AI cases
The AI case interview is still a case
Same clock, same close. The new layer is whether you know how the model actually works — and what a wrong answer costs in dollars.

What an AI case is
An AI case interview is a standard consulting case with a technology layer you cannot hand-wave. The client still needs a decision: invest, pilot, kill, or sequence. The interviewer now also wants to hear whether you can tell a chatbot from an agent, RAG from fine-tuning, and a demo score from the cost of a missed fraud claim.
This showed up first in specialist loops — McKinsey QuantumBlack, BCG X, Bain Vector. It is now leaking into generalist first rounds. You will still open with the objective. You will still do the math out loud. You will not get a pass for naming a model.
- If the prompt is a metric that moved, start like a diagnosis case, then ask what data and what model would even apply.
- If the prompt is “should we buy gen AI,” size the value pool before you pick a method.
- If the prompt is a model that already exists, read the error table before you recommend scale.
How the room is scored
Interviewers still grade structure, math, exhibits, judgment, and the close. The AI overlay is a judgment test: did you pick the cheapest method that works, and did you price the misses?
- Structure: objective, value, data/workflow, method, harness, pilot.
- Math: time saved minus extra leakage minus run-cost. Build cost in year one.
- Exhibits: precision/recall and hallucination rates are charts. Treat them like a P&L.
- Close: yes/no or a sequence, one number, a kill criterion, Monday next step.
CoachNed’s AI module is one 12-minute lesson and one original case, Clearclaim Mutual. Independent practice. Not a McKinsey, BCG, or Bain product.
Prep that actually transfers
Do not cram architecture diagrams. Learn the five-word map (AI, ML, gen AI, LLM, agent), the error-cost habit, and RAG vs fine-tune. Then sit a case that forces those words into a recommendation.
Specialist data-science loops still exist. Use the data science case guide for experiment design and model metrics. Use this module when you are a generalist who now has to sound literate.
FAQ
Is an AI case interview the same as a data science case?
No. A data science case often wants experiment design, metrics, and a model spec. An AI case in a generalist loop wants a business decision plus enough ML fluency to not recommend an unbounded agent with write access.
Do I need to code?
Not in a generalist McKinsey, BCG, or Bain case. You need to know what training vs inference means, what RAG is for, and how you would know the pilot worked.
How should I practice?
Read the 12-minute lesson, then sit Clearclaim Mutual out loud. The lesson is free. The scored case uses your case credit, same as any other CoachNed case.
Keep going
AI interview module
One lesson, one live case. The new generalist bar at MBB.
McKinsey AI interview
What generalist McKinsey rooms now ask, versus QuantumBlack.
RAG vs fine-tuning
The decision framework interviewers want, in one page.
LLM interview questions
Questions MBB generalists actually get, with the short answers.