How to start hiring on skills, in one role

Why company-wide rollouts fail

A skills-based hiring initiative that starts with a policy has to be agreed by every hiring manager before it can be tested by any of them. So it gets negotiated down to whatever nobody objects to, which is usually the degree requirement and nothing else, and eighteen months later somebody notices that the decisions are made exactly as they were before.

One role does not need agreement. It needs one manager willing to try it on one opening, and it produces something the next conversation can point at: this is what the rubric was, this is what we saw, this is who we hired and how it went.

Step one: write the three things, before you post

Not a job description — those are written to attract, and they list everything. Three capabilities that, if this person turns out not to have them, mean the hire has failed. Three, because four means you have not decided, and every list of eight is a list of zero.

This is the step teams skip and it is the one doing most of the work. It forces the disagreement that would otherwise surface in week six of interviewing, when two people on the panel discover they have been assessing different jobs. Having that argument in advance is uncomfortable and takes forty minutes; having it after eight interviews costs the process.

Write it down before any candidate exists, and do not change it once they do. A capability list edited after meeting someone is a description of that person.

Step two: everyone does the identical thing

Whatever you assess with — a short exercise, a set of situations, a structured interview — every candidate gets the same version. This is not fairness theatre; it is the property that makes comparison possible at all.

The commonest violation is invisible: the panel goes off-script for the candidate they like, asking follow-ups that let them show more. It is generous and it destroys the comparison, because now one candidate has been given room the others did not get.

It also means resisting the customised exercise for the senior candidate. If the exercise cannot work for everyone at that level, it is the wrong exercise.

Step three: score alone, then talk

Everyone scores against the written rubric independently and submits before any discussion. Then you discuss — and what you are discussing is the disagreements, which is the only part of the conversation that carries information.

Discussion before independent scoring produces consensus around whoever spoke first, or whoever is most senior in the room. Everyone recognises this and almost nobody structures against it, because the meeting feels productive either way.

The five mistakes that kill this in month one

In roughly the order they happen.

• Assessing everyone instead of the shortlist. Skills-based hiring at the top of a 300-application funnel is expensive and unnecessary. Cut to a manageable number first, however you like, then assess properly.
• Writing the rubric after seeing the first submissions. The rubric now describes what you saw, and it will keep describing whoever you already prefer.
• Adding a fourth, fifth and sixth capability during the process. Every addition dilutes the weight of the ones that actually decide it.
• Making the exercise long enough to be 'thorough'. Length filters for free time, not ability, and your strongest candidates have the least of it.
• Not recording the outcome. If nobody writes down whether the people you hired this way worked out, the next role starts from opinion again and the whole thing was a gesture.

What to measure to know whether it worked

Two numbers, both cheap to collect, and one of them is uncomfortable.

First: how many candidates you interviewed per hire, before and after. Skills-based hiring should reduce this — that is where the time comes back — and if it does not, the assessment is not being used to decide anything.

Second, and this is the uncomfortable one: how many of the people you assessed well and rejected went on to do the job well somewhere else. You will not always find out. When you do, it is the only external check the process ever gets.

Frequently asked

Do we need to buy an assessment tool to do this?

No. The three steps above work with a document and a spreadsheet, and doing them badly with an expensive tool is worse than doing them well with neither. A tool is worth it when you are running the same assessment repeatedly and the scoring consistency starts costing you real time.

How long should the assessment take a candidate?

Under an hour unpaid, and enforced rather than suggested. Past that you are sampling who had a free evening. If you genuinely need a full day of someone's work to decide, pay for it — and know that payment fixes the ethics without fixing the sampling problem.

What if hiring managers refuse to score before discussing?

Make the submission the price of attending the meeting. It sounds bureaucratic and it takes about three minutes per candidate. The first time two people arrive at opposite scores for the same person, nobody argues about the rule again.

How do we handle candidates who ask for feedback?

A rubric-based process is the only one that can answer that question honestly, which is a benefit worth using. You can tell someone which capability came out low and what specifically they did that scored that way. That is not possible after an unstructured interview, which is why most companies have a policy of saying nothing.