CoachNed
Sign up free

The work

Generative AI consulting is a workflow problem with a language model in it

Firms do not get paid to “do GPT.” They get paid to pick a process, put a model in the right step, and make the errors cheaper than the status quo. That is also what the interview tests.

Ned reviewing a generative AI workflow
Read the 12-min lessonSit the AI case

What clients are actually buying

Search “generative AI consulting” and you will find vendor pages. If you are applying to McKinsey, BCG, or Bain, translate those pages into case language: a value pool, a workflow, data, a method (prompt, RAG, fine-tune, agent), a harness, a pilot.

Typical workstreams: knowledge assistants on a policy corpus, copilots for analysts and adjusters, customer-service drafts with escalation, coding assistants behind a review rule, and (more rarely) agents that are allowed to act. The build vs buy question is real. The first question is still “which step in the process deserves a model.”

  • McKinsey: QuantumBlack plus generalist teams on functional transformations.
  • BCG: BCG X when it is a product; core BCG when it is strategy and adoption.
  • Bain: Vector / AIS when it is digital delivery; generalist Bain when it is the case in the process.

How this shows up in your interview

You are not pitching a consulting firm. You are showing you would not embarrass them in front of a CIO. That means: you can explain an LLM in one sentence, you do not fine-tune to solve a document lookup, and you will not give an agent the payout button on day one.

Independent overview for candidates. Not a vendor directory and not affiliated with the firms named.

Prep in two moves

Learn the simple stack in the AI lesson. Then sit a case where generative AI is the client’s proposed solution and your job is to accept, reshape, or kill it with numbers.

FAQ

Is generative AI consulting a separate job from generalist consulting?

Sometimes (X, QuantumBlack, AIS). Often it is generalists who can talk about the technology without hiding in jargon. Read the job email.

What is the most common interview miss?

Starting with the model. Start with the decision and the $ of a mistake.

Do I need a portfolio of GPTs I built?

Not for a generalist MBB case. A specialist loop may want a take-home. Do not invent a GitHub in the room.

Keep going