What this covers
Patrick interviews Ryan Caldbeck of CircleUp about building quantitative models for private consumer investing, a field where sourcing has traditionally run on trade shows, cocktail parties, and star analysts. The conversation starts with CircleUp's first model, a Classifier trained to replicate the founders' own yes-or-no decisions, and moves on to its later platform, Helio. Caldbeck says 8 of the 10 most predictive inputs in that first model were data the company never had to ask applicants for. He also explains when interpretable models are needed and when a black box is acceptable, as in entity resolution and industry classification.
The middle of the discussion is about data and its surprises. Standard providers cover only about the top 50 grocery retailers, so CircleUp collects doors data itself. Caldbeck uses Halo Top to show how shared data produces crowded reactions: roughly 18 of the top 20 consumer PE firms called within two weeks, after the growth had already happened. Some of his claims are contested, including that offline retail matters far more than Silicon Valley assumes and that D2C is good for testing but poor for scaling. Other findings are odder. Units sold per store per week showed no relationship with success, which he explains through what retail buyers want. Computer vision on snack bar packaging found that more words on the front went with slower revenue growth. Companies tweeting "send me a direct message" were three times more likely to fail.
The closing stretch covers how hard private-market quant is to prove to LPs, given short backtests and long feedback loops. It also covers how CircleUp runs itself: quarterly OKRs capped at three to five goals, and a mission that led it to decline dozens of public-company requests to use Helio.
Caldbeck argues that quantitative models can be built for private consumer and retail investing by replacing subjective judgment with interpretable signals (distribution, brand, and product uniqueness) validated against revenue growth, and that consumer's repeatable business models and abundant data make it the most tractable private market for this approach.
- Revenue growth serves as the objective outcome measure because private-market exits are too sparse to model directly
- Consumer companies share repeatable business models and leave abundant public data trails, unlike tech
- Distribution breadth, brand engagement, and product uniqueness predict success, while widely tracked metrics like velocity do not
Returns hinge on a few core public-market factors and on disciplined early-stage valuation.
- In public markets, a small number of thematic factors such as value, momentum, quality, and volatility drive most predictive power, and finding marginal improvements on top of them is very hard, a point Patrick raised as a framing for the consumer factor search.
“a pretty small number of thematic factors, let's call them, that really drive the vast majority of the predictive power.”
- Early-stage investors may identify great future brands or growth but still lose money when valuations are too high, so valuation discipline is critical to returns.
“any good revenue growth could be destroyed by too high a valuation.”