Most research on founder diversity stops at the size of the funding gap. How little capital reaches women-led or Black-led startups is well documented at this point. The question that matters more, and that almost nobody asks, is what happens to the capital that does reach them.

Regulation Crowdfunding is a good place to ask it. Every offering since May 2016 is publicly disclosed, which means you can work with the entire population rather than a sample, and unlike venture capital the money comes from ordinary people rather than institutions. Jack Jesaitis, who spent the summer with us as a research intern, took all 11,140 offerings filed between 2016 and 2025 and tested whether underrepresented founders convert capital into results more efficiently than everyone else.

Two of the three groups he could measure do. Companies with a female founder and no minority founder are funded at twice the odds of companies with neither, and raise about 46 percent more, after controlling for deal size, industry, geography and security type. Companies with a minority founder and no female founder show the same pattern at a smaller magnitude.

Founders who are both do not get the sum of those two advantages. That is the finding worth your attention. If two independent effects simply stacked, a founder who is both female and a minority should be funded at roughly 2.7 times the odds and raise about 66 percent more. The actual numbers are 1.5 times the odds, and an amount-raised advantage that disappears entirely into statistical noise. Their rate of coming back for a second raise, 11.8 percent, is the lowest of any group in the data, lower even than founders with neither identity. That shortfall appears independently in three unrelated outcome measures, which is what moved it from an interesting coefficient to the central finding of the report.

We cannot tell you why. The data establishes that something structural happens at the intersection of both identities. It does not establish what that something is. Backer network access, compounding bias and industry composition are all plausible, and this dataset cannot separate them. That is the honest state of it, and it is the question we would most like someone else to take further.

The full report is at cclear.ai/research/doing-more-with-less. It is free and ungated. The underlying CCLEAR data is available under license to anyone who wants to extend the work.