A shortlist lands in a founder's inbox by lunchtime. Ten profiles look relevant. The search took a fraction of the usual time.
That may be useful. It is not yet evidence that the hiring decision improved.
AI can reduce the work of finding and organising candidates. The value arrives only when the people are relevant, the claims can be checked and the hiring team understands why each person is there.
Start with the decision the shortlist must support
A shortlist is not a deliverable in isolation. It should help a team decide whom to meet and what it needs to learn from them.
Agree the few capabilities the role depends on, the context in which they must be used and the trade-offs you can accept. A founding engineer with product judgment and a mature-platform specialist may both be excellent, but they solve different problems.
If the brief is unresolved, technology can make the wrong interpretation arrive faster.
Ask what the evidence actually says
A matching title is not evidence of equivalent work. A skill listed in a profile is not the same as a result delivered with that skill.
For each candidate, look for the source of the claim, the work they owned and the setting in which they did it. Separate what is documented from what a model has inferred. If the source does not establish seniority, scale or a specific outcome, the shortlist should say so.
An honest unknown is more useful than a confident paragraph written to fill a gap.
Check relevance, not just volume
The useful test is how many introductions deserve a hiring manager's attention, not how many names appear on the screen. Review a small sample against the agreed brief, including people the system ranked lower.
Look for recurring misses. Is location wrong? Has domain exposure been mistaken for hands-on experience? Has a leadership title been treated as proof of building a team? Those errors can look persuasive at a glance.
Correct the criteria or the underlying data before increasing the volume.
Keep a human accountable for the call
AI may help map a market, organise evidence and highlight questions. It should not turn an uncertain inference into a fact or quietly reject someone because their CV uses different language.
Someone must be able to explain why the candidate was included, what is still unknown and what the next conversation should test. Keep sensitive candidate information within the agreed process and check that the tools used respect that boundary.
A person can move quickly without surrendering judgment. In fact, clearer evidence often makes the conversation shorter.
Measure the completed hiring step
Compare the time and effort required to reach a credible first conversation, the quality of those conversations and the amount of rework when a shortlist misses the brief.
Include the time spent reviewing outputs and correcting errors. A faster first list that creates more interviews with the wrong people has moved work rather than removed it.
The goal is not to defend a tool or a traditional process. It is to help a founder make a better decision with less wasted effort and a more honest experience for candidates.