Data driven investing in funds

For most LPs, Evan Finkel said, choosing a venture fund goes something like this. Someone sends a deck. There is a call or two. The fund turns out to be top quartile, as all funds are, with a couple of good markups, and the cheque goes out. That holds even at large, well-known LP organisations. Evan runs venture investing at Integra Global Advisors, a US multi-family office, and has built a process that asks far more of managers. It rests on an awkward fact about venture. No single number becomes reliable in time to act on.

Integra backs small pre-seed and seed funds, mostly a manager's first or second, many of them run by solo GPs. The first filter is size. Integra caps funds at $100 million, so a fund targeting $110 million is dropped automatically, however good it might be. An AI system Integra has trained then screens for a credible GP, a sensible strategy and a good fit between the person and the thesis. That last test matters more now that every fund is an AI fund, as every fund was a crypto fund five years ago. Many talented investors, Evan said, have no real right to win in the space they target.

From about the second call, Integra asks for the portfolio companies' KPIs and how they have moved over time, from the quality of revenue and customers to the quality of hires. TVPI, Evan said, is "not a particularly important part of our process". The team also reads the investment memos itself, without handing them to Claude or ChatGPT, and asks GPs what they got right and wrong, and why. Most LPs ask far simpler questions, Evan said, and need not copy Integra's whole process. "You just need to care a little bit and pretend it's your money and ask some deeper questions."

Joel Larsson, a co-founder of Pale blue dot, which organises The Drop, moderated, and asked when an LP can trust a number such as TVPI. Venture deals in small samples, Evan said. By the time a GP has made enough investments for the record to be statistically stable, they are probably near retirement age. Baseball has the same problem. Batting averages take several seasons to stabilise, so judging a batter on one season means trusting a statistic that is not stable at all. Evan's answer is to track several KPIs over time and ask whether they tell a similar story.

Track records do carry over, with one condition. Persistence is strong in venture and private equity, Evan said, as long as the strategy does not change. The usual failure is an exceptional GP who raised $10 million and is tempted to raise $50 million or $100 million, and the strategy changes with the size.

Joel doubted that history travels well even when a strategy stays put, and called the doubt a gut feeling with no data behind it. A deep-tech fund raised two years ago has perfect timing, one raised seven years ago less so. Evan agreed the error bars should be wider and said data should never be the only input. The alternative is a finger in the air, deciding a manager is great because of two or three markups, and Evan found it very hard to believe that was a better strategy.

Integra also spends a lot of time with founders and almost none with other LPs or co-investors. The founder world is small, Evan said, and founders know who the good GPs are, which makes them a leading indicator of whether a GP will keep attracting good ones.

Better data is coming. Soon, Evan said, AI systems will collect portfolio data and turn it into a form LPs can digest. Using it is another matter. Integra's own quarterly commentary on venture is read less often than Evan would like. Joel's firm sends LPs polished reports built for human readers, and has started cutting data from them to see whether anyone complains.

Evan Finkel gave this Expert Session at The Drop 2026 on 16 September, moderated by Joel Larsson of Pale blue dot.

You just need to care a little bit and pretend it's your money and ask some deeper questions.