AI agents help consultants across separable jobs: gathering and synthesizing research, and drafting the deck, model, or proposal. The big firms built their own — McKinsey's Lilli, BCG's Deckster. What none of them should own is the recommendation itself: the integrative judgment a client pays for stays human.
The consulting workflow, and which parts an agent can take
Consulting is a pipeline: research the problem, synthesize it into a story, and advise the client on what to do. An AI agent is genuinely useful for the front of that pipeline — pulling sources, summarizing prior work, drafting the slides and the model — because a consultant reviews everything before it reaches a client. The part to keep human is the advice: the read of a specific situation that carries the engagement's value and its risk. So the honest way to test any "AI agent for consulting" pitch is to ask, at each stage, does the agent do the work, or make the call? It's the same automate-the-work-not-the-decision line we drew for recruiters, marketers, accountants, and financial analysts — the professional-services pattern holds here too.
Research — synthesize the inputs, don't skip the reading
The front of an engagement is reading: prior decks, market data, expert calls, the client's own documents. This is where the big firms pointed their first agents. McKinsey rolled out Lilli, an internal assistant that pulls from the firm's own body of past work and proprietary research so a consultant can synthesize findings without re-reading every source by hand. This is the safest stage to automate because the output is checkable — every claim traces back to a document you can open. The Harvard–BCG field experiment found that on tasks inside AI's capability frontier, 758 BCG consultants using GPT-4 completed 12.2% more tasks, finished 25.1% faster, and produced results rated over 40% higher quality than a control group. Research synthesis is squarely inside that frontier — as long as a person still verifies the sources the agent leaned on.
Draft — build the deck and the model, keep your name on the logic
Consultants live in slides, and drafting them is the second job agents took over inside the big firms. BCG built Deckster, a slide assistant trained on the firm's templates that drafts and grades presentations, alongside GENE, a GPT-4o-based partner for brainstorming and content. At the enterprise end, Deloitte launched Zora AI, an agentic platform of ready-to-deploy agents for finance, HR, and other back-office functions — closer to a digital workforce than a drafting aid. The individual consultant's version is smaller but the discipline is identical: let the agent turn your research into a first-draft deck or model, then own the logic and the numbers before anyone sees them. It's the same draft-vs-publish split we mapped for content repurposing — automate the draft, keep a hand on what ships.
Advise — the recommendation stays human
Here the same Harvard–BCG study draws the line for you. The researchers deliberately included a task engineered to sit outside AI's frontier — one that required weighing conflicting evidence and reaching a nuanced business judgment. On that task, consultants using GPT-4 were more likely to reach the wrong conclusion than consultants working without it: confident, well-formatted, and incorrect. That's the recommendation stage. An agent that synthesizes cleanly can still lead you to a plausible answer that's wrong for the client in front of you, and it has no stake in being right. The deliverable a client actually buys — "here is what you should do, and why, given your specific situation" — is the judgment that sits outside the frontier and stays with the consultant.
Which agent should a consultant start with?
Start where your engagement actually stalls. If you drown in reading before every project, begin at the research layer — a synthesis agent that summarizes sources into a checkable brief, built the desk-level way, not the McKinsey way. The reproducible pattern is the same one behind Issue #001's Gmail triage agent: a bounded task, a checkable output, no platform team required — the headline-vs-desk-level split that separates Lilli and Deckster (uncopyable) from an agent you can stand up this week. If client-meeting prep eats your mornings, the routes in AI agent for meeting prep apply directly. New to agents? Begin at Agent 101, or pick a builder from the no-code AI agent tools guide. Whatever you start with, hold the line: automate the research and the draft, own the advice.
FAQ
What AI agents can help consultants? Match the agent to the stage. For research, a synthesis agent — the desk-level cousin of McKinsey's Lilli — summarizes sources into a checkable brief. For drafting, an agent turns that research into a first-draft deck or model, the way BCG's Deckster drafts slides. The recommendation itself stays with you. Start with whichever stage costs you the most time.
Can an AI agent write my consulting recommendation? It can draft the deck around it, but the recommendation should stay human. The Harvard–BCG field experiment found that on a task requiring integrative business judgment — the shape of a real recommendation — consultants using GPT-4 were more likely to reach the wrong conclusion than those without it. AI is strongest on the research and drafting inside its frontier, and weakest on exactly the call a client pays for.
Do McKinsey and BCG really use AI agents internally? Yes. McKinsey rolled out Lilli as an internal assistant grounded in its own research, and BCG built Deckster and GENE for slides and brainstorming; Deloitte launched Zora AI, a platform of deployable agents. These are internal, enterprise tools — study the pattern, but reproduce it at desk scale.
Do I need to be a big firm to use consulting AI agents? No — and that's the point. Lilli and Deckster run on infrastructure an independent consultant doesn't have, but the pattern (research synthesis, deck drafting, human judgment) is reproducible with tools you can run yourself. See the headline-vs-desk-level examples and the no-code tools guide to build the desk-level version.
Where should a solo or boutique consultant begin? With the stage that eats the most hours — usually the reading. Start with a research-synthesis agent that turns sources into a brief you verify, then add deck drafting. The meeting-prep and content-repurposing walkthroughs show the bounded-task, checkable-output setup on a single desk.
One real AI agent a week, straight to your inbox. Free, no upsell.