First principles in a case interview is not a brand of cleverness. It is the habit of reducing the client's decision to units, variable costs, and physical constraints, then rebuilding the recommendation from numbers you can ask for. Candidates who treat it as a philosophy speech, or who open every novel prompt with Company–Customer–Competitor, both lose the same points: the interviewer never sees a testable driver. After this piece you should be able to decide when unit decomposition beats a memorized spine, run one yourself on an operations case, and stop decomposing the moment the next layer no longer changes the ask.
Bain says its case interviews reward thinking in a logical, structured way, with precision and creativity and no single "right" answer. McKinsey describes the problem-solving interview as watching how you structure an ambiguous challenge, pick the issues that matter, and deal with facts and data. Neither firm is scoring whether you can name a mental model. Both are scoring whether your structure bottoms out somewhere real.
What "first principles" actually means on the scoring sheet
Strip the jargon. In the room, first principles means three moves in sequence.
- Name the unit of the business. Not "the market," not "the customer journey." The thing that generates revenue or burns cash: a tempered square meter, a delivery stop, a patient visit, a flight segment.
- Write the unit's economics as an equation. Price minus the costs that scale with that unit, plus the constraints that cap how many units you can sell (machine hours, warehouse cube, regulatory storage limits, crew legality).
- Ask for the data that would change the recommendation. If a branch cannot become a chart, a rate, or a yes/no fact in the next two minutes, it is decoration.
That is the whole method. Analogy thinking says "glass plants usually compete on service and brand." First-principles thinking says "contribution per good m² equals price minus blank, energy, labor, and scrap, subject to furnace throughput." One of those sentences leads to a data request. The other leads to a tour of the 3Cs.
When the memorized framework is the wrong first move
Standard spines are fine when the prompt maps cleanly onto them. Profit down with no other signal: revenue and cost is enough to open. Classic market entry with a clear geography and product: attractiveness, capability, and economics will get you started. I do not punish candidates for using a simple tree when the prompt is simple.
I do stop listening when the prompt is about physics or process and the candidate still opens with Company, Customer, Competitor. Typical tells:
- Margin collapsed after a process change, a supplier switch, or a capacity expansion.
- The ask is utilization, yield, scrap, cycle time, or "why did unit cost jump."
- The industry has a hard constraint the client cannot negotiate away (kilns, cold chain, runway slots, licensed beds).
In those rooms, a memorized marketing or strategy label wastes the first three minutes. You sound polished. You have not yet named the unit.
Where first principles wastes time
The method has a failure mode, and I see it often enough to name it.
Over-decomposition. You do not need atomic theory to price industrial glass. Two to three levels is enough for a 25-minute case: objective → unit equation → the one or two drivers that move the answer. Going further into "why is aluminum expensive" or "what is the thermodynamics of tempering" burns the clock and signals poor judgment.
Philosophy theater. Quoting physicists or founders does not score. If your opening sounds like a podcast intro, the interviewer is waiting for a branch list.
False rigor. Inventing twelve MECE-looking buckets that never become numbers is not rigor. It is a longer version of topic listing.
Ignoring a clean profit tree. If the prompt is "profits fell 20 percent last year" with no process hook, opening with revenue and cost is first principles enough. Forcing a custom unit model when the interviewer has a standard exhibit ready makes you look inflexible.
Ned's rule. Decompose until the next layer would change what you ask for next. Then stop and ask. Depth without a data request is theater.
A decision table before you speak
| Prompt signal | Default spine | First-principles override |
|---|---|---|
| Profit down, no process detail | Revenue / cost | Stay with the simple tree |
| "Unit cost rose" or "yield fell" | Often still revenue / cost | Open with the unit cost equation |
| Capacity, scrap, cycle time, energy | Generic ops buckets | Units × constraint × cost per unit |
| Enter a new segment | Market entry spine | Build unit economics of the new segment first |
| Pricing power with thin margins | 4Ps or value pricing | Contribution waterfall before any price change |
Use the right column only when the left column would miss the constraint. Otherwise keep it simple.
Worked example: Northline Tempering
Prompt (invented client, invented numbers): Northline Tempering fabricates tempered architectural glass for commercial contractors. Contribution margin per square meter fell from about 20 percent to under 12 percent in two quarters. Leadership wants a "pricing and positioning" review. How would you approach it?
The memorized answer (weak)
Candidate opens with Product, Price, Place, Promotion, or with Company–Customer–Competitor. Asks for brand perception, competitor list prices, and whether the sales team is "aligned." Fifteen minutes later you still do not know whether the plant is losing money on each good square meter or simply selling fewer of them. The interviewer has an exhibit on scrap rates sitting in the folder. You never earn it.
The unit answer (what scores)
Step 1 — Name the unit. Northline sells tempered glass by the square meter of good product that ships. Scrap that never leaves the plant is not revenue; it is burned cost.
Step 2 — Write the unit equation.
Contribution per good m² = Price − (Blank + Energy + Direct labor) × (1 / (1 − Scrap rate))
Or, equivalently, allocate scrap as a burden on good output. Same math.
Step 3 — Put numbers on it. The interviewer confirms:
| Input | Value |
|---|---|
| Average selling price | $38 per good m² |
| Float-glass blank | $22 per m² processed |
| Tempering energy | 20 kWh per m² processed |
| Loaded labor | $4.00 per m² processed |
| Current scrap rate | 18% (was 8% two quarters ago) |
| Furnace nameplate | 1,200 m² processed per day |
Use a real industrial power price as a sanity check. As of 2024, the U.S. industrial average retail electricity price was 8.13 cents per kWh (EIA average retail price by sector). At that rate, energy is 20 × $0.0813 ≈ $1.63 per m² processed. Your case exhibit may round; the point is that energy is small next to blank and scrap, not that you memorize the EIA table.
Variable cost per m² processed = 22 + 1.63 + 4.00 = $27.63.
At 18% scrap, every good m² requires 1 / 0.82 ≈ 1.22 m² of processing:
Cost per good m² ≈ 27.63 × 1.22 ≈ $33.70.
Contribution ≈ 38 − 33.70 = $4.30, about 11%. Thin.
Step 4 — Compare levers with the same equation.
| Lever | New assumption | Cost per good m² | Contribution | Margin |
|---|---|---|---|---|
| Status quo | 18% scrap | $33.70 | $4.30 | 11% |
| Fix scrap to 8% | Yield restored | $30.03 | $7.97 | 21% |
| Raise price 10% | ASP $41.80, scrap still 18% | $33.70 | $8.10 | 19% |
| Cut blank 5% | Blank $20.90, scrap 18% | $32.55 | $5.45 | 14% |
A 10% price hike and a return to 8% scrap move the margin by roughly the same amount on paper. That is useful. It does not mean they are equally good first moves. Scrap also hits the physical constraint:
- At 18% scrap, good output ≈ 1,200 × 0.82 = 984 m²/day.
- At 8% scrap, good output ≈ 1,200 × 0.92 = 1,104 m²/day.
That is about 120 more good m² per day with no new furnace. If the plant is sold out on good product, scrap repair grows both margin and revenue. A pure pricing study grows neither if contractors walk or if volume is already capped by good yield.
Step 5 — Prioritize the first data request. Ask for scrap by furnace, shift, and glass thickness for the last six months, and for utilization of good capacity. Do not open with a brand survey. The 4Ps answer asked the wrong question first; the unit answer earns the exhibit the interviewer wanted to hand you.
That is first principles in a case: not a story about rockets, a reproducible decomposition that beats a memorized marketing spine on an operations prompt.
How I score the opening 90 seconds
| Signal | What I write on the sheet |
|---|---|
| Names the unit and an equation | Strong structure |
| Opens 3Cs / 4Ps on a yield problem | Template without judgment |
| Decomposes past the data layer | Over-engineered |
| States which lever to test first and why | Prioritization |
| Ties constraint to recommendation | Synthesis-ready |
In a typical first round of eight candidates, I usually hear two or three who name a real unit in the first minute. The rest recycle labels. That gap is not intelligence. It is practice.
Practice this on a live clock
Take any ops or margin prompt and force yourself to write the unit equation before you draw buckets. Then run one timed structure drill and compare your first branch to what a memorized framework would have chosen.
For a full case with an interviewer who will push your tree, start the five-minute typed rep at /start (no account needed), or the live voice case at /interview. Everything on CoachNed is open for seven days with no card; then $120 for a recruiting season or $49 a month. CoachNed is not affiliated with McKinsey, BCG, Bain, Deloitte, or any firm named above.
Frequently asked questions
What is first principles thinking in a case interview?
It means reducing the client's decision to a unit of economics and the constraints that bind that unit, then rebuilding your structure and data requests from that equation instead of from a memorized framework label.
When should I use first principles instead of a standard framework?
Use it when the prompt turns on process, yield, capacity, energy, scrap, or another physical or operational constraint. Keep a standard profit or market-entry spine when the prompt maps cleanly onto those trees.
How deep should my decomposition go?
Two to three levels is enough for most interviews. Stop when the next layer would not change what you ask the interviewer for next.
Do interviewers expect me to cite first principles or famous examples?
No. They expect a clear structure, prioritized drivers, and numbers. Named philosophies and founder anecdotes do not appear on the scoring sheet.
How can I practice unit-level structure?
Write the unit equation for every practice case before you speak, then use timed structure drills and full cases until naming the unit feels automatic under pressure.
Sources
- Interviewing | Bain & Company — Bain's public description of case interviews rewarding structured, precise, creative problem-solving. Checked 2026-09-24.
- Interviewing at McKinsey | McKinsey & Company — McKinsey's public description of the problem-solving interview: structure, key issues, facts and data, recommendations. Checked 2026-09-24.
- 2024 Total Electric Industry — Average Retail Price (cents/kWh) | U.S. EIA — U.S. industrial average retail electricity price of 8.13 cents/kWh in 2024, used as the energy cost sanity check in the worked example. Checked 2026-09-24.