ML underwriting, four enterprise customers, and a sale

I decided to refocus Jumpstart. The original marketplace was not going to work because the two sides wanted fundamentally different things. The available projects were too risky for the investors I could reach, and the investors who could move real capital operated on multi-year timelines and relationship-driven processes that looked nothing like a liquid marketplace.

I made three decisions at once. First, focus on U.S. solar markets, where at least there were limited sovereign risks. Second, focus on residential rather than commercial. I wanted to leverage my data science background, and there was far more data to work with at the residential scale. 90% of commercial and industrial entities do not even have a credit score, which makes commercial financing vastly more complicated. Third, focus on a single side of the marketplace by providing a SaaS service, before trying to build out both sides. Residential gave me a tractable problem.

I built an initial model using public solar-program data, census data, weather data, and macroeconomic data. The first version landed around 0.75 F1 and 0.85 AUC-ROC. I would not compare those metrics directly to FICO, because FICO is a score and those are classification metrics. They are not the same kind of output. The point was not that we had found some magic number. It was that we had built a system with enough real predictive signal to create customer trust and open a path into a market that standard credit thresholds were screening out too crudely. We eventually got to four paying enterprise customers.

The habit I built here that I still use every week is treating sales as a learning system. I maintain a live hypothesis document. I kill things when the data comes back cold. I try to be honest about what I am actually seeing versus what I want to be seeing. An advisor I worked with later told me I was unusually willing to change direction when a thesis died. That willingness started at Jumpstart.

Jumpstart was the first time I saw a company change shape because reality forced it to. The marketplace idea was ambitious and intellectually satisfying. The underwriting product was what the market could actually absorb. You do not get to keep a thesis just because you like it.