This VC uses AI to spot startups before the data catches up

Marc Palet co-wrote this article.

Deal sourcing – the process of finding, assessing, and securing investment opportunities before competitors snap them up – is a core function for every venture capital firm. A portion of a fund’s returns hinges on the quality of the deals that its principals bring in.

It’s also one of the most time-consuming parts of the job, so it’s no surprise that every fund is trying to figure out how to use AI to help identify investable companies.

Made by Ulla/Tech in Asia with the help of AI

When we implemented AI in our deal sourcing, we expected that we would be able to do the same work faster. What surprised us were the new insights we gained and the work that we were able to do that we couldn’t before.

We found that the real edge wasn’t speed, but finding out about investable companies much sooner in the process. It’s an unstructured, open-world problem that old-school database monitoring was never built to solve, and it’s exactly what generative AI is suited to handle.

What we used to do

At OM Venture Capital, we used to source deals by running a few strategies in parallel.

We started with startup database searches, filtering based on criteria that fit our thesis. It was a useful approach, but these searches surfaced thousands of companies, with no way to prioritize which one to call first.

Worse, they couldn’t tell us which deals were currently actionable. A startup could go 18 months without raising any funds or shipping any meaningful products and still show up as a match for our investment.

It became clear that we needed signals that told us which companies were worth a call this week, not more matches.

See also: How I built an AI analyst for my firm, no engineers needed

So we also scoured newsletters and tech-focused sources that posted whenever a startup raised funds or made relevant headlines. These signals worked better than the searches did because the companies that showed up were ones that had something happening.

But every signal meant someone on the team had an article to read, and our sources started producing more articles than we could keep up with. They would generate over 2,500 signals a week, which was impossible for our team of four to manage.

As for deciding on which deals are actionable, other firms tackle this by licensing databases and watching for changes in the data such as headcount jumps, funding events, or traffic spikes.

What AI helps us do now

The results