Why the Payback Period Lies When the Cash Flows Are Uneven

One question, answered honestly; three questions, not answered at all

The payback period is defined without ambiguity: accumulate the net cash inflows year by year and record the moment the running total reaches the initial outlay. It is arithmetic, not judgement, and within its own terms it never lies. The trouble is that people use it to answer a question it was never asked — is this project worth doing, and is it better than that one — and for that it is not merely imprecise, it is capable of giving the wrong answer with complete confidence.

The first thing it throws away is everything after the cut-off. That is not a subtlety, it is the entire design: the measure stops at the moment of recovery by construction. Two projects identical up to the payback date and wildly different afterwards get the same score. A machine that runs profitably for ten more years and a machine that seizes up the following month are indistinguishable to it. Whatever else you do with the number, do not let it rank alternatives with different lives.

The second thing it throws away is the time value of money, and the third — the one nobody mentions — is the shape of the cash inside the year. A payback quoted as 1.67 years on a project returning 60,000 in each of two years is not a measurement, it is a straight-line interpolation inside year two. If the money actually arrives as an annual settlement in December, the true payback is 2.00 years, four months later than the number on the slide. Seasonal businesses, annual rebates, harvest cycles and grant instalments all break the interpolation in the same direction: the cash is later than the fraction implies.

The ranking inversion, computed

Two projects, each costing $100,000. Project A returns $50,000 in year one, $50,000 in year two, then $5,000 a year for three more years: it is the classic front-loaded case, a marketing push or a piece of equipment for a contract that ends. Project B returns $30,000 a year for five years: the classic steady case, a machine that simply works. Payback for A is exactly 2.00 years. Payback for B is $100,000 divided by $30,000, which is 3.33 years. On the payback rule, A wins by sixteen months and it is not close.

Now count the whole life. A returns $115,000 in total; B returns $150,000. Discount at 8 % and A is worth $210 of net present value — technically positive, economically indistinguishable from doing nothing — while B is worth $19,781. The gap is $19,571 in B's favour, on an investment of $100,000. The internal rate of return tells the same story from the other end: 8.13 % for A, which is barely the cost of capital, against 15.24 % for B. Payback preferred the project that returns a fifth of the value.

It is worth being precise about why, because the reason is not that payback is crude. Payback ranked correctly on the question it was asked: A really does return the money faster. The error is one of substitution — using a liquidity measure as a value measure. That distinction is what makes the rest of this article possible: once you know payback is measuring recovery speed and nothing else, you can say exactly when recovery speed is the thing you care about, and stop pretending the rest of the time.

The end-of-life cost that payback cannot see

Add a third project. Project C costs $100,000, returns $40,000 a year for four years, and then costs $60,000 in year five to dismantle, restore, decontaminate or dispose of. Its payback is 2.50 years — better than steady Project B, which needs 3.33. Its net present value at 8 % is minus $8,350, and at the 4 % real financial discount rate the European Union prescribes as a benchmark for appraising co-financed investment, minus $4,120. Its undiscounted total is exactly zero: $160,000 in, $160,000 out. Payback ranked a value-destroying project second out of three.

This is where the most common patch fails. Discounted payback — accumulate the discounted inflows instead of the raw ones — is often offered as the fix, and it does fix one of the three defects. It does not fix truncation. Project C still shows a discounted payback of 2.90 years at 8 %, still inside the horizon, still apparently fine, while its net present value is negative. Discounted payback tells you when the discounted money comes back, which is a better question than the undiscounted one and still not the question you are asking.

One more thing about Project C is worth noticing, because it also disposes of the obvious alternative. Its cash flows change sign twice — out, in, out — so the internal rate of return is not uniquely defined for it: the polynomial has more than one root and the number your spreadsheet returns depends on the guess you seeded it with. Net present value has no such difficulty, which is why the discipline that has to appraise public investment across the Union settles on discounting rather than on any rate-shaped summary. Where a project has a restoration obligation, a lease dilapidation, a battery replacement or a decommissioning bond, net present value is not one option among several. It is the only one of the three that reads the whole timeline.

The shortcut is exactly backwards

There is a rule of thumb that circulates wherever payback does: divide one by the payback period and you have roughly the annual return. A four-year payback becomes 25 %, a two-and-a-half-year payback becomes 40 %. It is worth taking seriously because it is not arbitrary — for a project that pays a level annuity, the payback period is exactly the annuity present-value factor, so its reciprocal is exactly the capital recovery factor. That means the shortcut has a closed-form error: the reciprocal exceeds the true internal rate of return by r divided by ((1+r) raised to the number of years, minus one). That term is always positive and it shrinks as the life lengthens.

Compute it and the pattern is startling. Fix the payback at four years, so the shortcut always says 25 %. If the asset lasts exactly four years the true return is 0 % — you got your money back and nothing else. Five years: 7.93 %. Six years: 12.98 %. Eight years: 18.62 %. Ten years: 21.41 %. Fifteen years: 24.01 %. Twenty years: 24.70 %. Fix the payback at two and a half years instead, so the shortcut says 40 %, and a three-year life returns 9.70 %, a four-year life 21.86 % and a five-year life 28.65 %. The overstatement is 30.3 points, 18.1 points and 11.4 points respectively.

Read the pattern the right way round and it undoes the usual advice. The shortcut is nearly right only when the asset lives four or five times as long as its payback — a building, a grid connection, a long-lived machine. It is worst on short-lived assets, and short-lived assets are exactly the case in which payback is normally recommended. So the two defences of the measure pull apart: it is legitimate to use payback on a three-year piece of equipment because the horizon is genuinely short, and it is not legitimate to turn that payback into an implied return, because on a three-year life the reciprocal is out by thirty points. Use payback as a duration. Never use it as a rate.

When payback is nonetheless the right tool

The honest case for payback begins by admitting what it measures: how long the business is exposed. That is a liquidity question, and there are situations in which liquidity, not value, is the binding constraint. A business with eleven months of cash cannot fund a project that pays back in eighteen, however large its net present value, because it will not be there to collect. In that world the correct cut-off is not a preference and not a habit — it is derived from the cash runway, and the project that recovers inside the runway is the one that can be attempted at all. The measure is answering exactly the question that matters.

The second legitimate case is a horizon that is genuinely short and genuinely known. A three-year lease with no renewal, equipment whose useful life is set by a certification that expires, a contract with a fixed end date: here the truncation defect largely disappears, because there is nothing after the cut-off to throw away. The third is a risk that ends the project from outside — expropriation, a licence that may not be renewed, a regulatory reversal, a technology that a competitor is about to obsolete. In all three, the cash flows beyond the horizon are not merely uncertain, they are the ones you positively do not believe, and a measure that refuses to count them is expressing a view rather than committing an error.

In every other case, use it as a screen and not as a rule. Compute the payback because it is free and because it names the exposure; then compute the net present value, because that is the number that says whether the project makes the business richer. Where the two disagree, as they did for Projects A and B, the disagreement is the finding: it means the ranking is being driven by timing rather than by value, and someone has to decide explicitly whether the business is short of cash or short of good projects. The public appraisal frameworks resolved this a long time ago — the European Union's rules for co-financed investment prescribe discounting at a 4 % real financial benchmark, and the United Kingdom's Green Book prescribes discounting at a social time preference rate of 3.5 % in real terms for the first thirty years and 3.0 % from year thirty-one, with a review opened in December 2025 recommending the headline move to 3.0 %. Neither of them makes recovery speed the criterion. Neither should you, unless you can say out loud which constraint is binding.

Frequently asked questions

Is discounted payback the fix?

It fixes one defect out of three. Discounting the inflows before accumulating them removes the assumption that a euro in year three is a euro today, which is real progress. It does nothing about truncation, and truncation is the defect that produces wrong decisions. Project C in this article — four years of inflows followed by a large dismantling cost — has a discounted payback of 2.90 years at 8 % and a net present value of minus 8,350. The discounted measure said the project was fine because it, too, stopped counting at the moment of recovery. Nor does discounting fix the within-year interpolation, which still assumes cash arrives evenly through the year in which recovery falls.

What payback period should I require?

There is no derivation for it, which is the honest and unwelcome answer. Net present value has an accept rule that follows from something real — accept when the value discounted at your cost of capital is positive. Payback has a cut-off that somebody chose. The only defensible way to choose it is to derive it from the constraint it is standing in for: if payback is a proxy for liquidity, set the cut-off at the point where the business would run out of cash under an adverse scenario, and check it against the actual runway rather than against last year's rule. If you cannot connect the cut-off to a constraint, you are not applying a criterion, you are applying an inherited number — and you should compute the net present value alongside it and be ready to override.

When do payback and net present value agree?

When the projects being compared have the same life and the same shape of cash flow, and differ only in scale. Two identical machines, one bigger, will be ranked the same way by both. The moment lives differ, or the profiles differ — one front-loaded, one flat, one with a tail or an end-of-life cost — the agreement is coincidence rather than structure. That is a useful practical test: before trusting a payback ranking, ask whether the alternatives have the same life and the same profile. If they do, the ranking is safe and cheap. If they do not, compute the net present value, because you have no reason to expect the two to agree and this article shows three cases where they do not.

Can I use one divided by the payback period as an annual return?

No, and the error is calculable rather than vague. For a project paying a level annuity, the payback period equals the annuity present-value factor, so its reciprocal equals the capital recovery factor — which exceeds the true internal rate of return by r divided by ((1+r) to the power of the number of years, minus one). Concretely: a four-year payback implies 25 %, but the true return is 0 % if the asset lasts four years, 7.93 % over five, 12.98 % over six, 18.62 % over eight and 24.01 % over fifteen. The shortcut only converges on the truth for assets that live several times their payback period. Since the usual argument for using payback at all is that the horizon is short, the shortcut is worst precisely where the measure is most defensible.

If it is this flawed, why does everyone still use it?

Because it is free to compute, impossible to misunderstand, and it silently encodes a constraint that the value calculation leaves out. Net present value assumes you can raise the money and survive the wait; a small business often cannot, and payback is the crude way that assumption gets challenged. It also travels well: a workshop foreman, a bank manager and a shareholder will all understand two and a half years, and none of them will agree on a discount rate. The reasonable position is not to abolish it but to demote it — report the payback because it describes the exposure, decide on the net present value because it describes the value, and treat any disagreement between them as a question about which constraint is actually binding.