They guess. We compute.

Why semantic job matching failed for twenty years — and the one structural change that finally unlocks it.

Everyone has felt it. You search for a job, and by the third page of results the machine has stopped understanding you — roles that have nothing to do with what you asked. You give up. Not because the right job isn't out there, but because you can't find it, and you can't trust the search.

Multiply that across the economy and it has a price tag. Skills mismatch costs around 6% of GDP, and by 2030 up to $15 trillion in lost output (BCG). The most important matching problem in the economy still runs on guessing job titles.

It's not for lack of trying

For twenty years this looked solvable. Dozens of funded startups and the biggest platforms on earth have chased semantic job matching. Most were quietly acquired: Bright into LinkedIn, Trovix into Monster, Textkernel into Bullhorn for €300M, Eightfold to a $2.1B valuation. The demand was never in doubt — and notice the pattern: the giants didn't crush this category, they bought it.

And yet the experience never fundamentally improved. Why?

The structural flaw: everyone guesses

Because almost every system works the same way — top-down. Take a job, let AI break it into data, then estimate how similar two things are. Guessing has no floor. Change one word in a posting and the results move. It can't be exact, and it can't explain itself. You get ranked guesses, not answers — which is why search scatters into noise, and why bolting "more AI" onto the same approach just produces more confident noise.

The unlock: compute, don't guess

JobID works the other way around — bottom-up. Underneath everything sits one shared model of work: skills, tasks, context. Every job, CV, training and search is translated into that single model. We don't estimate whether two things are similar — we compute it against the same ground truth. The result is matching that is exact, explainable, and never breaks into irrelevance, with a transparent score and the precise reason behind every match.

Think Nutrition Facts. Once every product is described in the same standard units, you can finally compare and rank them. JobID does that for work.

They guess. We compute.

For a person, that means discovery that understands what you can actually do — not the titles you have to guess — and a search that quietly becomes your profile, so the right work starts finding you. For a company, the same engine matches employees to new roles, trainings and career paths, across departments and languages.

Why now

Work is being rewritten by AI. Titles inflate, skills shift monthly — and AI needs structured, computable work data that simply doesn't exist yet. Whoever owns the shared model of work owns a layer that LinkedIn, SAP and Indeed all need and none of them has built. It's a ~$870B market today, heading to $1.7T by 2030, with no infrastructure player yet.

Where we are

This isn't a pitch for an idea. The hard part — the research — is done: four years of R&D, a working engine with millisecond retrieval out of an ontology built at exactly the right depth, a working prototype, and a Letter of Intent from Siemens. What's left is execution.

We're looking for the investors who want to be early on the open data standard for work — before it becomes obvious.