NPV Says Yes. Should You? What Discounted Cash Flow Can't Tell You
Net present value (NPV) and internal rate of return (IRR) are the standard tools of capital budgeting — Graham and Harvey's 2001 survey of CFOs found roughly three-quarters use each always or almost always. But the two metrics can rank the same projects differently, and each has documented failure modes. IRR's pitfalls: it implicitly assumes interim cash flows are reinvested at the IRR itself, flattering high-IRR projects (McKinsey published a cautionary tale on exactly this in 2004); projects with non-conventional cash flows can have multiple IRRs or none; and as a percentage it is blind to scale — a small project with a 50 percent IRR can create far less value than a large one at 20 percent. NPV's subtler failure modes: the answer is highly sensitive to the discount rate and to the terminal value, which often carries the majority of the calculated value in long-horizon models; the point estimate projects false precision; and standard NPV treats investments as now-or-never, ignoring the value of deferring, staging, or abandoning (the real-options critique). None of this means abandoning discounted cash flow. It means demoting the number from verdict to premise: an NPV is the conclusion of a hidden chain of assumptions, and each assumption — base case, discount rate, terminal growth — is an attackable claim. Before accepting any DCF verdict, identify the two or three assumptions that move the answer most, state them as explicit claims, and test whether the decision's sign survives plausible attacks on them. The model informs the decision; the argument makes it.
About three-quarters of CFOs rely on NPV and IRR routinely (Graham & Harvey, 2001) — and the two metrics can pick opposite winners from the same list of projects. IRR flatters small, quick, high-percentage bets; NPV hides its judgment calls inside a discount rate and a terminal value. The fix isn't ditching the math. It's demoting the number from verdict to premise — one input to an argument that still has to be made.
- NPV and IRR can disagree — percentage return and value created are different questions; know which one you're asking
- IRR's hidden flattery — it assumes you reinvest every interim dollar at the IRR itself, which is why 50% IRRs breed in slide decks
- NPV's hidden judgment calls — discount rate and terminal value often decide the answer before the operating forecast gets a vote
- The number is a premise — a DCF verdict is the conclusion of attackable assumptions; attack them before you obey it
Better process, better decision quality: five connected pieces on judging, choosing, and challenging the tools behind big decisions.
- 1.Decision Quality: The Six Elements of a Good Decision — Before You Know the Outcome
- 2.Business Decision Frameworks: Which One to Use, When — The Complete Chooser's Guide
- 3.Prospect Theory: Why Your Team Fears Losses Twice as Much as It Values Wins
- 4.Real Options: Why Keeping the Door Open Is a Decision Too
- 5.NPV Says Yes. Should You? What Discounted Cash Flow Can't Tell YouYou are here
Two Numbers, Opposite Verdicts
Here is a choice every investment committee eventually faces, reduced to its skeleton (illustrative numbers, deliberately simple). Project A: invest 1 million, and the model says it returns 58 percent — IRR of 58, NPV of plus 1.4 million. Project B: invest 20 million for a 22 percent IRR and an NPV of plus 6 million. You can fund only one. The percentage points at A. The value points at B. Which number is right?
Both — they are answering different questions. IRR answers 'how fast does each invested dollar grow?' NPV answers 'how much richer does this make us?' A company is not in the business of maximizing percentages; it banks value, and B creates four times more of it. Yet in room after room, the 58 wins the applause, because a big percentage feels like a better deal than a big number.
This matters because these two metrics are not fringe tools. When John Graham and Campbell Harvey surveyed hundreds of CFOs for their landmark 2001 study, roughly three-quarters reported using NPV and IRR always or almost always. The most standardized decision arithmetic in business — taught identically in every finance course on Earth — can still deliver opposite verdicts on the same afternoon. Before your next business case lands, it is worth knowing exactly where each number bends the truth.
What the Two Numbers Actually Say
Net present value discounts a project's forecast cash flows back to today at a rate reflecting the money's risk and opportunity cost, then subtracts the investment. Positive NPV means the project clears that hurdle and creates value in today's currency. Internal rate of return runs the same machine backwards: it finds the discount rate at which the project's NPV equals exactly zero — the break-even growth rate of the invested cash.
They earned their dominance honestly. Against the pre-DCF world of payback periods and gut feel, discounting imposed real discipline: money has a time cost, risk has a price, and projects across divisions become comparable on one scale. The canonical textbooks — Brealey and Myers chief among them — have spent decades, with good reason, telling managers to trust NPV. The trouble starts where the discipline gets mistaken for a verdict.
Where IRR Flatters
IRR's failure modes are the best documented — McKinsey ran a cautionary tale on them in 2004 after watching practitioners systematically misread the metric. Three are worth memorizing.
The reinvestment fantasy
IRR silently assumes every interim cash flow is reinvested at the IRR itself. A project sporting a 50 percent IRR is claiming you can redeploy each dollar it throws off at 50 percent — usually fiction. High IRRs inflate themselves; the McKinsey piece showed real projects whose economics collapsed once reinvestment was modeled at realistic rates.
Multiple answers, or none
Projects whose cash flows change sign more than once — build, earn, then decommission or reinvest — can have two, three, or zero IRRs. The spreadsheet will still cheerfully print one of them without mentioning the others.
Percentage blindness
A percentage carries no scale. Ranking by IRR systematically favors small, quick projects over large, patient ones — the cold-open trap. If the decision is between mutually exclusive projects, NPV ranks; IRR misleads.
Where NPV Hides Its Judgment Calls
NPV's problems are quieter, which makes them more dangerous — the number looks like measurement while containing forecasts stacked on judgments. Three assumptions typically carry the verdict. The discount rate: a point or two of movement swings marginal projects across the approve line, and the rate itself is an estimate wearing a formula. The terminal value: in long-horizon models, the value assigned to everything after the forecast window often makes up the majority of the total — meaning the least knowable part of the model contributes the most to the answer. And the base-case forecast: revenue curves drawn by the team whose project is being judged, a conflict of interest so routine nobody names it.
Add the structural blind spot: standard NPV prices a project as now-or-never and management as passive, assigning zero worth to the ability to stage, defer, or abandon — flexibility that is often the most valuable thing about an uncertain investment. We unpack that inversion in the real options piece; here it is enough to note that the tidy final number encodes a dozen contestable choices, and prints none of them.
So Should We Ditch DCF?
No — and that objection deserves a straight answer, because 'the model has flaws' is exactly the argument every pet-project sponsor reaches for when the NPV comes back negative. Discounted cash flow remains the most disciplined way we have to make money-across-time comparable. Abandoning it does not liberate the decision; it hands the decision to charisma and volume, which have worse error bars than any terminal value.
The failure is not in computing the numbers. It is in the hand-off — treating the output as the decision rather than an input to it. In decision-quality terms, DCF serves the information and values elements superbly: it forces forecasts into the open and prices time and risk. What it cannot supply is the reasoning element — the step that weighs the model against everything the model cannot see, and against attacks on its own assumptions. When the printout is treated as the verdict, that step silently never happens.
The Number Is a Premise, Not a Verdict
Here is the reframe that changes how business cases get reviewed. 'This project has an NPV of plus 6 million' sounds like a fact. It is actually the conclusion of a hidden argument: IF customers adopt on this curve, IF margins hold at this level, IF the right rate for this risk is 9 percent, IF the business is worth this multiple in year ten — THEN the value today is 6 million. Every IF is a claim. Every claim can be supported or attacked with evidence. The spreadsheet just runs the arithmetic on whichever claims survive.
Structured this way, the review writes itself: the model's key assumptions become explicit claims in an argument tree, each carrying its evidence — market data supporting the adoption curve, the competitor move attacking it, the sensitivity run showing the sign flips at 11 percent. The financial model and the strategic objections finally argue in the same arena instead of alternating slides. That is the practice our decision quality thesis demands — and it is what argument mapping was built for: the DCF feeds the argument; the argument, tested, makes the decision.
The Technique: The Three-Assumption Audit
Before accepting any DCF verdict, require one artifact alongside the model: the three assumptions that move the answer most, each stated as a plain claim with its plausible range. Almost always they are the discount rate, the terminal assumption, and one operating driver — adoption, price, or margin. If the sponsor cannot name them, the model has not been understood by its own authors.
Then run the only sensitivity test that matters for the decision: does the verdict's sign survive the pessimistic-but-plausible end of each range? A project that stays positive when all three claims are attacked is a decision the model genuinely supports. A project whose sign flips inside plausible ranges is not being decided by the model at all — it is being decided by whoever chose the base case, and that argument should happen in the open.
The Diagnostic Question
In your last approved business case, could anyone in the room have named the single assumption that moved the NPV most — and what evidence supported it? If not, the committee approved an argument nobody read.
What to Do With This
Rank mutually exclusive projects by NPV, never IRR
Percentages are blind to scale. When you can only do one, the question is value created, and that is NPV's question.
Distrust heroic IRRs on principle
Anything far above your realistic reinvestment rate is flattering itself by assumption. Ask for the modified calculation or the NPV.
Demand the three driving assumptions with every model
Rate, terminal, one operating driver — stated as claims with ranges, not buried in cell C47.
Check what fraction of value sits past the forecast window
When the terminal value carries most of the answer, the model is mostly a belief about the distant future — review it as one.
Price the flexibility the model ignores
If the project can be staged, deferred, or abandoned, standard NPV is undervaluing it. Bring the option to the argument explicitly.
The Spreadsheet Isn't the Decision
Two projects, two trusted numbers, opposite verdicts — and the resolution was never going to come from a third metric. It comes from remembering what the numbers are: compressed arguments, built from claims about the future, some of which deserve to survive scrutiny and some of which don't. The committee's job was never to read the bottom line. It was to interrogate the reasoning above it.
Three-quarters of CFOs run on these two numbers, and they are right to — as instruments. The organizations that get burned are the ones that promoted the instruments to judges.
NPV says yes. The argument decides whether you should.
Sources & Further Reading
The CFO survey: roughly three-quarters use NPV and IRR always or almost always
McKinsey's field warning on the reinvestment-rate assumption inflating IRRs in practice
The canonical textbook treatment of NPV, IRR, and the documented pitfalls of each
Why now-or-never NPV misvalues irreversible, deferrable investments — the real-options critique
The six-element framework separating the information a model supplies from the reasoning a decision requires
Frequently Asked Questions
What is the difference between NPV and IRR?
Net present value (NPV) discounts a project's forecast cash flows to today and subtracts the investment, answering 'how much value does this create?' in currency. Internal rate of return (IRR) finds the discount rate at which NPV equals zero, answering 'how fast does each invested dollar grow?' as a percentage. They usually agree on accept/reject for a single project, but can rank competing projects differently — and when they conflict on mutually exclusive choices, NPV is the reliable guide because percentages ignore scale.
Why can IRR be misleading?
Three documented reasons. First, IRR implicitly assumes interim cash flows are reinvested at the IRR itself, which inflates high-IRR projects — McKinsey published a 2004 cautionary tale on exactly this practice. Second, projects whose cash flows change sign more than once can produce multiple IRRs or none, and spreadsheets print one without warning. Third, IRR is scale-blind: a small project at 50 percent can create far less value than a large one at 20 percent, so ranking by IRR favors small, quick bets.
What are the limitations of NPV?
NPV's weaknesses are its hidden judgment calls: the answer is highly sensitive to the discount rate (itself an estimate), the terminal value often contributes the majority of calculated value in long-horizon models despite being the least knowable input, and base-case forecasts are typically authored by the project's own sponsors. Structurally, standard NPV also treats investments as now-or-never, assigning no value to the ability to defer, stage, or abandon — the real-options critique.
Do companies actually rely on NPV and IRR?
Yes, overwhelmingly. Graham and Harvey's 2001 survey in the Journal of Financial Economics — the standard reference on practice — found roughly three-quarters of CFOs use NPV and IRR always or almost always for investment decisions, making them the most widely used capital-budgeting tools by a wide margin.
Should NPV make the final investment decision?
No — it should inform it. An NPV is the conclusion of a chain of assumptions (adoption curve, margins, discount rate, terminal value), each of which is a contestable claim. The disciplined pattern is to treat the number as a premise in the investment argument: state the three assumptions that move the answer most, test whether the verdict's sign survives plausible attacks on them, and weigh the model alongside factors it cannot see. If the sign flips within plausible ranges, the base-case author — not the model — is making the decision.
What is the three-assumption audit?
A lightweight review discipline for any DCF-backed proposal: alongside the model, the sponsor must name the three assumptions that move the answer most — almost always the discount rate, the terminal assumption, and one operating driver — each stated as a plain claim with a plausible range. The committee then checks one thing: does the decision's sign survive the pessimistic-but-plausible end of each range? It converts an unreadable model into an arguable case in about fifteen minutes.
How do real options change NPV analysis?
Standard NPV assumes the investment is now-or-never and management is passive afterward, so uncertainty only reduces value. If a project can be staged, deferred, or abandoned, the downside is capped while the upside remains — flexibility with real economic value that plain NPV scores at zero. Dixit and Pindyck showed that for irreversible but deferrable investments, committing whenever NPV is merely positive destroys the often-larger value of waiting for information.
Argumentree Team
Decision Intelligence
The Argumentree team explores the science of better decisions—from 18th-century mathematics to modern AI.
The Rest of the Money Stack
DCF is one instrument in the financial-decision toolkit. These cover its complements and its correctives.
The flexibility value that now-or-never NPV assigns zero — and the discipline that keeps it honest
The chain rule that puts models in the information link — and reasoning in its own
The full chooser's guide across strategy, money, risk, and execution frameworks
Let the model argue — not rule.
Argumentree turns a business case's assumptions into explicit claims with evidence, lets the sensitivity attacks land in the open, and records which arguments actually carried the decision.
Start Free 14-Day Trial →