Report: 10 minutes with a fake client may help decide who gets a McKinsey job

A graduate who wants to be a business analyst at McKinsey may soon have to spend 10 minutes talking to a client who does not exist.

The Verge reported on October 14, 2026, citing The Wall Street Journal, that the consulting firm plans to add an AI-simulated client conversation to some entry-level interviews. The test would be for undergraduates in some North American offices, and hundreds of candidates may take part. ^[1] I have seen no statement from McKinsey itself confirming it, so this is a reported plan, not a settled one.

Here is my view. If McKinsey puts a simulation between young people and a job, it should publish what the exercise is meant to measure and how its assessors score it before it uses it on a single applicant. And whatever it publishes, a 10-minute chat with a synthetic client measures how well someone performs for a machine in a scenario somebody else designed. That is not the same as showing how well they will do the job.

What is reported, and what is not

The details are modest. The simulation is meant to reflect 80% of the scenarios McKinsey consultants might meet in their day-to-day work. ^[1][2] The AI would not score anyone. Human assessors would be present and would judge how candidates handle the situations put to them. ^[1] The simulated client is not built to behave like an unreasonable or out-of-control one. ^[1]

That is the whole public account. We do not know what the client will ask candidates to do. We do not know what the assessors will be looking for, which offices will run the exercise, or whether anyone has checked that doing well in it says anything about doing well at work. Nothing public answers any of that so far.

The best case for the firm

I want to give McKinsey its strongest argument, because it is a real one.

The machine is not the judge. A person is in the room, and a person decides. That matters. The loudest fear about AI in hiring is software that quietly ranks people and throws most of them out before a human ever sees a name. This is not that, at least as reported.

Nor is the simulated client a trap. A test that mimicked a tyrant would mostly reward people who have practised being bullied. A calm, plausible client is a fairer thing to ask a 21-year-old to talk to.

And let us not pretend the old way was a gold standard. Human interviewers are tired, biased, charmed by the wrong things and inconsistent from one afternoon to the next. If a standard scenario gives every candidate the same starting point, that is arguably better than a chat that depends on who happens to be across the table.

All of that is fair. None of it answers the question that counts.

A person in the room is not a window

A human assessor is only as accountable as the criteria that person has been handed. Those criteria are written by the firm. So is the scenario. So are the lines the client speaks, the conditions under which candidates are watched, and the meaning of a good performance.

A graduate walks into a room where every rule was set in advance by the party deciding whether to hire them, and nobody else gets to read the rules.

That is what I mean by a gatekeeper. Technology that helps a consultant draft a slide is a workplace tool. Technology that stands between an applicant and the job is a different thing, and it needs a different standard of scrutiny. The person who is judged cannot see the yardstick, and the public cannot either.

Take the 80% figure. It sounds like rigour. But it describes how much of the job the simulation is meant to resemble, and the account does not say how that was counted. It says nothing about whether the exercise predicts who succeeds. A test can touch every situation in a job and still tell you nothing about who will be good at it, just as a menu that lists every dish tells you nothing about the cook.

Performing for the machine

Now think about the candidate. You are 21, you have been given 10 minutes, and the thing you are talking to is software playing a client. You are also being watched by people with a score sheet you have never seen.

What gets measured in that room? Maybe it is the judgement McKinsey wants. Maybe it is also comfort with talking to a machine, how fast you work out what it responds to, how you speak under observation, and whether your style of communicating happens to suit the script. Some candidates will have used these systems for years and some will barely have touched one. Some will have communication styles or disabilities the script never imagined.

People warm up at different speeds, and plenty of capable ones take a while to find their footing with a stranger, let alone a simulated one. A short, staged encounter rewards the fast starter, and the job does not always want one.

These are not wild worries. They are the open questions the story leaves hanging. How will the firm check whether the exercise disadvantages candidates with different communication styles or needs? Can a candidate ask for another format? How will the conversations be recorded, kept or used afterwards? None of it has been answered publicly so far.

Why it reaches beyond one firm

If this were a one-off, it would be a curiosity. It is not likely to stay one. McKinsey is among the best-known names in graduate recruitment, and what it does gets noticed. My forecast, and it is only that, is that if the exercise looks efficient, other employers will ask why they do not have one too. Each of them could then write its own rules in private, and applicants would meet a different invisible test at every door.

That is why the demand has to come first, and why it is not hard. Tell candidates what the exercise is for. Publish the qualities it is meant to detect and the guidance assessors follow in scoring them. Show the evidence, if there is any, that performance in the simulation tracks performance in the job. Say how the firm will look for unfair effects, and say what a candidate can do if the format does not suit them.

None of that gives away the scenario or ruins the test. A driving examiner does not hide that you will be marked on mirrors and junctions. The rules can be public while the route stays secret.

What to watch

First, whether McKinsey confirms the plan and says in its own words what it is testing. Second, whether it publishes anything beyond a description. A firm that earns its living telling other organisations to measure things properly can surely do it to itself.

Until then, the fairest thing to say about the simulated client is that it is the only participant in the interview whose behaviour has been fully explained. Everyone else, candidate and assessor alike, is still guessing what it is for.