Most AI Hiring Tools Automate the Wrong Step
Teams adopting AI in hiring almost always start with outreach. It's the easiest step to automate and the least important. Here's the step that actually decides whether a search works.
Teams adopting AI in hiring almost always start in the same place: outreach. Connection requests, follow-ups, multi-step sequences. It's the easiest part to script and the easiest to put a number against.
It's also the part that matters least. If the definition of the role is vague, automation doesn't fix it — it distributes the vagueness faster, to more people, with better grammar. You get more conversations with candidates who were never right, and a hiring manager who quietly stops opening the tool.
The bottleneck in hiring has never been sending. It's deciding what good looks like before you go looking.
Five People, Five Different Roles
Ask five members of a hiring panel to describe the same job and you'll get five jobs. One is hiring for domain knowledge, one for raw slope, one for a specific tool, one for someone who won't need managing, one for someone who'll accept the band.
None of them are wrong. But until those definitions are reconciled into one, everything downstream is guesswork. Sourcing gets built on whichever version the recruiter happened to hear. Screening is inconsistent because each reviewer applies their own bar. Outreach describes a role that doesn't match what the candidate will be interviewed for. And rejections can't be explained, which means nothing is learned from them.
The cost surfaces three weeks later, as a pipeline of plausible profiles the hiring manager rejects one by one without being able to say why. It's ultimately the same failure that produces a bad hire — an unexamined bar, applied inconsistently.
Why Forms Make It Worse
The standard software answer to this is a form. Title, seniority, location, must-haves, nice-to-haves, salary range. Submit.
Forms fail for a specific reason: they ask for conclusions rather than reasoning. A field labelled "must-have skills" invites a list of technologies. It doesn't surface the thing that actually determines fit — the shape of the problem this person will be handed in month one.
Faced with an empty field, most people write the generic version of the role. Not because they lack an opinion, but because writing a specification is a different skill from having one. A good recruiter doesn't hand you a form. They ask questions, push back on vague answers, and reflect the role back until you say "yes, that's it." That conversation is where the specification actually gets built — and it happens to be something a language model can do at any hour, at your pace, without a calendar invite.
What Belongs in a Usable Brief
A brief is usable when someone else could screen candidates from it without asking you follow-up questions. In practice that means:
- The problem, not the title. What's broken or missing that this hire fixes.
- Career stage, not years. "Has owned a roadmap end to end" is checkable. "5–7 years of experience" isn't.
- Environment fit. Company size, stage, degree of structure they'll be walking into. A strong operator from a 2,000-person org often fails at 20, and the reverse is just as true.
- Hard constraints. Location, work authorisation, language, compensation ceiling. The things that disqualify regardless of how good the profile looks.
- Explicit exclusions. The backgrounds that look adjacent but consistently don't work out. Usually the most valuable lines in the whole document, and the ones nobody writes down.
From Brief to Scorecard
A brief written in prose is useful to humans. It becomes useful to software when it's decomposed into criteria that can be evaluated independently — a weighted list of things a profile either demonstrates or doesn't. This is what skills-based hiring actually requires, and why dropping degree requirements on its own achieves nothing.
The important output isn't the score. It's the breakdown. A candidate at 78% tells you almost nothing. A candidate at 78% who matched on five criteria and missed on two — one of which turns out not to matter — tells you what to change. That's also where scoring stops being a black box. If the assessment shows its reasoning per criterion, a hiring manager can disagree with it specifically instead of dismissing the whole system.
Then it compounds. Sourcing targets the shape of the profile instead of the keywords in a title. Screening stays consistent across reviewers and across weeks. Outreach can reference something real, which is the only thing that reliably moves reply rates. And when the manager rejects a high-scoring candidate, you know which criterion was wrong.
That last one matters most. Hiring is an iterative search. Without a written definition of the target, you can't iterate — you can only re-roll.
Automate the brief, and everything after it gets cheaper. Automate the outreach first, and you've built a faster way to be wrong.
Umamy turns a conversation with the hiring manager into a structured brief and a weighted scorecard — then assesses every candidate against it, with the reasoning attached.