctly human signals?
- People scoring as career votes: Great people vote with their career. When top tier engineers, designers, or go to market leaders join a team early, that’s a powerful endorsement, often revealing more than just the product pitch. We want to build models that recognize and score those patterns of affiliation.
- Hiring velocity and pedigree fusion: Speed matters. So does who’s joining and from where. Hiring three engineers in a month says one thing. Hiring three ex-Stripe engineers tells a clearer one. We want to not only see how fast companies are hiring, but also who they are hiring AND from where.
- Proactively timing rounds: One of the most frustrating experiences in VC? Finding a company after the round has already been done. Our infrastructure should help us predict fundraising intent vs. giving us a frustrating datapoint after the fact.
- Market blooms: A single seed company may not tell you much. But when ten similar companies pop up in the course of three months, something’s happening. Our models should detect these blooms as clusters of innovation that indicate where the puck is going. Done well, this helps us think about the right companies and quantify the emergence of categories.
The dual challenge of discovery and prioritization
Finding the most companies does not equal investing in the right ones. We need to find the right ones and, just as importantly, at the right times.
That’s the core of the infrastructure challenge:
- Building the long list (discovery): How do we surface stealth teams, under the radar GTM traction, or latent hiring signals that suggest an impending inflection?
- Making the short list (prioritization): When we’re staring at a list of 25 seemingly strong companies, how do we make sense of it? How do we know where to spend time and who might break out the fastest?
Why we haven’t built it yet, and why we’re doing it now
There’s a reason this hasn’t been done, or hasn’t worked well when attempted.
The first is there’s often either no data or bad data. The art of venture is finding the new, next thing, so data is sparse, especially in early stage and emerging sectors, aka where the money is. This makes VC a weird edge case. It’s non-standardized, non-repeatable, and intuition heavy.
The perhaps biggest reason, though, is usability difficulties. There’s a lack of explainability in high-performance models. In an environment where we want people to show their work, we’re making our models do the same. Our most precise early models didn’t pass the investor trust test. The best infrastructure has to be accurate and usable. And in venture, usable means investors can interrogate it, pressure test it, and gut check it.
Who we’re looking for
To me, data science isn’t a support function. We’re embedding it as part of our strategy.
This is where you come in. We’re hiring someone to lead this work: someone who is excited to build high-trust infrastructure that earns investor buy-in, and ultimately assists with their decisions. You’ll have the chance to leave your fingerprints on how a modern VC firm is run.
The next generation of venture firms won’t just invest better. They’ll see better, earlier. And that’s only possible if the systems behind them get smarter, faster, and more useful.
If that sounds like a challenge worth chasing, we should talk.
