Consulting AI toolkit: the junior analyst's stack for 2026
McKinsey now tests graduates on how they use AI

McKinsey now tests graduates on how they use AI.
Not whether. HOW.
In a pilot reported in January, candidates use the firm's own AI tool, Lilli, during a case interview. Interviewers watch how they prompt it, whether they show "curiosity and judgment", and whether they push back on its output (The Irish Times, via the FT).
So if you want to work in consulting, your AI habits are now part of the interview. Here's the stack worth learning, and how to use it without getting burnt.
How AI is changing junior consulting work
The same article has the number everyone quoted: McKinsey boss Bob Sternfels said the firm runs 20,000 AI "agents" alongside 40,000 staff, aiming for one agent per person within about 18 months.
But the most useful evidence for juniors is older and better. A Harvard and BCG experiment with 758 BCG consultants found:
→ on tasks AI handles well, consultants with GPT-4 finished 12.2% more tasks, 25.1% faster, at more than 40% higher quality → below-average performers improved by 43%, top performers by 17% → on a task chosen to sit outside AI's abilities, AI users were 19 percentage points LESS likely to get the right answer
That last line is the whole job now. Knowing when the tool is out of its depth.
And the tools keep moving into the apps you'll live in. In March, Anthropic's Claude for Excel and PowerPoint started sharing context across open files, with ready-made skills for things like model auditing, comps analysis and deck reviews.
What uni won't teach you about consulting with AI
A business degree teaches frameworks. It rarely teaches the boring bit that keeps a firm out of the news.
In 2025 Deloitte agreed to refund part of an A$440,000 report for the Australian government after a researcher found a made-up quote from a judge and references to academic work that doesn't exist. The revised version disclosed that Azure OpenAI was used in writing it (AP).
Nobody marks you down at uni for an unchecked footnote. A client will. (Lawyers learnt this first, as we covered in AI-invented citations in court.)
The junior analyst's AI stack
- Research: a search assistant plus primary sources. Use ChatGPT or similar to map a market fast, then open every source it cites. No link, no fact.
- Reading packs: a source-grounded tool. Load annual reports and interview notes into something like NotebookLM and ask questions against those documents only. Much easier to check.
- Spreadsheets: an AI add-in inside Excel. Good for cleaning data, explaining formulas and auditing a model. Bad at knowing whether your assumptions are sensible.
- Slides: an AI add-in inside PowerPoint. Fine for a first draft of a deck in the firm's template. The storyline is still your job.
- Your own checklist. The cheapest tool in the stack and the one that matters most.
The critical-thinking move: 5 checks before anything goes to a manager
- Is this inside the frontier? Summaries and first drafts, usually yes. Novel judgement on messy data, often no. The BCG study says that's exactly where people get worse with AI.
- Can I trace every number? Each figure on a slide should go back to a cell, and each cell to a source.
- What did the tool assume? Ask it to list its assumptions, then challenge the two that move the answer most.
- Would it survive the client's expert? Pick the weakest claim and try to break it yourself first.
- Did I disclose the AI use? Deloitte had to add that line after the fact. Follow your firm's rules from day one.
What to do this month
- Run a mock case with AI. Take a free practice case, use a chatbot as your "Lilli", and write down where it helped and where you overruled it. That's your interview story.
- Audit one model. Download a public company's financials, build a quick model, then ask an AI add-in to find errors. Count what it missed.
- Build a 5-slide deck from one annual report. Draft with AI, then rewrite the storyline by hand. Compare the two.
- Tighten your applications too. Our AI job search toolkit covers tailoring without sounding like everyone else.
More on the field in our consulting hub.
so TLDR: firms now hire for judgement with AI, not just AI use. Learn the stack, then learn exactly where it breaks.
When an AI tool and your own analysis disagree, what's your first move to work out who's right?
Sources
- McKinsey challenges graduates to use AI chatbot in recruitment overhaulThe Irish Times (Financial Times) · 14 JAN 2026
- Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and QualityHarvard Business School Working Paper 24-013 · 22 SEP 2023
- Advancing Claude for Excel and PowerPointAnthropic · 11 MAR 2026
- Deloitte to partially refund Australian government for report with apparent AI-generated errorsAssociated Press (via News4JAX) · 07 OCT 2025
Get the next issue by email
theVANTAGEd is written by students to bring critical thinking to AI across real industries. One issue every two weeks.