March 17, 2026

Beyond Search: Why Ranking Is the Hard Part of Sourcing

Search returns thousands of matches. The bottleneck was never finding companies, it's ordering them by who is actually worth your time. That ordering is a real machine-learning problem.

Run a search across private companies and you'll get thousands of matches. That's not the problem. The problem is that a list of thousands is almost as useless as no list at all. The bottleneck in sourcing was never finding companies. It's ordering them.

Filtering and ranking sound similar. They are completely different problems, and most tools only do the first one.

Filtering Treats Every Match as Equal

A filter answers a yes-or-no question: does this company match my criteria? Everything that passes lands in one flat pile, in no particular order. A company you'd kill to own and one you'd never call sit side by side, indistinguishable. The tool has done its job and left the actual work, figuring out where to start, entirely to you.

At small list sizes that's fine. At the scale real sourcing produces, it just moves the bottleneck from finding to triaging. And triage by hand has a hard ceiling. A person can give real attention to maybe a few dozen companies before the list blurs into noise, so what actually happens with a flat pile of two thousand is that you work the top of it, the alphabetical accident or the order the database happened to return, and the rest never gets looked at. The filter didn't help you find the best companies. It just guaranteed the best ones are sitting somewhere in the part you'll never reach.

Ranking Is a Different Problem

Ranking asks a harder question: of everything that matches, which few are most worth your time right now? That's not a filter. It's a score, a single number that has to fold together everything known about a company and an owner and put the best ones on top.

Getting that number right is where the actual machine learning lives, and it's genuinely hard. Hard in a specific way: the thing you most want to predict, whether an owner will actually sell, is the thing you have almost no labeled examples of. Most companies never transact, so you're learning a rare event from sparse, messy evidence. And you only ever see outcomes for the companies someone chose to pursue, never the ones that got passed over, so the data you learn from is shaped by past decisions rather than what would have happened. Get the score wrong in the cautious direction and you bury a great company on page nine. Get it wrong the other way and you send your best hours chasing names that were never going to move.

Weak Signals, Combined

No single fact tells you a company is worth pursuing. Owner tenure, a recent change in the business, ownership structure, industry, signs of a succession gap, each one is a weak signal on its own. The skill is combining many weak signals into one reliable ranking, weighting them correctly, and not being fooled when a few are missing or wrong, which in private-company data they usually are.

That's a modeling problem, not a search problem. It needs noisy, sparse data turned into a calibrated score, and it needs to hold up when the inputs are incomplete.

A Good Ranking Learns From You

The best version of this doesn't rank the same way for everyone, and it can't, because there's no universal definition of "best company." A business one buyer would kill for is one another wouldn't return a call about, so a single shared ranking is wrong for almost everyone by construction. The only fix is a ranking that learns the person. It watches which companies you actually pursue and which you pass on, and treats every one of those as a label: this is what good looks like to you, this isn't. Your passes are as informative as your pursuits, sometimes more, because they're where your taste diverges from the generic score. The ranking that was decent on day one becomes genuinely yours over time, and that's also why it's hard to copy: a competitor can buy the same data, but not the thousands of small judgments you've taught it.

That's how we think about it at Equitle. Finding the companies is the easy half. Ranking them, so your attention goes to the handful most likely to turn into something, not a flat list of thousands, is the half that actually saves you time.

Shrey Parekh

CTO, Co-Founder, Equitle