McKinsey’s AI Skills Test Reshapes the Final-Round Interview

McKinsey is now testing job candidates on its internal Lilli AI in the final round of hiring, a move that is reshaping the shape of the consulting interview and, in the same stroke, reviving the case for liberal arts hires whom the firm had spent years deprioritizing. The Financial Times report on AI skills tests in hiring captures a shift that goes deeper than a new interview question: the firm is screening for AI judgment before the case interview, and the gate is sitting on top of the same pipeline that feeds the case interview and the Personal Experience Interview.

What was once a chat-based screening is becoming a final-round component. The pilot is small, the rollout is regional, and the stakes for candidates who do not clear the bar are real.

How McKinsey’s New AI Interview Works

The AI round sits alongside two McKinsey mainstays. Candidates still face the case interview and the Personal Experience Interview. In some U.S. and North America final rounds, a third assessment has been added: a live, time-bound session with Lilli, the firm’s proprietary AI platform. The structure of that session is laid out in how the McKinsey AI interview is structured, based on internal sources reported by interview-prep firm CaseBasix.

Candidates are given a business scenario similar to real consulting work. They prompt the AI, read what it produces, and decide what to keep, adjust, or discard. The interview is not a test of advanced prompt engineering. The expected bar is baseline comfort: can the candidate ask a clear question, iterate when the first answer is thin, and explain the reasoning that links a raw AI output to a structured recommendation.

Four interactions tend to define the session:

• Asking the AI a clear, scenario-specific question.
• Reviewing the response for relevance, accuracy, and gaps.
• Refining or re-prompting when the first answer is incomplete.
• Synthesizing the result into a structured, client-ready answer.

McKinsey has not framed the round as a replacement for the case interview. The career page frames AI use more broadly: “We welcome those who share our curiosity about AI and its potential.” The firm also tells candidates in advance what the round will test, and how to prepare, according to a McKinsey spokesperson cited in coverage of the pilot. The pilot is non-evaluative in its current form, designed to help McKinsey understand future use cases.

The Liberal Arts Reversal

For decades, McKinsey’s hiring funnel was tuned for analytical, business-school profiles. The new AI interview is forcing a reversal, and the candidates the firm had deprioritized are the ones now getting a second look.

On the HBR Ideacast in January, McKinsey Global Managing Partner Bob Sternfels said AI models have developed an expertise in problem-solving that used to be the firm’s main screen. The interview is what opens the door; the AI skill is what gets asked about once inside.

“We will be looking more at liberal arts majors, whom we had deprioritized,” Sternfels said on HBR Ideacast, citing the firm’s need for creativity beyond the “logical next steps” that AI can already produce.

The same Sternfels’s HBR Ideacast interview on agentic work makes the underlying logic explicit. The firm needs people who can impose judgment on AI output, not just people who can produce it. The interview design reflects that. Sternfels has also pointed to Ravi Kumar S, the CEO of IT firm Cognizant Technology Solutions, as a peer recruiter now hiring more liberal arts candidates for the same reason.

The reversal is the hidden pivot of the pilot. The AI round is the new filter, but the candidates being told they now belong in the pipeline are the ones whose value proposition is hardest to score on a spreadsheet: writers, historians, philosophers. McKinsey is betting that human judgment, applied to AI output, is the durable edge the firm is buying.

An Agentic Workforce Reshapes the Job

McKinsey’s hiring experiment is downstream of a deeper shift inside the firm. Eighteen months ago, the firm had roughly 3,000 AI agents supporting its work, and 40,000 employees far outnumbered that fleet. As of January, the count of AI agents has crossed 20,000, a more than 500% expansion, CEO Bob Sternfels said on HBR Ideacast.

That ratio is now closer to one human for every two agents, and it is changing the kind of work consultants are paid to do. Lilli processes more than 500,000 prompts per month across the firm, per the AI interview rollout timeline and BCG, Bain parallels. The volume is one signal of how the agentic fleet is being used; the direction is clear from Sternfels’s own framing on HBR Ideacast. As where the AI winners and losers are stacking up shows, the firms that have spent on agents are now under pressure to translate that spend into new business models.

McKinsey is telegraphing its answer. The firm is migrating away from pure advisory work, away from the fee-for-service model, toward an outcomes-based model where fees are tied to the impact delivered for clients. If that shift is real, the AI interview stops looking like a hiring novelty and starts looking like a hiring signal: the firm needs people who can supervise agents and own the outcome, not just people who can bill hours. Early testers have described the experience as “prompt anxiety,” a term McKinsey used internally when Lilli first launched, and the firm has reported that one hour of prompt training lifted adoption inside its own ranks.

What Other Major Employers Are Doing

McKinsey is the most visible case, but it is not alone. Three patterns are emerging across large employers that want AI fluency baked in from the start.

• Johnson & Johnson has built an internal skills inference system that maps 41 future-ready skills, such as data management and process automation, across its workforce and scores each employee on a zero-to-five scale, per the FT’s look at how J&J and DHL use AI for skills inference.
• DHL runs a “career marketplace” that compares what its staff can do today with what open roles need, then directs people to the right training and managers to the right internal candidate, reducing external hiring costs.
• Bank of America gives employees AI simulations of difficult client conversations, so they can practise, get feedback, and improve before the real call.

The direction in each case is the same: the AI test is moving upstream, into hiring for some firms, into workforce development for others. The candidate side of this trend is visible across the pipeline, as the story of a UMass Lowell grad who built an AI advisor and landed at Nvidia shows. The same competition for AI fluency is showing up in the entry-level race, where the candidates who can show a working AI build have a concrete edge over those who list AI as a skill line on a resume.

The Risks of Hiring on the AI Test

Two cautions sit on top of the celebratory framing. Both come from outside the firms that are running the pilots.

Nick van der Meulen, an MIT scientist who studies technology-driven organisational change, told the FT that AI assessments should be recognised as a rough assessment of skills. Van der Meulen notes that such systems are not 100 per cent accurate, and that problems can arise if employees do not put effort into keeping their digital footprint complete. The output is a starting point for a conversation between a staff member and a manager, not a final score.

Nimmi Patel, head of skills, talent and diversity at Tech UK, the British trade body, is blunter. Patel says high-stakes evaluation and growth decisions are best suited to remain under human supervision, delivered through a hybrid approach that pairs AI inference with human review. The risk in either model is the same: a system that scores a candidate or an employee on inputs the model cannot fully see.

McKinsey has answered the worry with the design of its pilot. The firm has stressed that the AI round is not used in day-to-day performance management at J&J, and that participation in the AI-driven skills platform is optional at both J&J and DHL. Sternfels has framed the next stage of the agentic bet flatly: in another 18 months, every employee at McKinsey will be enabled by one or more agents, he predicted on HBR Ideacast. The firm is hiring for that workforce now.

Frequently Asked Questions

What is McKinsey’s AI interview?

McKinsey’s AI interview is a final-round assessment, currently being piloted in the U.S. and North America, in which candidates are asked to complete a short, live task with Lilli, the firm’s proprietary AI platform. The bar is baseline comfort with prompting, iterating, and applying judgment, not advanced prompt engineering.

How does the McKinsey AI interview fit with the case and PEI?

The AI round sits alongside the case interview and the Personal Experience Interview as a third component in some final rounds. The case and PEI still drive most of the scoring. The AI round is an additional signal, not a replacement, and the round is currently non-evaluative.

Is McKinsey the only consulting firm doing this?

McKinsey is the first major consulting firm confirmed to use an AI tool directly inside a final-round interview. BCG has its own internal AI tool, Deckster, and Bain has Sage, and the consulting prep firm Management Consulted has reported that other firms are expected to follow McKinsey’s lead.

Are employers outside consulting also testing AI skills?

Yes. Johnson & Johnson uses an AI-driven skills inference system built around 41 future-ready skills. DHL runs a career marketplace that matches staff to open roles using AI, and Bank of America uses AI simulations to help employees practise difficult client conversations.

How should a candidate prepare for an AI skills test?

Candidates who know the round is coming should practise prompting, refining, and synthesising on a public AI tool, then rehearse explaining why they used a given prompt and what they did with the output. McKinsey’s own guidance is to treat Lilli as a junior teammate, not an answer engine.