What we learned searching 1B+ profiles
Four lessons from searching a database of 1B+ people. Precise criteria beat broad queries, hard facts beat skill keywords, and the best candidates are not applying.
Searching a database of 1B+ people teaches you quickly what actually finds great candidates and what just produces long lists. We built 10xTable Sourcing on top of that database, and we ran real searches for real roles until the results were good enough to send to hiring managers. Four lessons survived that process. Precise criteria beat broad queries, location has to be a first-class criterion, hard facts filter better than skill keywords, and the best candidates for your role are almost never in your applicant pool.
Lesson 1, precise criteria beat broad queries
With a billion profiles, finding people is not the problem. Everything matches something. Type "senior backend engineer, fintech experience, Europe" and you get an enormous list of plausible people. Plausible is the enemy. A long list of maybes is not a shortlist, it is homework.
What works is the opposite of a clever query. Break the role into explicit criteria, each one checked on its own against each profile. The test for a good criterion is whether a stranger could answer yes or no from the profile alone. "Strong engineer" fails that test. "Has built and operated production machine learning systems" passes it.
There is a calibration to this. A criterion that is too loose lets almost everyone through and tells you nothing. A criterion that is too strict starves the search and you end up staring at three names. The craft is writing criteria tight enough to mean something and loose enough to leave a real pool. We got this wrong in both directions before we got it right, and it changed our results more than anything else on this list.
Lesson 2, location must be a first-class criterion
Our early searches treated location as a preference folded into the query text. The results looked reasonable and were not. We got people "connected" to the target city. They had worked there once, studied there, or mentioned it in a bio. If your role needs someone in Berlin, "was in Berlin five years ago" is a different person from "is in Berlin now."
The fix was simple and complete. We promoted location to an explicit criterion, checked on its own for every profile, exactly like the role criteria. Once we did that, on-location results went from a frustrating minority of each search to effectively all of it.
The general lesson is bigger than geography. Anything non-negotiable about your role must be a criterion the search verifies, not a hint you hope the search picks up. Hints get diluted. Criteria get enforced.
Lesson 3, hard facts filter better than skill keywords
The most tempting way to search is by skills. It is also the least reliable, for two reasons.
First, profiles do not report skills consistently. Some of the strongest people we surfaced listed almost none, because they were busy doing the work. Meanwhile, thin profiles listed everything. Filter on a skill keyword and you systematically select for self-promotion, not ability.
Second, skill filters starve the search. Every keyword you require cuts away real candidates who simply phrased their experience differently. Stack three or four and you have filtered out most of the people you were looking for.
What works is a split. Filter on hard, categorical facts, meaning what this person actually does day to day, whether they are an individual contributor or a manager, where they are. These are stable, checkable, and phrased consistently enough to filter on. Then treat skills as questions to verify on each surviving profile, with the supporting evidence shown next to the answer. Facts decide who enters the pool. Skills, verified with evidence, decide how the pool ranks.
Lesson 4, the best candidates are not applying
The uncomfortable one. When we compared search results against real applicant pipelines for the same roles, the overlap was close to zero. The strongest profiles for a given role almost never appear among its applicants.
It makes sense once you say it out loud. The people you most want are employed, busy, and not reading job boards. Your inbound is not a sample of the market. It is a sample of who happened to be actively looking, at the moment your posting was live, in the places you posted it. Screening that pool perfectly still means choosing the best of who showed up.
Sourcing inverts the order. Define what great looks like first, search everyone, then go win the people who match. Screening your applicants still matters, and doing it fast with evidence matters. But if the process starts at "who applied," the ceiling on your hire was set before you read a single CV.
What we do with these lessons
All four are built into Sourcing. You describe your role in plain language. We turn it into precise, checkable criteria, with location and your other non-negotiables enforced as first-class checks, hard facts doing the filtering, and skills verified per profile with the evidence shown. The search runs across 1B+ people, and every answer about every candidate is clickable through to its source, the same standard we hold in Screening.
You do not need an ATS, an integration, or a call with us to try it. Describe a role and look at who comes back.