Most candidates treat cost-benefit analysis as a two-column list: benefits up, costs down, recommend go if the gap is positive. That is accounting theater. In the room, I am listening for whether you can run a marginal decision against a counterfactual — what changes from here, when the cash moves, who pays, and what you learn by waiting.
Governments write the method down because the naive version keeps approving the wrong projects. After this piece you will be able to structure a go/no-go case the way serious appraisal actually works, show where a live programme's "positive BCR" misleads, and reproduce one full numerical recommendation.
What governments mean by cost-benefit analysis
Private cases and public appraisal share the same spine. Compare options, measure costs and benefits in the same units, bring future flows to present value, and decide against an explicit rule. The United States sets that spine in OMB Circular A-94: discount every future benefit and cost, use a real rate on real (inflation-stripped) flows, and report sensitivity when the rate moves the answer. The United Kingdom's Green Book (2026) is blunter: social CBA is the recommended method for comparing shortlisted options, with a published Social Time Preference Rate of 3.5% real for years 1–30, then a declining schedule.
Two distinctions matter in interviews. The social rate is not your client's WACC — in a corporate case you discount at the opportunity cost of capital you state and defend — but A-94 and the Green Book still teach the logic: timing is part of the decision, not packaging. And social CBA counts effects on society, not only the sponsor's P&L. In a commercial case you usually stay inside client economics, yet you still name incidence: who captures the benefit and who absorbs the cost.
Four places naive CBA lies
| Trap | What candidates do | What the sheet rewards |
|---|---|---|
| Sunk costs | Fold yesterday's spend into "total project cost" | Forward costs and benefits only; sunk money is irrelevant to the next decision |
| Discount rate | Add eight years of savings and compare to today's capex | Present-value the stream; show how a higher hurdle flips the call |
| Option value | Force a binary full rollout | Price a pilot, delay, or staged commit when uncertainty is the bottleneck |
| Incidence | Treat "labor savings" as free money | Say who loses the wage, who gains the margin, and whether that changes the client's objective |
The Green Book states the sunk-cost rule plainly: costs already incurred that cannot change must not enter the appraisal of what to do next; what can still matter is the opportunity cost of locking those assets into this use. That is the difference between "we have already spent $24 billion, so we should finish" and "from today, is finishing better than stopping?"
Discounting is the second lie. Raising the rate shrinks distant benefits more than near-term costs. For a typical investment — cash out early, cash in later — a higher discount rate reduces NPV. A-94 insists on discounting and on showing sensitivity. If your recommendation dies when the rate moves from 8% to 12%, say so before the interviewer asks.
Option value is the third. A full rollout destroys the option to learn cheaply. A pilot preserves it. You do not need Black-Scholes in a first round; you need to quantify the cost of information and the downside of being wrong at scale.
Incidence is the fourth. Automation "saves" labor only by transferring income away from workers or a contractor. A retailer may still want the project. A city measuring social CBA may not. Naming the transfer stops you from treating private gains as social wins.
Where a "positive BCR" misled in public: HS2
High Speed 2 is the clearest recent demonstration that the frame of the CBA can dominate the arithmetic. The UK National Audit Office's 2026 reset report records that DfT's May 2026 Accounting Officer assessment put the benefit-cost ratio for completing the programme in a high-level range of about 1.5 to 6.4 — medium to very high value for money — while also stating that if today's costs had been known at Notice to Proceed in 2020, the BCR would have been about 0.3 to 0.4, poor value for money. That split exists because Green Book appraisal of the live decision excludes sunk spend and scores continuation against a cancellation counterfactual that itself costs real money.
How large is that counterfactual? HS2 Ltd's published cancellation cost advice (May 2026) estimates remediation under a "realistic" scenario at £33 billion to £58 billion (2025 prices), after applying the Green Book's 66% optimism-bias uplift for early-stage non-standard civil works — and that range still excludes sunk cost to date. Completing the scheme was estimated to need another £46.8 billion to £61.7 billion left to spend. Once you treat avoided remediation as a benefit of finishing, a continuation BCR can look healthy even when a clean-sheet CBA would never have cleared.
That is not a gotcha about railways. It is the interview lesson: two analysts can share the same traffic forecasts and still disagree because one is answering "was this ever a good idea?" and the other is answering "given where we are, what should we do tomorrow?" Both questions are legitimate. Confusing them is not.
When a case hands you a half-built factory, a partly paid software licence, or a brand already launched in two cities, ask which decision the client needs: ex ante quality or forward value for money. Then pick the counterfactual out loud.
Worked example: Meridian Cold Storage
Prompt: Meridian Cold Storage runs twelve freezer warehouses in the Midwest. Management wants to automate pick-and-pack at the flagship site. Capex is $18 million. A design study already cost $2.4 million. Should they proceed?
Invented but consistent figures (all real dollars; ignore inflation for clarity):
| Item | Amount | Timing |
|---|---|---|
| Design study (already paid) | $2.4M | Sunk |
| Automation capex | $18.0M | Year 0 |
| Labor cost avoided | $3.2M per year | Years 1–8 |
| Extra energy and maintenance | $0.6M per year | Years 1–8 |
| Contribution from higher throughput | $1.1M per year | Years 1–8 |
| Severance and redeployment | $1.2M | Year 0 |
| Company real hurdle rate | 8% | Stated |
Net operating gain each year = 3.2 + 1.1 − 0.6 = $3.7M.
Naive CBA (what I stop listening to). Benefits = 3.7 × 8 = $29.6M. Costs = 18.0 + 2.4 + 1.2 = $21.6M. "Net" = +$8.0M. Recommend go. Three errors: sunk study is in the costs, cash is not discounted, and the recommendation never states a counterfactual.
Forward CBA at 8%. Drop the $2.4M study. Present value of an eight-year $3.7M annuity at 8% is 3.7 × [(1 − 1.08^−8) / 0.08] = 3.7 × 5.747 = $21.3M. Upfront outlay = 18.0 + 1.2 = $19.2M. NPV ≈ +$2.1M. Still positive, but the cushion is thin.
Sensitivity on the discount rate.
| Real hurdle | Annuity factor (8 yrs) | PV of $3.7M stream | NPV vs $19.2M outlay |
|---|---|---|---|
| 8% | 5.747 | $21.3M | +$2.1M |
| 12% | 4.968 | $18.4M | −$0.8M |
| 15% | 4.487 | $16.6M | −$2.6M |
At the company's stated 8% the project clears. At a stress rate of 12% it fails. Say that in the recommendation: "I would proceed at the stated hurdle, and I would kill it if capital is actually closer to mid-teens."
Option value of a pilot. Automate two aisles for $2.5M, keep the rest manual for a year, then decide. After the pilot, assume a 60% chance the $3.7M run-rate proves out and a 40% chance it lands near $2.0M (adoption friction, shrink, downtime). In the bad state, walk and lose only the pilot. You do not need a perfect tree — you need to show that information has a price, and that a big-bang rollout destroys the option to learn cheaply.
Incidence. The $3.2M labor saving is a transfer from warehouse labor (and possibly a 3PL) to Meridian's margin. If the objective is owner returns, name the transfer and move on. If the case includes community or ESG constraints, the social ledger diverges from the P&L — the Green Book's society-wide view exists for that reason.
Recommendation I would accept: pilot first unless delay is extremely costly; full automate only if the pilot clears 8% under conservative throughput with severance funded. Reject any go based on undiscounted totals that include the study.
Ned's rule. Never subtract "total benefits" from "total costs." Subtract the present value of incremental benefits from the present value of still-avoidable costs, against an explicit do-something-else baseline — then say whose pocket moves.
What the interviewer is scoring
McKinsey and Bain both frame cases as how you think through a business problem, not whether you can name a framework. CBA earns points when it becomes a decision — proceed, reject, pilot, or delay — and loses them as vocabulary.
I listen for four moves in the opening minutes: objective and decision rule, realistic alternatives including status quo, the drivers you will quantify first, and the risk or timing that could reverse the call. Candidates who list benefits without a counterfactual are already behind. Candidates who ask whether yesterday's spend is sunk are already ahead.
If the case is still diagnostic — profit fell, churn rose — do not force a CBA tree. Diagnose first; appraise options second. The guide to picking a case structure covers that sequencing.
Practice this on a live decision
Before any math, write: objective, alternatives, what is sunk, discount rate, and who bears each major cost. Then run base and stress-rate numbers. CoachNed's structure drill forces that spoken tree; the math drill is for the annuity step under time pressure.
Everything is open for seven days, no card; then $120 for a recruiting season or $49 a month. Try a five-minute typed rep at /start, or the ~18-minute voice case at /interview.
Frequently asked questions
What is a cost-benefit analysis framework in a case interview?
A decision method, not a brand-name structure: define the objective, compare options against a baseline, quantify incremental costs and benefits over time, discount when timing matters, and recommend proceed, reject, pilot, or delay.
How do you calculate cost-benefit analysis with a discount rate?
Estimate flows by year, use a real rate on real dollars, divide each year's net by (1 + r)^t, and subtract today's avoidable outlay. The Green Book uses 3.5% real for the first thirty years in public appraisal; corporate cases usually use the client's cost of capital. Always show one higher-rate sensitivity.
Should sunk costs be included in cost-benefit analysis?
No. Include only forward costs, plus any opportunity cost of keeping an already-bought asset in this use. HS2's continuation BCR and its clean-sheet BCR diverge for exactly this reason.
When should I recommend a pilot instead of a full rollout?
The Green Book states the sunk-cost rule plainly: costs already incurred that cannot change must not enter the appraisal of what to do next; what can still matter is the opportunity cost of locking those assets into this use. That is the difference between "we have already spent $24 billion, so we should finish" and "from today, is finishing better than stopping?"
When NPV is positive but fragile, when one or two assumptions drive the answer, or when being wrong at full scale costs more than learning. Price the pilot and state what evidence flips you to go or kill.?
High Speed 2 is the clearest recent demonstration that the frame of the CBA can dominate the arithmetic. The UK National Audit Office's 2026 reset report records that DfT's May 2026 Accounting Officer assessment put the benefit-cost ratio for completing the programme in a high-level range of about 1.5 to 6.4 — medium to very high value for money — while also stating that if today's costs had been known at Notice to Proceed in 2020, the BCR would have been about 0.3 to 0.4, poor value for money. That split exists because Green Book appraisal of the live decision excludes sunk spend and scores continuation against a cancellation counterfactual that itself costs real money.
Is cost-benefit analysis the same as NPV?
NPV is the usual metric inside a CBA. CBA is broader: which effects count, whose welfare is in scope, risk adjustments, and the counterfactual. In most private-sector cases, a clean NPV against a clear baseline is what the interviewer wants.
Sources
- OMB Circular A-94 (November 2023) — US federal guidance on benefit-cost analysis, discounting of future flows, real vs nominal rates, and sensitivity. Checked 2026-09-24.
- The Green Book (2026) — GOV.UK — UK appraisal rules: social CBA, STPR of 3.5% real for years 1–30, sunk-cost exclusion, optimism bias, society-wide costs. Checked 2026-09-24.
- Green Book supplementary guidance: Discounting (PDF) — Derivation of the Social Time Preference Rate and declining long-horizon rates. Checked 2026-09-24.
- NAO — High Speed Two reset — Completion BCR about 1.5–6.4 vs 0.3–0.4 if today's costs had been known in 2020; Green Book frame of continuation vs cancellation. Checked 2026-09-24.
- HS2 cancellation cost advice — GOV.UK — Remediation counterfactual of £33–58bn (2025 prices) with 66% optimism bias, excluding sunk spend. Checked 2026-09-24.
- McKinsey — Interviewing — Cases as business problem-solving, not memorized templates. Checked 2026-09-24.
- Bain — Hiring process — Interview emphasis on how candidates think through problems. Checked 2026-09-24.
CoachNed is not affiliated with McKinsey, BCG, Bain, Deloitte, or any firm named above.