In most first rounds I have scored, the best arithmetic in the building belonged to an engineer. The offer usually went to someone else. The gap was not communication, and it was not "business sense", whatever that is. It was order of operations.
Engineers run the scientific method forward: observe, measure, conclude. Saying the answer before the data is in feels like malpractice, because in a lab it is. A case interview scores the same method run backward: state the answer you would give if the interview ended now, then design the cheapest test that could prove it wrong. The interviewer is not waiting for your conclusion at minute 24. They are scoring whether you had one at minute 2 and what you did to it afterward. The same goes for physicists, chemists, and statisticians trained to put data first.
By the end you will be able to open a case with a stated hypothesis, decide how many digits a decision needs, and close with a recommendation a partner can act on. The engineering stays. Only the order changes.
What the firms say they want, and what they plan to teach
ABET, which accredits US engineering programs, defines engineering design as "an iterative, creative, decision-making process" involving "generating multiple solutions, evaluating solutions against requirements, considering risks, and making trade-offs" under constraints (ABET criteria, 2026–2027). That is also a fair description of a 25-minute case.
None of the big three asks for a business degree. McKinsey's advanced professional degree page recruits PhDs, MDs, and JDs and tells them they will "gain the business, leadership, and problem-solving skills" once inside. BCG's advanced degree candidates page offers "customized training programs for non-business background consultants". Bain's advanced degree page values "a diverse range of academic backgrounds in our consulting roles". The US Bureau of Labor Statistics lists a bachelor's degree as the typical entry-level education for management analysts, median pay $101,190 as of May 2024 (BLS Occupational Outlook Handbook).
So business knowledge is not the test; the firms will teach it. The test is what training cannot install: committing to a view on thin evidence, testing it, and changing it without sulking. That is exactly what engineers withhold, because training taught them to.
Nor are you rare. In MIT's 2025 graduating student survey, 12.4% of bachelor's graduates with an accepted job (185 respondents) and 23.4% of master's graduates were heading into professional, scientific, and technical services, the category that includes consulting (MIT Institutional Research). The interviewer has seen your profile often, and an engineer who leads with the answer rarely.
The scoring sheet only scores what you commit to
Firms do not publish their scoring sheets, and anyone quoting exact weights is guessing. Every sheet I have used was a handful of boxes, each filled from evidence that happens out loud. Mine has five.
| Box | What the interviewer needs to hear |
|---|---|
| Structure | A decomposition a team could be handed |
| Analysis | Quick, correct arithmetic with units |
| Judgment | A view, with reasons, on what the numbers mean for the decision |
| Synthesis | Recommendation, two reasons, one risk, one next step |
| Communication | Answer first, then support |
Structure and analysis are where engineers collect. Judgment is where they leak, because it is scored from how many times you took a position and what you did when the data pushed back. One view at minute 24 is a data point; three views, two of them revised, is a pattern. The sheet rewards the pattern, not the final view.
This is why "communicate better" is useless advice for engineers. Your sentences are fine; they arrive in the wrong order. Barbara Minto's The Minto Pyramid Principle, which her site calls a standard text at the major consulting firms, says the same of writing: answer first, then support (Barbara Minto). A lab report is the exact inverse. You have written dozens. That is the habit to break.
Same case, two clocks
The two runs against a 25-minute clock, on a case about automating a manual production step:
| Minute | Forward run (how engineers are trained) | Backward run (what the sheet rewards) |
|---|---|---|
| 0 to 2 | Questions on process, cycle times, defect rates | Two questions: what "worth it" means to the client, and the volume outlook |
| 2 to 4 | A structure listing every cost and benefit of automation | "Hypothesis: automation only pays if volume grows. I need today's labor cost, the cell's all-in cost, and the forecast." |
| 4 to 12 | Requests process data, builds a cost model, checks it | Gets the three numbers, does the payback, says which side of the line the client is on |
| 12 to 18 | Requests more data to tighten the model | Finds the one input that could flip the decision and spends the time there |
| 18 to 24 | Presents the model and its uncertainties | Revises the hypothesis once, says so, asks about the order book |
| 24 to 25 | "So, to summarize what we've found so far..." Time is called | "Do not automate yet. Two reasons, one risk, one thing to check next week." |
The forward run is not stupid; with six weeks it is roughly right. In the room it produces a model with no owner and a recommendation the clock deleted.
A worked example: Kestrel Composites
Kestrel Composites and its numbers are invented; the arithmetic is the point. Kestrel makes 20,000 carbon-fiber bicycle frames a year. Layup is manual: 25 technicians at a fully loaded $56,000 each, so $1.4 million a year. A supplier has quoted an automated layup cell at $4.8 million; it needs 5 operators plus $220,000 a year in maintenance and consumables, roughly $500,000 a year to run, with capacity for 36,000 frames. The CFO wants capital projects to pay back inside 3.5 years. Should Kestrel buy the cell?
Minute 2, the hypothesis. "At today's volume the cell probably does not pay back fast enough, so the case turns on whether volume grows. Let me test that with the payback."
Payback today. Cash saving is $1.4m minus $0.5m, or $0.9m a year. Payback is $4.8m divided by $0.9m, about 5.3 years. Well outside 3.5. The hypothesis survives.
The flip point. A 3.5-year payback needs a saving of $4.8m divided by 3.5, about $1.37m a year. Add the cell's $0.5m running cost and manual labor would have to cost about $1.87m, which at $56,000 each is 33 to 34 technicians. Each technician handles 800 frames a year (20,000 divided by 25), so 34 technicians is about 27,000 frames. The decision flips at roughly 27,000 frames a year, 35% above today.
Minute 12, the revision. "Automation is a bet on volume, not on the robot. If the order book supports 27,000 frames within two years, buy the cell. If not, wait. What does the forecast look like?"
Notice what was never computed: the cell's cycle time, uptime, and scrap rate, which the forward run models first. Capacity of 36,000 clears any plausible forecast, and a 10% error in running cost moves the flip point by roughly 700 frames. The input that deserves precision is frames per technician: at 900 rather than 800, the flip point is about 30,600 frames, and the recommendation might change. That is where the next question goes.
Ned's rule. Say the recommendation you would give if the case ended now, then spend the rest of the time trying to break it. If nothing breaks it, you finished early. If something does, you found the insight, and the interviewer watched you find it.
Decision-grade precision, not lab-grade
Engineers are trained to report uncertainty honestly, which in a case becomes reporting it endlessly. The fix is a precision budget: spend precision on inputs that can move the answer across the decision line, and nowhere else. Three silent steps. Find the line (payback hurdle, breakeven volume, margin target) and name it out loud. Measure the distance from your estimate to the line; for Kestrel, 5.3 years against 3.5 is a long way. Then compare that distance with plausible input error: if the error cannot close it, stop calculating and state the conclusion; if it can, that input is the case, so ask about it.
| Kestrel input | Plausible range | Moves the decision? | Precision to spend |
|---|---|---|---|
| Cell capital cost | $4.3m to $5.3m | No: payback stays 4.8 to 5.9 years today | None; quote it as given |
| Cell running cost | $0.45m to $0.55m | Barely: about 700 frames either way | One sanity check |
| Frames per technician | 700 to 900 | Yes: 24,000 to 30,600 | Ask how the 800 was measured |
| Volume forecast | Unknown | Yes: it is the decision | Most of the remaining time |
Round early and say so: "call it $0.9m a year" is a consultant's sentence; "$0.9 million, or more precisely $896,000" is a lab notebook's. Do it without a machine. Bain's own digital assessment page says calculators are allowed only in the online assessment, "not during subsequent interviews", and I know of no major firm that permits one in a live case. If your last mental arithmetic happened in a MATLAB console, a week of quick math will fix it.
Where the engineering scores, and where it leaks
Do not hide the engineering. Point it at the boxes that pay.
| Habit you already have | Box it fills | How it sounds in the room |
|---|---|---|
| Dimensional analysis | Structure | "Profit is volume times price minus unit cost, less fixed cost. I'll take those in turn." Units keep the tree honest. |
| Bottleneck thinking | Judgment | "The line runs one shift, so the constraint is labor, not machines. Capex is the wrong lever until the second shift is full." |
| Failure-mode thinking | Synthesis | "The main risk is that 800 frames per technician was a good year. I'd check last year's payroll before signing." |
Three habits leak points, into silence rather than wrong answers: deriving from first principles when you could state the equation and plug in; tool talk, since nobody is scoring whether you would build this in Python; and exhaustive enumeration, since a tree with eleven branches reads as an inability to prioritize. BCG's own case interview preparation page lists exactly one item under "Don't": "Overcomplicate Solutions". The same page says what matters most is "how you approach the problem and the quality of your reasoning". Reasoning you never voice cannot be scored.
The behavioral half has the same bug
Engineers tell behavioral stories the way they write methods sections: the system in loving detail, the decision in one clause at the end. The interviewer scores the decision: what you chose, who disagreed, what you did about it, what it cost you.
Run the same reversal. Open with the decision and the outcome in one sentence ("I pulled the launch by three weeks over the product lead's objections, and here is why"), then supply context as it is asked for. Put a person in every story who wanted something different from you; without one it is a project summary, and the case already covered that. The behavioral simulator asks four follow-ups per story, roughly where engineers discover theirs had no people in it.
Practice this today
Take one case, any case. At minute two, before you have a single number, say aloud the recommendation you would give if the case ended now, and write it on the top line of your page. Run the case, then compare the top line with your actual recommendation. If they match, you know what the following 20 minutes were for: breaking a hypothesis that held. If they differ, you can name the exact number that changed your mind, which is the sentence interviewers most want.
Do it against a clock and out loud; engineers do this correctly in their heads and never say it. The live voice case with Ned runs about 18 minutes and debriefs on seven scores, judgment and synthesis among them. With five minutes, the first rep is three typed turns with instant scores, no account needed. Repeat the structure drill until a hypothesis in your opening feels normal rather than reckless.
Everything on CoachNed is open for seven days with no card; after that it is $120 for a recruiting season or $49 a month. CoachNed is independent and not affiliated with McKinsey, BCG, Bain, ABET, or any other organization named here.
Frequently asked questions
Do consulting firms hire engineers without an MBA?
Yes. McKinsey, BCG, and Bain each run recruiting tracks for advanced-degree candidates, and BCG says it hires "across all advanced degree backgrounds and concentrations". The BLS lists a bachelor's degree as the typical entry-level education for management analysts.
Do engineers get operations or manufacturing cases?
Not by design, as far as any firm has said publicly. Cases come from one pool, so expect profitability, pricing, and market entry at least as often as anything with a factory in it. Practicing only plant cases is how engineers over-fit.
Should I state a hypothesis if I might be wrong?
Yes. A hypothesis stated at minute two and revised at minute twelve scores higher than a correct conclusion at minute twenty-four, because the interviewer saw your judgment work twice. Being wrong early and saying so is not penalized; being silent is.
Sources
- Criteria for Accrediting Engineering Programs, 2026–2027 — ABET — definition of engineering design. Checked 2026-09-24.
- Advanced professional degree — McKinsey careers — APD recruiting and on-the-job business training. Checked 2026-09-24.
- Advanced Degree Candidates — BCG Careers — eligible degrees and non-business-background training. Checked 2026-09-24.
- Case Interview Preparation — BCG Careers — BCG's case description and its single "Don't". Checked 2026-09-24.
- Advanced degree career opportunities — Bain & Company — Bain on academic backgrounds. Checked 2026-09-24.
- Digital assessment — Bain & Company — calculator rules for assessment and interviews. Checked 2026-09-24.
- Management Analysts — BLS Occupational Outlook Handbook — entry-level education and median pay, May 2024. Checked 2026-09-24.
- Student Placement — MIT Institutional Research — 2025 Graduating Student Survey industry breakdown. Checked 2026-09-24.
- Barbara Minto — The Minto Pyramid Principle — the answer-first method. Checked 2026-09-24.