We didn't set out to start a software company. We built a spreadsheet to stop arguing with each other about scores, and it got too good to keep to ourselves.
Scoring by feel, for a decade
Harlem Capital has spent ten years underwriting founders who don't look like the ones most firms default to, and building the operating muscle to place talent inside our portfolio once we've backed them. For most of that decade, both jobs ran on feel. A partner would meet a founder, form an impression, and defend it in IC with a mix of notes, gut, and whoever argued loudest. Hiring for portfolio companies was the same exercise wearing a different hat: a recruiter would forward five resumes, and we'd rank them the way you rank anything with no rubric, by vibe.
That worked well enough when deal volume was low and the team was three people who'd built the same instincts together. It stopped working once we scaled. Two partners could look at the same founder scoring sheet and land on opposite conclusions, and neither one could point to the line that explained the gap. That's a bad way to run due diligence on a check size that doesn't come back.
Nothing on the market fit
Before we wrote a line of our own code, we shopped. The background-check vendors were built for compliance teams verifying employment dates and criminal records, not for a partner trying to figure out if a founder's last company actually shipped or just raised. The ATS scoring tools were built for high-volume corporate recruiting funnels, tuned to filter out resumes at scale, not to help a two-person deal team reason about one candidate for a portfolio company's first sales hire. Both categories were solving adjacent problems with adjacent data. Neither one touched the specific judgment calls a fund actually has to make.
We weren't imagining this gap. A March 2026 Forbes Technology Council piece on the buy-versus-build decision in AI makes the case we kept landing on ourselves: buy the commodity and the regulated, build the layer where your judgment is the product. Our diligence process and our portfolio talent pipeline were exactly that layer. A February 2026 GoingVC survey of AI-native venture firms found the same thing happening across the industry: GPs rebuilding sourcing, diligence, and portfolio support in-house because nothing off the shelf was shaped for how a fund actually decides.
A shared rubric, not a shared vendor. Every score tied to the specific evidence that produced it, the same way we now break down a founder's public commit history in what a founder's GitHub actually tells you. The rubric came first. The software came later, once the rubric had survived enough real IC meetings to trust.
When the tool became the point
We kept the internal tool alive for two years before anyone outside the firm saw it. It scored inbound founders. It scored candidates for portfolio CEOs who needed a first VP of Sales. It settled more IC arguments in ten minutes than a partner memo settled in a week. Other GPs who sat on our portfolio company boards started asking where the scores came from, and then asking if they could use the same thing.
That's the actual founding moment of ScoringFactory. Not a whiteboard session about market size, but a tool we'd already leaned on for two years of real decisions, that other funds wanted access to because it had already been stress-tested on money that wasn't imaginary. Matt, my co-founder, took the internal build and rebuilt it as a product a fund outside Harlem Capital could run on its own pipeline. I stayed on the GTM and investor side, because the pitch to another managing partner is easier when it comes from someone who still sits in IC every week.
Why staying close to a fund helps everyone
The instinct with a spinout is to detach it from the mothership so it can chase a bigger market unencumbered. We did the opposite on purpose. ScoringFactory still runs on Harlem Capital's live deal flow and live portfolio hiring, every week, which means the rubric gets tested against real outcomes instead of hypothetical ones. When a scored founder raises a Series A eighteen months later, or a scored VP candidate flames out in month four, that result feeds back into the model. A tool built by consultants who've never sat in an IC meeting doesn't get that loop.
It also keeps us honest about who the product needs to work for. Harlem Capital's mission is 1,000 investments in diverse founders over 20 years, and the funding data keeps reminding us why that mission still matters: Black founders posted their strongest funding quarter in nearly four years this June, and quarters like that are still the exception, not the rule. A scoring tool that quietly re-encodes pedigree bias into every rubric would work against that mission inside our own fund before it ever hurt anyone else's. Building where we could see the damage first was the whole point.
Other funds don't need our mission. They need a rubric that survives contact with a real IC and a real portfolio hire. Staying inside that loop is how we keep building the version that actually does.