Every Project Looks Good on a Spreadsheet: 6 Rules for Deciding Which Capital Projects Get Funded

Bent Flyvbjerg spent decades building a database of more than 16,000 capital projects across 136 countries. His finding, published in How Big Things Get Done (2023), is the most sobering number in corporate finance: 8.5% of projects came in on budget and on time, and only 0.5% delivered the budget, the schedule and the benefits that were promised when they were approved.

Half a percent. Those projects were all approved by intelligent people looking at spreadsheets that said yes.

The problem is almost never the arithmetic in the business case. It is that the business case was written by the person who wanted the money, and nobody was assigned to write the case against it.

Our group was founded in Azagra in 1890 and today runs eight active verticals, from wine and mineral water to real estate, hospitality and electrical installations, selling into more than 75 countries. In a diversified group, every vertical arrives each year with a list of things it needs to buy, and every list is longer than the cash available. Here are the six rules we use to decide which ones get funded.

Rule 1: Sort the list into four buckets before you compare a single number

The most common mistake in capital budgeting is putting every request into one ranked list and funding from the top down. That produces a portfolio where a bottling line with a 34% projected return beats a warehouse roof with no return at all, right up until the roof fails in November.

Split requests into four categories and treat them as separate budgets:

• Maintenance: replacing what already exists so the business keeps running. Not optional, not ranked against growth.
• Compliance and safety: legally or contractually required. Also not optional, and the only question is the cheapest compliant way to do it.
• Cost-out: investments that reduce a known, measurable cost. Highest confidence of any category, because you already have the baseline.
• Growth: new capacity, new markets, new lines. Highest returns on paper, lowest confidence in reality.

A project only has to beat the other projects in its own bucket, and each bucket gets its own share of the capital before anyone starts ranking. That single change removes most of the political theater from the process, because the person asking for a roof no longer has to inflate a return to compete with a growth project.

Rule 2: Set one hurdle rate, publish it, and know why it sits above your cost of capital

Every company should have a single number that a growth project must clear. Most do not, which means the required return is negotiated privately, project by project, by whoever argues best.

Here is the interesting part. Research by Jagannathan, Matsa, Meier and Tarhan, published in the Journal of Financial Economics (2016), surveyed CFOs and found that firms routinely set hurdle rates well above their own cost of capital, with the average hurdle sitting several points higher. Finance textbooks treat that as an error. It is not.

The gap is deliberate and rational, and it covers three things textbooks ignore: management attention is scarcer than money, forecasts submitted by sponsors are optimistic, and capital committed today cannot be redeployed if something better shows up in eighteen months. The hurdle rate is not a discount rate. It is the price of your own capacity to execute, and that capacity is the real constraint in almost every private company.

Publish the number. A written 15% hurdle is an argument you have once a year instead of forty times.

Rule 3: Use payback to size the risk and NPV to size the prize, and never payback alone

In their survey of 392 CFOs, Graham and Harvey found that roughly three in four always or almost always use net present value and internal rate of return, while 57% still use the payback period. Payback has an unfashionable reputation among academics and a devoted following among owners, and the owners are right for a reason the textbooks understate.

Payback is not a valuation method. It is a risk measure. It answers one question: how long am I exposed before I get my money back? In a business where the forecast beyond year three is fiction, a short payback is worth more than an elegant terminal value.

Use them together, and read them as two different questions:

• NPV and IRR: is this worth more than it costs, over its life?
• Payback: how long before I stop being wrong in an expensive way?

Reject any project where the payback runs past the useful life of the technology, however good the NPV looks. A ten-year payback on equipment that will be obsolete in seven is not an investment, it is a subscription to a mistake.

Rule 4: Make the sponsor state what changes in the operation, not just what changes in the spreadsheet

Every business case has a line that says revenue increases or cost falls. Very few say who, specifically, will do something differently on the Monday after the machine is installed.

Require three things in writing before any approval:

• The honest do-nothing case. Not "we lose the market," which is rhetoric. What actually happens to volume, cost and margin over three years if we spend zero? Most growth cases collapse the moment the do-nothing case is written by someone who is not the sponsor.
• The operating change. Which shift, which process, which headcount, which customer. If nobody's daily work changes, the savings are not real.
• The one assumption that breaks it. Every case has a single variable that decides everything, usually a utilization rate or a price. Name it, and state the value at which the project stops working.

When a sponsor cannot tell you the number at which their own project fails, they have not evaluated it, they have justified it.

Rule 5: Approve in tranches, not in totals

The classic failure is approving 2 million and discovering at 1.4 million that the assumption was wrong, at which point the cost already spent becomes the argument for spending the rest. That is the sunk cost trap, and it is not a theoretical problem. It is the standard way large projects go wrong.

Break every significant project into stages with money attached to each: design, order, build, commission. Release the next tranche only when the stage before it hit its gate. Write into the approval, at the start, the condition under which the project stops.

Do this while everyone is calm, because the moment the project is in trouble, the sunk money will do the arguing.

Rule 6: Post-audit the result, and have the same person present it

This is the rule most companies skip, and it is the one that makes all the others work. Twelve to eighteen months after a project goes live, compare what it delivered to what was promised, in the same format as the original request.

Three rules for the post-audit:

• The same sponsor who asked for the money presents the outcome. Not finance, not a consultant.
• The purpose is calibration, not punishment. You are measuring the accuracy of your forecasting, not running a trial.
• Track the bias, not just the variance. If sponsors' revenue forecasts land 20% high on average across a dozen projects, that is not bad luck, that is a coefficient. Apply it to the next submission.

The first post-audit changes one project; the knowledge that there will be a post-audit changes every business case written after it. Sponsors write more conservative numbers when they know they will stand up in a room in eighteen months and explain them.

Key Takeaways

• Sort capital requests into maintenance, compliance, cost-out and growth, and give each bucket its own budget. Never rank a roof against a growth project.
• Publish one hurdle rate for growth capital. Set it above your cost of capital deliberately, because management attention, not money, is the binding constraint.
• NPV tells you whether a project is worth more than it costs. Payback tells you how long you are exposed. You need both answers, and payback alone is not one of them.
• No approval without a written do-nothing case, a named operating change, and the value of the one assumption that breaks the project.
• Fund in tranches with defined gates, and write down the stopping condition before the first euro is spent.
• Post-audit every material project and have the original sponsor present the result. Track forecast bias and apply it to the next request.
• Of more than 16,000 projects studied by Flyvbjerg, only 0.5% delivered on budget, on time and on benefits. Assume your process, not your luck, is what decides which side of that you land on.

Frequently Asked Questions

How do you budget for capital expenditures?

Start from the maintenance and compliance requirements, because those are commitments rather than choices, and fund them first. Whatever remains of your available cash flow and debt capacity is what growth and cost-out projects compete for. Set the total envelope before you look at individual requests, or the list will expand to fill whatever number you are willing to defend.

What are the steps involved in the capital expenditure approval process?

A workable process has six: a written request with the do-nothing case and the key assumption; categorization into maintenance, compliance, cost-out or growth; financial screening against the hurdle rate with NPV, IRR and payback; approval at the authority level matching the amount; staged release of funds against defined gates; and a post-implementation audit presented by the original sponsor. Skipping the last step is what keeps companies making the same forecasting error for a decade.

Are WACC and hurdle rate the same thing?

No, and they should not be. The weighted average cost of capital is what your financing costs; the hurdle rate is the minimum return you require, which in practice sits above WACC. Survey evidence in the Journal of Financial Economics shows firms deliberately set hurdles several points higher to account for forecast optimism, scarce management bandwidth and the option value of keeping capital available.

What is a good payback period for a capital project?

It depends on the category, not on a universal number. Cost-out and maintenance investments are usually expected to pay back in one to three years because the baseline is known; growth projects are commonly given three to five. The real test is relative: the payback must be comfortably shorter than the useful life of whatever you are buying.

Should sunk costs be ignored in decision-making?

Yes, in the decision, though not in the post-audit. Money already spent cannot be recovered by spending more, so the only relevant question at any stage gate is whether the remaining cost buys the remaining benefit. The sunk cost belongs in the post-audit, where it teaches you something about your forecasting, rather than in the room where the next tranche is approved.

One thing to do this week

Take the largest capital project your group approved two years ago. Pull the original business case, put the promised numbers next to the actual numbers, and calculate the percentage gap on the single most important assumption. That percentage is your forecasting bias, and you should be applying it to every case on your desk right now.

Explore the eight businesses of Manzanos Enterprises, and for the decisions that sit around this one, read capital allocation: the CEO's most important job, the five tests before you own or lease your building and how to negotiate with your bank on credit terms and covenants.