011 min
Why Cost-Benefit Analysis uses one unit
A product lead at a subscription app has a decision to make. The company runs its own billing system, and two engineers spend a quarter of their time keeping it working. A vendor offers a paid billing service instead. Every figure in it is hypothetical.
The team writes a list of pros and cons. The pros are fewer outages, fewer failed payments and less maintenance. The cons are a yearly fee, a migration project and a new vendor to depend on.
The list settles nothing. "Fewer outages" and "a yearly fee" are measured in different units, so nobody can say whether three pros are worth more than two cons. Each person gives the items a different size in their own head, and the most confident person in the room decides.
Cost-benefit analysis removes that problem by turning every item into one unit, usually money. Once every effect has a price, the items can be added up and compared. The same method works at any scale, from a software purchase to a national bridge.
021 min
Where Cost-Benefit Analysis comes from
The method began with a question about bridges. In the 1840s the French engineer Jules Dupuit asked how to judge the value of a road, a bridge or a canal. A toll shows only part of the answer, because some users would pay a lot to cross and others almost nothing. Dupuit's idea was to add up what every user is willing to pay, and to treat that sum as the benefit to society. Alfred Marshall later built this idea into formal economic theory.
Governments then made it a rule. In the United States, the Army Corps of Engineers started to use the method after a 1936 law required it for proposed federal waterways. A later federal flood law went further and set the test that benefits, whoever receives them, must be larger than the estimated costs.
That last phrase, "whoever receives them", is easy to read past. It will matter again when this article reaches the method's limits.
033 min
How Cost-Benefit Analysis works
Suppose the product lead from the billing example decides to run the analysis. The method has the same steps in every setting.
- Frame the decision. Name the options, including doing nothing. Say who is affected and how many years the analysis covers.
- List the costs and benefits of each option.
- Give each one a money value.
- Discount amounts that arrive in later years.
- Compare the totals.
- Test the result by changing the uncertain inputs.
Which costs and benefits to list
Direct costs, such as fees and wages, are the easy part. Indirect costs, such as shared overhead, are harder to see. Intangible effects, such as lower morale or damage to a reputation, are the hardest, and they still belong on the list.
Two kinds of cost are often missed. The first is opportunity cost, the value of the best alternative use of the same time or money. An engineer who spends a month on a migration is not building something else, and that lost work is a cost even when no cash leaves the company. The second is an external effect, a cost or benefit that falls on people who are not part of the decision, such as pollution from a factory.
One kind of cost is left out on purpose. A sunk cost is money already spent and gone. It is the same whichever option is chosen, so it cannot help choose between them.
Giving everything a money value
Some items have market prices, such as a salary or a licence. For items without a price, analysts estimate what people would be willing to pay. This is Dupuit's idea again, applied to things like time saved on a journey or a cleaner river.
Discounting later amounts
A dollar received in five years is worth less than a dollar today, because a dollar today can be invested and grow. The idea is called the time value of money. Discounting turns each future amount into its present value, using a rate of return.
The University of Utah course notes give a clear case. A sum of $1,000 due in five years, discounted at 10 percent a year, has a present value of $621. The reason is that $621 invested today at 10 percent a year grows to about $1,000 in five years.
The choice of rate matters more as the years get further away. A higher rate reduces distant amounts faster, and so it can decide whether a long project passes.
Comparing the totals
Two summary numbers are common. The net present value (NPV) is the present value of benefits minus the present value of costs. A positive NPV means the benefits are larger. The benefit-cost ratio (BCR) divides the present value of benefits by the present value of costs. A ratio above 1 means the benefits are larger.
Testing the result
Every input is a forecast, so a careful analysis changes the uncertain ones and watches whether the answer flips. This is sensitivity analysis. The analysis can run before the action, to forecast, or after it, to compare what happened with what was forecast.
041 min
Cost-Benefit Analysis and declining discount rates
The Green Book shows that the discount rate need not be a single number. The UK government uses a rate of 3.5 percent a year for the first 30 years, 3 percent for years 31 to 75, and 2.5 percent after that. The rate falls as the years get further away, so a benefit in year 100 is reduced by less than a fixed 3.5 percent rate would reduce it. The design reduces the problem of discounting away the far future while keeping a firm rate for near-term projects.
051 min
A worked Cost-Benefit Analysis
Here are the product lead's numbers. They are hypothetical, in US dollars, and all amounts arrive at the end of each year over a five-year horizon, discounted at 8 percent a year.
The vendor charges $60,000 a year, and the migration project takes engineers' time worth $90,000, paid once at the start. The service saves about $110,000 a year of engineering time and recovers about $30,000 a year of payments that used to fail.
| Item | Amount | Present value at 8% |
|---|---|---|
| Vendor fee (cost) | $60,000 a year | $239,563 |
| Migration project (cost) | $90,000 once | $90,000 |
| Engineering time saved (benefit) | $110,000 a year | $439,198 |
| Recovered failed payments (benefit) | $30,000 a year | $119,781 |
| Total costs | $329,563 | |
| Total benefits | $558,979 |
The net present value is $558,979 minus $329,563, which is $229,416. The benefit-cost ratio is about 1.70. On these numbers the vendor passes.
The lead does not stop there. Nobody knows the savings exactly, so the lead asks how wrong they can be. The yearly benefit can fall by about 41 percent, to about $82,500, before the net present value reaches zero. That margin is wide enough to make the decision comfortable, and it shows which forecast to check first.
061 min
A public case: the Clean Air Act
The same method has been used on decisions that affect millions of people. In 2011 the US Environmental Protection Agency published a study of the Clean Air Act Amendments of 1990, looking at costs and benefits from 1990 to 2020. Its central estimate of the benefits was more than 30 times its estimate of the costs.
The agency also reported a range, and the range matters as much as the centre. Even its lowest estimate of the benefits was about three times the costs. A decision that passes at every point in a range of estimates is far more convincing than one that passes only at the best guess.
Most of the benefit came from one item. About 85 percent of the economic benefit was fewer early deaths from fine particles in the air, so the analysis depended heavily on how much money a prevented death is worth. The next section explains why that kind of number is where the method is most open to challenge.
072 min
Where Cost-Benefit Analysis misleads
Cost-benefit analysis gives an answer with a decimal point, and that makes it look more certain than it is. Four problems account for most of the criticism.
What has no price
Some benefits and costs resist a money value. The standard example is a human life. One way around the problem is cost-effectiveness analysis, which does not price the outcome. It compares the cost of each option per unit of outcome, such as the cost per life saved. Health economists often measure the outcome in quality-adjusted life years, which combine length and quality of life in one number.
Who gains and who loses
A positive net benefit does not say who receives the benefit. The usual test, called Kaldor-Hicks efficiency, accepts a project when the winners could in theory pay the losers and still be better off. It does not require that they do. In the usual statement of this problem, positive net benefits decide the case, whoever benefits and whoever loses.
This is the flood-law wording from earlier, "whoever receives them", showing its cost. A project can pass while the same group bears its costs every time.
Ratios that can be gamed
A benefit-cost ratio changes when an item is moved from one side to the other. Take hypothetical totals of $12 in benefits and $10 in costs, a ratio of 1.2. Treat $5 of the costs as a reduction in benefits instead, and the totals become $7 and $5, a ratio of 1.4. The net benefit is $2 either way. The Utah notes make this point and advise ranking competing projects by net present value instead of by ratio, because the ratio is so easy to change.
Distant years and future people
Discounting reduces costs and benefits that fall far in the future. The Utah notes warn that the choice of rate can be used to steer the result, and that the rate should be stated openly. For projects that last generations, such as climate policy, a high rate makes future harm almost disappear from the total, which is a question of fairness between generations as much as of arithmetic.
What the method is for
In 1996 the journal Science published a statement by eleven economists, including Kenneth Arrow and Maureen Cropper, that took a middle position. The authors said the analysis should be neither necessary nor sufficient for sound policy, and that it is still a useful framework for organising scattered information. The Utah notes reach the same view: treat the method as a way to think through a decision, not as a rule that makes the decision.
083 min
Why Cost-Benefit Analysis inputs are often wrong
So far the arithmetic has been the easy part. The harder part is whether the forecasts that go into it are right. Bent Flyvbjerg, a researcher at the University of Oxford, studied this for transport projects.
Costs usually end up higher
Flyvbjerg collected data on 258 transport projects in 20 nations on five continents. Nine in ten ran over their cost estimates. Rail projects ran 44.7 percent over on average, bridges and tunnels 33.8 percent and roads 20.4 percent. Across the 70 years the data covers, the estimates did not become more accurate.
Demand is forecast too high
Costs are only one side of the analysis. The benefits of a rail line depend on how many people ride it, and those forecasts were also off. In Flyvbjerg's data, actual rail passenger numbers averaged 51.4 percent below the forecast. Nine in ten rail projects overestimated their traffic. For roads, errors went in both directions about equally.
The Channel Tunnel
The tunnel between France and the United Kingdom shows the two errors together. When the tunnel's owner raised money from investors, it said that 10 percent was a reasonable allowance for unexpected construction costs. Construction costs ran 80 percent above budget, and revenues were half of the forecast. The project's internal rate of return, the yearly return it earned on the money invested, was negative.
Why the errors lean in one direction
Random mistakes would produce overruns and savings in equal numbers. Here, almost all the errors point the same way, so Flyvbjerg looked for causes that act on everyone. He names two.
The first is the planning mistake that psychologists call optimism bias, a habit of expecting better outcomes than past cases support. A close relative is the planning fallacy, where people underestimate how long and how costly their own plans will be. The optimism bias is unconscious, so being careful does not remove it.
The second is strategic misrepresentation. Promoters compete for scarce funds, so they have a reason to make costs look low and benefits look high. Flyvbjerg treats this as a rational response to the incentives, not as a failure of skill.
Why the worst projects get approved
The incentives also act on selection. A project that looks best on paper is the one most likely to be approved. If the analysis of that project understates costs most, then the projects that win approval are the ones with the largest hidden errors. Flyvbjerg calls this survival of the un-fittest.
Two corrections
The UK Treasury's Green Book, which sets the rules for appraisal in the UK government, requires an explicit adjustment for optimism bias. Practitioners add to their cost and time estimates and reduce their benefit estimates.
Flyvbjerg recommends a second correction called reference class forecasting. It takes an outside view: instead of forecasting the project's own events, the analyst finds a group of similar past projects, builds the spread of their real outcomes, and places the new project in that spread. Governments and companies in several countries have adopted the method.
091 min
Cost-Benefit Analysis vs nearby concepts
Cost-benefit analysis is often mixed up with ideas that share a word or a purpose. The deciding fact is what each one measures.
| Concept | What it measures | How it differs |
|---|---|---|
| BusinessReturn on investment | ||
| Not in the library yetCost-effectiveness analysis | ||
| Not in the library yetPros and cons list | ||
| BusinessOpportunity cost |
The most common confusion is between the first two rows. A return on investment answers whether a project earns more than it takes in. A cost-benefit analysis answers whether society, or the whole organisation, ends up better off once every effect is counted, including effects on people outside the deal.
101 min
How to apply Cost-Benefit Analysis
A product team can use the method without a finance department. Five habits make the difference.
List "do nothing" as an option.
The comparison is always between the options and the current course, and the current course has costs too.
Price staff time.
Engineering hours are a real cost even when no invoice arrives, because those hours have another use.
Give a range for every forecast.
Write a low, a likely and a high value, and run the analysis on each.
Find the break-even value before looking for approval.
The break-even is the value at which the net present value reaches zero, as in the billing example. If a reasonable forecast sits close to it, the decision is not as safe as the headline number says.
Correct for known bias.
Raise cost and time estimates and lower benefit estimates by the amount similar past projects missed. Ask someone who does not gain from approval to review the inputs.
Then write the assumptions down and return after the project to compare what happened with what was forecast. That after-the-fact check is how a team learns how wrong its own forecasts usually are.
?3 questions
Questions people ask
Who should check a cost-benefit analysis?
What should a team do when the result is close to zero?
Is a pros and cons list ever enough?
▶3 videos
Watch
Understanding Cost-Benefit Analysis
Cost benefit analysis
EnhanceTuition
Cost Benefit analysis | Cost Effective analysis | PSM lecture | Community Medicine lecture | Arpit
Arpits PSM made easy
§6 sources
Sources
Show all 6 sourcesHide sources
Cost-benefit analysis, encyclopedia summary covering the history from Dupuit, the US adoption and the main criticisms.
Four Problems to Watch in Benefit-Cost Analysis, course notes, University of Utah.
Benefits and Costs of the Clean Air Act 1990-2020, the Second Prospective Study, US Environmental Protection Agency, 2011.
Is There a Role for Benefit-Cost Analysis in Environmental, Health, and Safety Regulation?, Arrow and ten co-authors, Science, 1996 (abstract page).
Survival of the Un-fittest: Why the Worst Infrastructure Gets Built, Bent Flyvbjerg, draft of the paper in Oxford Review of Economic Policy.
The Green Book: appraisal and evaluation in central government, HM Treasury, 2026 edition (HTML version).





