Cost Benefit Analysis 2026: Only Incremental Cash Counts

Four cost benefit analysis conventions, four answers

Before any number goes down, ask which convention the client underwrites on. The firms that publish worked cases do not agree, and the disagreement changes the sign of the answer.

Bain's coffee shop case puts the whole cost of opening into the first year. Its threshold is therefore a one-year capital recovery. McKinsey's Electro-Light case sets a lump sum of launch cost against unit margin and attaches no horizon at all. Oliver Wyman's oil and gas case computes an annual profit of 17.5 crore, then refuses the project because the payback exceeds the client's five-year limit, checked 30 September 2026 at oliverwyman.com.

The fourth convention comes from a government rather than a firm. It is the only one of the four that names a rate. Office of Management and Budget Circular A-94 states: "Constant-dollar benefit-cost analyses of proposed investments and regulations should report net present value and other outcomes determined using a real discount rate of 7 percent. This rate approximates the marginal pretax rate of return on an average investment in the private sector in recent years", checked 30 September 2026 at whitehouse.gov.

Four sources, four bars: one-year recovery, undiscounted volume, a five-year payback limit, and a 7 percent real discount rate. The same cash flows pass some of those and fail others, which is why the convention question comes first.

The automation business case as it is first presented

A distributor picks 2.4 million order lines a year across one warehouse. A vendor proposes a goods-to-person system. Capital cost is 14.6 million dollars, useful life ten years, and the finance team uses a 9 percent nominal cost of capital.

The benefit case arrives from operations with three lines, and the numbers are stated in constant dollars.

The mispick line: error rate falls from 1.4 percent of lines to 0.3 percent, which is 1.1 points, and 1.1 percent of 2.4 million lines is 26,400 errors, at 38 dollars each, so 1,003,200 dollars.

Simple payback is 14,600,000 divided by 3,099,200, which is 4.71 years. Net present value at 9 percent over ten years uses an annuity factor of 6.4177, so 3,099,200 times 6.4177 is 19,889,605 dollars, less the 14.6 million of capital, which is a net present value of 5,289,605 dollars. On those numbers the project is approved in the meeting.

Three tests that remove non-incremental benefit from the analysis

Every one of the three benefit lines above survives a casual reading and two of them do not survive these tests.

1. Is it cash? An avoided allocated overhead is not cash leaving the building. Only the cash part of a cost saving counts.
2. Is it incremental to this decision? A saving already committed to another approved project cannot be counted twice.
3. Would it happen anyway? A benefit that arrives from a trend regardless of the investment belongs in the baseline.

Applied here, two lines move. Of the 58 picker roles, 14 were already going to be removed by a route-density project the board approved in June, so only 44 are incremental to the automation: 44 times 52,000 is 2,288,000 dollars. And the 38 dollar mispick cost contains 11 dollars of allocated warehouse overhead that stays whether or not the error happens, so the cash cost is 27 dollars: 26,400 times 27 is 712,800 dollars.

The cost lines pass all three tests unchanged. Maintenance, software and power are cash, incremental, and would not occur without the system.

The same case after the tests

Rebuild it with the revised lines and nothing else changed.

The revised net benefit is 2,288,000 plus 712,800 less 920,000, which is 2,080,800 dollars. Payback is 14,600,000 divided by 2,080,800, which is 7.02 years. Net present value is 2,080,800 times 6.4177, which is 13,353,862 dollars, less 14.6 million, so negative 1,246,138 dollars.

The project moved from plus 5.29 million to minus 1.25 million dollars, a swing of 6.54 million, and no assumption about the technology changed. Two thirds of the swing came from the headcount already promised elsewhere, and one third from eleven dollars of overhead inside a unit cost.

That is the whole discipline, and it is why the benefit list has to be interrogated line by line before any discounting happens. Discounting a wrong benefit precisely is still wrong.

A published business case that names the error in its own hint

Deloitte's Recreation Unlimited case puts the same trap in front of a candidate and then labels it. Question four reads: "If Recreation Unlimited chooses to improve their website it will cost them $10M (one-time cost), but they expect that each visitor will spend an additional $1. The profit margin on any products purchased on the website is 50%. How would you evaluate the potential opportunity of improving the website? The client has also provided us with the data point they have 20M people visit the site annually", checked 30 September 2026 at usrecruiting.deloitte.com.

Under the question the firm prints a hint that separates two calculations: "Calculation A: Contribution is the case influx generated by the additional visitors and spending. It does not account for the profit margin on products sold. Calculation B: Includes profit margin and is the actual profit generated."

The firm's answer table is an image, so the arithmetic below is ours. 20 million visitors spending one dollar more each is 20 million dollars of incremental revenue, which is calculation A. At a 50 percent margin that is 10 million dollars of incremental profit, which is calculation B. Against a one-time cost of 10 million dollars, the investment returns its cost in a single year.

Both numbers are correct. Only one of them can be set against the 10 million dollar cost. A candidate who uses 20 million reports a six-month payback on a project that takes a year. The error is a factor of two, on a case that prints its own warning.

The discount rate is a decision

Our revised case is negative at 9 percent. Change only the rate and watch what happens. The benefits were stated in constant dollars, so a real rate is the right one to use here. A-94 puts the rule plainly: "Nominal and real values must not be combined in the same analysis."

At the 7 percent real rate the circular prescribes, the ten-year annuity factor is 7.0236. So 2,080,800 times 7.0236 is 14,614,668 dollars, less 14.6 million of capital, which is a net present value of about 15,000 dollars. The project is worth roughly nothing, which is a different conclusion from a 1.25 million dollar loss and from a 5.29 million dollar gain.

Solve for the rate at which it breaks even and it lands near 7.05 percent. So the answer to this business case is a sentence about the cost of capital rather than a sentence about warehouses: approve it below 7 percent and refuse it above, and the vendor's technology is not in dispute.

A-94 asks for exactly that presentation: "Analyses should show the sensitivity of the discounted net present value and other outcomes to variations in the discount rate." Deloitte's own five-step approach ends with outlining next steps and expected results, checked 30 September 2026 at deloitte.com. On this case the next step is a treasury conversation, and saying so is the recommendation.

Where a benefit case is scored

The weights below are ours, read from the rubric code our own grader runs. No consulting firm publishes a marking scheme, and the 7 percent rate above belongs to a federal circular rather than to us or to any firm.