McKinsey's 10-Minute AI Client Scores No One. The Partners Do.

If you're an undergrad applying for a McKinsey business analyst job in some North American offices this fall, you may finish your case and then spend about 10 minutes talking to an AI client on a laptop. The avatar won't score you. The partners in the room will, and they're watching for whether you listened and changed course, not whether you got through the AI.

The Wall Street Journal reported the pilot on Oct. 14, 2026. It changes how you should prepare. It also changes how you should write the "client-facing" bullets those same partners are going to ask you about.

What is the McKinsey AI avatar interview?

The McKinsey AI avatar interview is a roughly 10-minute laptop conversation with a simulated client that runs after the case challenge, with partners or interview assessors watching. According to the WSJ, McKinsey is piloting it this fall with undergraduates applying for business analyst jobs in some North American offices, and hundreds of candidates are likely to go through it.

• Where: inside McKinsey's offices, on a laptop.
• When: after the case, not in place of it.
• Personas: a mayor deciding where to build a stadium, and a "professorial" client who wants to understand the reasoning behind McKinsey's recommendations.
• Scoring: none from the tool. McKinsey partners told the WSJ the AI produces no scores or assessments of its own.
• Who judges: the partners or interview assessors in the room.

A role-played client is the interviewer-acted stakeholder in a consulting case: the person you present to, who pushes back, asks questions, and sometimes reveals facts that change the answer. McKinsey's pilot replaces the acting with software. The judging stays human.

Why the partners, not the avatar, decide

The partners decide because the avatar exists to hold the client constant, and that puts all the variation on you. Saleem Zafar, the Dallas partner who designed the pilot, has said evaluators look at three things: whether you heard what mattered, whether you acted on it or missed it, and whether you pushed your own agenda over the client's.

Coverage of the pilot calls the role-played client "the noisiest part of a consulting interview." When interviewers act the client themselves, each candidate meets a slightly different client, which makes candidates hard to compare. Give every candidate the same mayor with the same concerns and the same timing, and the differences between candidates stand out.

So "no score" doesn't mean low stakes. With the client's quirks removed, the partner's read on how you listened is the main thing separating you from the next candidate. Zafar's three criteria are what you're graded on:

1. Did you hear it? The client says something that reframes the problem.
2. Did you act on it? You changed the recommendation, or at least the order of your argument.
3. Did you serve the client or your script? You dropped the slide you rehearsed because it no longer answered the question.

Why practising against a hostile client is the wrong prep

Don't prepare for a hostile avatar. The simulation is meant to cover about 80% of everyday client situations and is not designed to play an "unhinged" client. You probably won't fail by getting rattled. You're more likely to fail by staying polite and sticking to your script.

One reported example shows how. The simulated mayor cares less about projected savings than about service disruption. A candidate who keeps returning to the financial benefits has missed the stakeholder's real objection, even if every number they quote is right.

This trips up strong case candidates because case prep teaches structure: lay out the framework, run the math, give the answer. The avatar round checks whether you'll drop the structure once the client tells you the structure missed the point. In practice that looks like:

• Notice the shift. "You mentioned disruption twice. That sounds like the bigger concern."
• Say it back. "So the question isn't the cheapest site. It's the site that keeps services running during construction."
• Change the recommendation out loud. "That moves me from Site A to Site B, and here's what that costs."

That last step is where most people stall. Agreeing with the client is easy. Saying "my answer just changed, and here's why" takes some nerve, and it's the step Zafar's criteria actually reward.

The avatar round is not the Lilli round

These are two separate pilots, and preparing for one doesn't cover the other. Since January 2026, McKinsey has also run an AI-assisted round in some US final rounds, mostly for business analysts, where candidates work with Lilli, McKinsey's internal AI assistant. CaseBasix, a coaching firm, first reported that round.

In the Lilli round you're in charge and the AI works for you. In the avatar round the AI is in charge of the conversation and you have to respond to it. Clever prompting won't help you with the mayor. What helps is catching the new fact, saying it back, and changing course.

If you want practice reps, McKinsey offers a free AI interview practice tool, which the firm says is meant to level the playing field for candidates who can't pay for consulting coaches. Management Consulted and Hacking the Case Interview both track how the AI rounds are changing.

Why your resume is the bigger filter

Your resume cuts more candidates than the avatar does. Among candidates who reach the interview stage, offer rates typically run 15% to 30%, and most filtering happens earlier, at the CV and cover letter stage.

Funnel estimates vary a lot. Prep sites put McKinsey at over 200,000 applications a year for about 2,000 new consultants. CEO Bob Sternfels has put the figure closer to a million applications a year. That works out to somewhere between roughly 100 and 500 applications per hire.

The resume also matters later in the process. McKinsey's evaluation still puts the most weight on case performance and the PEI. The PEI (Personal Experience Interview) is McKinsey's behavioral round. Where do interviewers find the story they probe? Often in your resume. A bullet that reads "stakeholder management" gets you asked about stakeholder management, and if there's no real story behind it, you'll be making one up while a partner watches.

This is also where the avatar round and your resume meet. Partners are watching for a candidate who changes course when the facts change, and the best evidence that you can is a bullet where you already did.

How to rewrite stakeholder management bullets into pivot stories

A strong stakeholder management bullet describes a pivot: what you planned, the fact that changed, what you changed in response, and what came of it. If a bullet has no "fact that changed," there's no pivot story behind it, and that's exactly what the avatar round tests and what a PEI probe will find.

Use this four-part structure:

1. Original plan: what you set out to do or recommend.
2. Fact that changed: the new information, usually from the stakeholder.
3. What you changed: the decision, scope, or recommendation you revised.
4. Result: what happened, measured if you can measure it.

Here are some stakeholder management resume bullet examples, before and after. The bracketed figures are placeholders. Fill them with your real numbers or leave them out.

Before: "Client-facing role managing stakeholder relationships for a student consulting project."

After: "Proposed a pricing increase to a campus café owner; after she said her real worry was losing regular customers, reworked the plan into a loyalty tier that raised revenue [X%] over [N] weeks without a price change."

Before: "Strong stakeholder management across cross-functional teams."

After: "Built a dashboard spec for the sales lead; when the ops manager said reps updated the CRM only on Fridays, switched the refresh to weekly and cut the report from [N] metrics to [N], which the team adopted."

Before: "Presented findings to senior leadership."

After: "Recommended closing the lowest-traffic branch; when the director said the branch anchored a community partnership, changed the recommendation to reduced hours, keeping [X] in savings while the partnership continued."

Every "after" bullet answers the question a partner will ask next ("what changed your mind?") before they ask it. They also read like a transcript of a good avatar round: heard it, said it back, changed the answer.

Most people don't remember these stories when they sit down to write a resume. They're buried in old project notes. Refolk builds your resume from your own history, so the moment where a stakeholder changed the plan comes back out and becomes a bullet, instead of getting flattened into "client-facing."

How to prepare for the McKinsey business analyst interview

Prepare for three rounds: the case, the PEI, and the avatar. The skill linking them is changing your answer out loud when the facts change. Here's a two-week plan:

1. Days 1 to 4: Pick three pivot stories. Each needs a real fact that changed and a decision you revised. These cover your PEI and your rewritten bullets.
2. Days 5 to 7: Rewrite your resume to match. Every "stakeholder," "client-facing," or "cross-functional" bullet should point to one of those three stories.
3. Days 8 to 10: Run cases with a partner who interrupts. Ask them to drop one new fact midway (a budget cut, a political constraint, a timeline). Practise the notice, say it back, change the recommendation sequence.
4. Days 11 to 12: Use McKinsey's free practice tool for structured reps, and run the Lilli-style AI collaboration separately if your office uses that round.
5. Days 13 to 14: Rehearse the mayor. Have a friend play a stadium-siting mayor who cares about disruption, not savings. If you mention savings a third time after they've told you that, start over.

The resume work in steps 1 and 2 is the slowest part, because you're tailoring the same stories to McKinsey, then to other firms and corporate strategy roles, each with its own posting. Refolk handles that: paste the posting and get your own resume back rewritten for it, with a cover letter drafted and a fit score that shows how well your stories line up with what the posting asks for.

Will the avatar round spread to other roles?

Maybe, but nothing is announced. Expansion to other roles in 2027 depends on the pilot working, and McKinsey hasn't publicly set the success metrics that would decide it.

If you're an experienced hire or an MBA candidate, don't assume you'll face it next cycle. Do assume the skill it tests is already being judged in your case and PEI rounds. Sternfels has said McKinsey built an AI "workforce" of 20,000 agents to support its human employees. A firm with 20,000 agents doing support work hires people for the judgment the agents can't provide: reading a client and changing course. That holds whether or not the avatar ever reaches your interview.

FAQ

Does the McKinsey AI avatar give you a score?

No. McKinsey partners told the WSJ the AI tool provides no scoring or assessment of its own. Partners or interview assessors in the room judge your performance, using criteria pilot designer Saleem Zafar described: whether you heard what mattered, whether you acted on it, and whether you put your agenda ahead of the client's.

Is the McKinsey AI client interview the same as the Lilli round?

No. The Lilli round, piloted in some US final rounds since January 2026, has you directing McKinsey's internal AI assistant as a tool. In the avatar round the AI plays the client and you respond to it. Prompting skill helps in the first and does little in the second, where listening and pivoting matter.

Does the avatar round count more than the case?

No. McKinsey's evaluation still weighs case performance and the PEI most heavily, and the avatar is a pilot that runs after the case. Most candidates are cut before any of it: offer rates among those who reach interviews typically run 15% to 30%, and most filtering happens at the CV and cover letter stage.

How should I practise a consulting interview AI role play?

Practise with a partner who drops a new fact into the middle of a case, then work on noticing it, saying it back, and changing your recommendation out loud. Skip hostile-client drills. The avatar is meant to cover about 80% of ordinary client situations and is not designed to play an "unhinged" client.