- **Title:** Consulting Case Interview Guide 2026: Prompt → Framework → Recommendation - **Canonical:** https://ip.adatepe.dev/blog/consulting-case-interview-guide - **Author:** Isabella Moretti - **Category:** Consulting - **Published:** 2026-04-21 - **Read time:** 16 min read - **Tags:** Consulting, Case Interview, McKinsey, BCG, Bain, MECE, Market Sizing # Consulting Case Interview Guide 2026: Prompt → Framework → Recommendation *Consulting Guide · Updated April 2026 · Reviewed by a former McKinsey Engagement Manager and ex-BCG Senior Associate (11 years combined, London / Munich / New York)* Consulting case interviews are theatre with a rubric. The interviewer has a prompt, a structure they expect you to impose on it, a number they expect you to land within a factor of two, and a recommendation they expect you to deliver in under 90 seconds. The 60 minutes in between is scored on structure, math fluency, hypothesis-driven thinking, and the five follow-ups you don't get to see. This guide walks one profit-decline case end-to-end with the rubric scoring shown at every step. It then covers the two other dominant case types — market-entry and market sizing — with their distinct starting questions and red flags. Pair this with our [McKinsey PEI guide](/blog/mckinsey-pei-personal-experience-interview) for the behavioral half of the loop, and the [behavioral interview guide](/tips) for the STAR mechanics consulting candidates also need. ## The shape of a case interview A standard first-round case runs 30–45 minutes. A second-round case runs 45–60. The structure is consistent regardless of firm: 1. **Prompt** (30–60 seconds). The interviewer reads the case prompt and may hand you a page of data. 2. **Clarifying questions** (2–3 minutes). You ask two to four focused questions. 3. **Silent structuring** (60–90 seconds). You ask for a moment, sketch an issue tree on paper. 4. **Lay out your structure** (90–120 seconds). You walk the interviewer through your tree, top-down, before diving in. 5. **Drive the analysis** (20–35 minutes). You work through branches of the tree, pulling specific data points from the interviewer as needed, running math out loud, forming and testing hypotheses. 6. **Synthesis** (2–3 minutes). You summarise findings, give a recommendation, and name two to three risks or caveats. 7. **Follow-ups** (5–10 minutes). The interviewer probes one or two parts of your analysis, sometimes with new data that changes the answer. Two things distinguish a strong case from a mediocre one: **driving the case** (you, not the interviewer, naming what to look at next) and **landing the recommendation** (direct answer first, reasoning second). ## Clarifying questions — what to ask, what not to Candidates often under-ask or over-ask clarifying questions. The rubric scores three or four well-chosen questions, not ten, and not none. What to ask: - **Business model confirmation.** "When you say 'the client sells subscription software to mid-market B2B customers', is this a new product line or their existing one?" - **Time horizon.** "Are we looking to reverse the decline in one quarter, one year, or three years?" - **Success metric.** "Is the client's goal profit, margin, revenue, or market share?" - **Scope.** "Are we evaluating the full product line or a specific segment?" What not to ask: - **Questions with inferable answers.** Don't ask "what industry is this?" if the prompt said it. - **Fishing expeditions.** "What's the management team like?" reads as directionless. - **Questions that reveal you don't have a structure yet.** Asking "should I use a profit tree or a market-entry framework?" out loud is a downgrade. The candidate who asks three targeted questions in 90 seconds, reflects them back to confirm understanding, and then says "I'll take a moment to structure" outperforms the candidate who rushes to a framework or asks eight surface questions. ## Issue tree / MECE construction The issue tree is the central artifact of the case. It's also where most candidates lose points. ### What makes a tree MECE MECE — Mutually Exclusive, Collectively Exhaustive — means your tree's branches don't overlap and together cover the problem. For a profit decline: - **Profit = Revenue − Costs.** Two branches, no overlap, together the whole profit. - **Revenue = Price × Volume.** Two branches, no overlap. - **Volume = Customers × Units per customer.** Two branches. Or equivalently: existing-customer volume + new-customer volume. Pick one split, not both. - **Costs = Fixed + Variable.** Or: Cost of Goods Sold + Operating Costs. Pick one split and be consistent down the tree. The trap: candidates mix bases of decomposition. A tree that has "Revenue → Price × Volume" and then "Costs → by business unit" is not MECE and scores poorly. ### The case-specific overlay Pure financial trees are not enough. After the financial decomposition, overlay the case-specific variables. For a SaaS company with a subscription decline, the tree might be: - **Revenue** - Price: contract value per customer segment - Volume: - New customer acquisition (by channel) - Existing customer retention (by segment) - Existing customer expansion (by product) - **Costs** - COGS: infrastructure, support - Operating: sales, marketing, R&D The overlay is where the case's teeth are. A client with a subscription decline has a story hiding in retention or expansion, not in price or COGS; the tree should foreground that. ### Walk the tree top-down Once drawn, present the tree top-down in 90–120 seconds. "I'd think about this problem as profit = revenue minus costs. On revenue, the drivers are price and volume; volume decomposes into new customer acquisition, existing retention, and existing expansion. On costs, the split is COGS and operating, with operating further broken down by function. Given the context — a subscription-software client — I'd start with retention and expansion on the revenue side, because those tend to be the more sensitive levers for SaaS profit." The interviewer should nod and either validate or adjust. "Good structure. Start with retention." ## Running the math out loud Once you're in a branch, the interviewer will hand you numbers. Running the math out loud is scored on three things: arithmetic fluency, unit discipline, and hypothesis formation. ### Worked example: SaaS profit-decline case *Prompt:* "A mid-market SaaS client has seen profit decline from $40M to $28M over two years despite revenue holding flat at $200M. The CEO wants to understand what's driving the decline and whether to cut costs or invest in growth. Walk me through how you'd approach this." *Clarifying questions:* "When you say profit, are we talking operating profit, net profit, or EBITDA? [Operating.] Has revenue actually held flat, or are net revenue and gross revenue diverging — perhaps because discounts are growing? [Gross revenue flat. Net revenue I'll give you later.] And is the $200M across all product lines? [Yes, single product line.]" *Structure:* "I'd start with a profit tree: profit = revenue − costs. Revenue is flat, so either the cost side is expanding, or revenue has shifted composition (discount-heavy mix) without moving the top line. I'd ask about cost-side growth first, because the math there is often simpler, then pivot to revenue composition if costs don't fully explain the gap." *Interviewer hands data:* "Here's the cost breakdown. COGS: $60M two years ago, $65M today. S&M: $60M two years ago, $80M today. R&D: $30M two years ago, $35M today. G&A: $10M two years ago, $12M today." *Math out loud:* "Let me compute the totals. Two years ago: 60 + 60 + 30 + 10 = 160. Revenue 200 minus 160 is 40 — matches. Today: 65 + 80 + 35 + 12 = 192. Revenue 200 minus 192 is 8 — but you said profit is 28, not 8. So I'm missing something. Either the revenue number I'm using is wrong — possibly net revenue is lower — or the cost breakdown doesn't sum to total costs." *Interviewer:* "Good catch. Net revenue today is $220M; gross revenue before discounts was $240M. Two years ago net and gross were both $200M." *Updated math:* "Then today's picture is: net revenue $220M minus costs $192M = $28M profit. Confirmed. So revenue has actually grown, but costs have grown faster. S&M has grown 33% — from $60M to $80M — while revenue only grew 10%. S&M efficiency has collapsed. The CEO's question about cutting costs vs investing in growth has an answer: the efficiency issue is inside S&M, not a broad cost problem." *Hypothesis:* "My working hypothesis is that the client is overspending on acquisition in a way that isn't paying back. I'd want to check two things: customer acquisition cost (CAC) trend and lifetime value (LTV) trend. If CAC has grown faster than LTV, that's a unit economics problem that reducing S&M will fix faster than cutting COGS. If LTV has shrunk — maybe retention has dropped — that's a product problem and S&M cuts won't help." The math was driven, the hypothesis was formed before asking for more data, and the next question is falsifiable. This is what scores. ## Synthesis and the recommendation The synthesis is the most under-practiced part of a case. Candidates run great math and then fumble the recommendation. The shape that lands: - **Lead with the answer.** First sentence: the recommendation. - **Two or three supporting reasons.** Each one sentence. - **Two risks or caveats.** What could flip the answer, and what you'd want to check next. - **Delivered in 60–90 seconds, standing up (metaphorically). No hedging.** ### Synthesis for the SaaS case "My recommendation is that the client should cut S&M spending by roughly 20% and reinvest those savings into retention-focused product investments. Three reasons. First, S&M spend has grown 33% while net revenue grew only 10%, indicating collapsing acquisition efficiency. Second, the cost problem is localised to S&M — COGS, R&D, and G&A are all growing in line with or slower than revenue, so a broad cost-cut is not warranted. Third, the persistence of the profit gap suggests the efficiency issue is structural, not a transient marketing-spend spike. Two caveats: I'd want to see LTV and CAC trends split by customer segment before committing to the 20% number — if the efficiency decline is concentrated in one channel, the cut should be targeted, not across-the-board. And I'd want to understand the retention picture, because if product stickiness is declining, S&M cuts won't fix the underlying issue." That's a scored recommendation. Direct, specific, with risks. ## Three case types: profit, market-entry, sizing Almost every consulting case collapses to one of three types. Each has a distinct starting question and distinct red flags. ### Profit *Starting question:* "Is this a revenue problem, a cost problem, or a mix problem?" *Red flag:* Jumping to a generic profit-tree without asking about time horizon or business model first. Profit trees are not interchangeable — a SaaS profit tree differs from a retail profit tree. *Follow-up mechanism:* The interviewer usually provides new data mid-case that forces a hypothesis revision. Candidates who cling to the original hypothesis even after contradicting data appears score poorly. ### Market-entry *Starting question:* "What's the size of the opportunity, who's already in the market, what would we bring that they don't, and can we profitably serve it?" *Red flag:* Treating market-entry as a pure sizing exercise. The attractive-market-size answer is a junior-consultant answer. A scored market-entry case names the competitive moat the client would have, the go-to-market path, and the conditions under which the firm would NOT recommend entry. *Framework overlay:* Market size × market attractiveness × competitive position × capability fit × expected profit. ### Sizing *Starting question:* "What's the right base — population, households, businesses, transactions — to start from?" *Red flag:* Opaque assumptions. "I'll assume 10% of customers do this" without naming why 10% and what would change it. *Framework overlay:* Always name your base, your adjustment factors with reasons, and one sanity check that uses a different base to see if the two approaches converge. | Case type | First question | Red flag | | ------------- | ----------------------------------- | --------------------------------------- | | Profit | Revenue, cost, or mix problem? | Generic tree, no business-model overlay | | Market-entry | Size × attractiveness × position? | Treating as sizing exercise only | | Sizing | What's the right base to start? | Opaque adjustment factors | ## The "what did we miss" follow-up Every case ends with some version of "what could change your recommendation?" This is not a rhetorical question. The interviewer expects you to name two or three specific things: 1. A data point you didn't get access to that would move the answer. 2. A time-horizon sensitivity (if the recommendation is three-year, what changes at one-year?). 3. A strategic alternative you didn't pursue (what if the client's goal were market share, not profit?). Candidates who say "I feel confident in my recommendation" score below candidates who name a specific flaw in their own reasoning. Consulting selects for intellectual humility on this question specifically. ## Frequently Asked Questions ### How is a McKinsey case different from a BCG or Bain case? All three use the same case structure (prompt → framework → math → synthesis). Differences are tonal: McKinsey tends to be more interviewer-led (you drive less, the interviewer probes more). BCG tends to be more candidate-led (you drive the whole case; the interviewer watches). Bain sits between the two with a slight BCG lean. Same rubric, different steering. ### Should I use a named framework like Porter's Five Forces? Rarely. Named frameworks off-the-shelf read as memorised. Interviewers prefer a custom issue tree built from the case's specifics. Use frameworks as mental scaffolding, not as the structure you present. ### How much math do I need to be fluent at? Addition, subtraction, multiplication, and division on two-to-three-digit numbers, plus percentages and ratios, delivered out loud without a calculator. Most cases don't need fancier math. Drill mental math until you can do 15% of 340 or 280 × 1.4 in your head in under 10 seconds. ### How long should my issue tree be? Two levels of depth is usually right; three is occasionally needed for complex cases. A tree that fills a whole page is usually too ambitious for 60 minutes. Prune branches that aren't likely to matter — a tree that says "every possible driver is on the table" reads as directionless. ### What do I do if I get stuck mid-case? Say so out loud and reset. "Let me pause. I've been working on retention for a few minutes but I haven't found evidence that it's the driver. Let me step back and check my tree — I think I should test the pricing branch next, because …" Interviewers reward visible recalibration over silent flailing. ### Can I prepare cases on my own? Up to a point. Solo practice builds math fluency and structure muscle. Mock cases with another person (ideally someone who has case-interviewed before) are where the case-driving and recommendation-landing skills get built. A minimum of 20–30 live mock cases is common before a McKinsey or BCG onsite. ## Keep reading - [McKinsey PEI 2026: Personal Impact, Entrepreneurial Drive, Courageous Change](/blog/mckinsey-pei-personal-experience-interview) — the behavioral half of the loop - [The Behavioral Interview Guide: STAR, Stories, and How to Actually Win](/tips) - [Product Manager Interview Guide 2026](/blog/product-manager-interview-guide) — product sense shares DNA with case structure - [Data Scientist Interview Guide 2026](/blog/data-scientist-interview-guide) — ambiguous-problem framing adjacent to case reasoning Ready to drill cases with scored feedback on structure, math, and synthesis? [Start a free trial](/pricing) — consulting-preset cases across profit, market-entry, and sizing, with issue-tree scoring and synthesis timing.