The McNamara fallacy is the mistake of judging a situation only by what can be readily measured while dismissing relevant evidence that is harder to quantify. A dashboard can report completed milestones, high utilization, and successful training attendance while leaving out whether customers can use the service or employees can apply what they learned.
The numbers may be accurate. The mistake lies in treating them as a complete account of the outcome.
This guide explains the meaning of the McNamara fallacy, compares it with Goodhart’s law, and shows how to review a decision using both metrics and evidence those metrics omit. The workplace examples and worksheet are practical illustrations, not findings from a study.
What is the McNamara fallacy?
The McNamara fallacy, also called the quantitative fallacy, occurs when ease of measurement determines which evidence is allowed to influence a decision. What is difficult to count gets excluded, and the exclusion is mistaken for proof that it does not matter.
For example, a manager may use training completion as the only evidence that a new process is ready. Staff demonstrations reveal that several routine tasks still require workarounds, but the manager dismisses those observations because they do not appear in the completion report. The failure is not counting attendance. It is rejecting relevant evidence about capability.
Distinguish three situations:
- A measurement limit: the team knows its dashboard covers only part of the outcome and seeks additional evidence.
- A measurement error: a reported value is inaccurate or its calculation is wrong.
- The McNamara fallacy: the team treats evidence outside its accepted measures as irrelevant, even when it could change the decision.
Not every missing metric is a fallacy. No report can include everything. The issue is whether relevant omissions are acknowledged and investigated or dismissed simply because they are awkward to quantify.
Where does the term come from?
Social researcher Daniel Yankelovich introduced the term in his October 15, 1971 speech, The New Odds. He named it after Robert McNamara and criticized the exclusion of relevant information from a strongly quantitative approach to management. The speech discusses both market research and the use of measures in Vietnam.
Yankelovich’s warning can be paraphrased as four steps: measure what is easy; discard or arbitrarily score the difficult evidence; assume it is unimportant; then make decisions as though it does not exist. These are a conceptual warning, not an experimentally validated sequence every organization follows.
The distinction matters for attribution: the formulation is Yankelovich’s critique, not a quotation in which McNamara names his own fallacy. It also does not establish a single-cause explanation for a war or any other complex historical outcome.
McNamara fallacy examples at work
Look for a decision in which an available count is accepted while relevant observations are ruled out. The following examples are hypothetical.
| Context and measure | Evidence being dismissed | A useful check |
|---|---|---|
| Training: completion rate | Employees cannot perform a common task without help. | Observe representative tasks and record where assistance is needed. |
| Customer support: tickets closed | Customers describe unresolved problems after closure. | Review case histories and follow up with a varied sample of customers. |
| Transformation: milestones delivered | Frontline teams depend on undocumented workarounds. | Trace an actual workflow from start to finish. |
| Capacity: utilization | Mentoring and coordination work are absent from the allocation categories. | Review how work is recorded and investigate recurring handoff delays. |
Some omitted evidence can eventually be measured. Repeated task observations, for example, may support a useful completion or error measure. That does not make the initial observations worthless while the team develops its measurement method.
Conversely, a vivid complaint is not automatically representative. Its value may be to reveal a failure mode worth checking. The next question is how often it occurs, under what conditions, and with what consequences.
For the broader design of an evidence set, see organizational transformation metrics. A useful measure answers a specific question; it does not replace every other question about success.
McNamara fallacy vs Goodhart’s law
The McNamara fallacy concerns evidence left out of a decision. Goodhart’s law concerns a measure losing its usefulness when it becomes a target. They overlap, but they describe different problems.
| Question | McNamara fallacy | Goodhart’s law |
|---|---|---|
| Main problem | Relevant evidence is dismissed because it is difficult to quantify. | Targeting a measure weakens its connection to the desired outcome. |
| Can the numbers be accurate? | Yes. They can accurately describe an incomplete picture. | Yes. A correctly counted result can become a poor proxy for success. |
| Diagnostic question | What important evidence are we refusing to consider? | How could this target be achieved without the intended progress? |
| Initial response | Investigate the omitted evidence and its relevance. | Review the target, incentives, proxy, and unintended effects. |
A support manager can commit the McNamara fallacy without introducing a closure target: they might simply disregard customer accounts because only numerical reports are accepted. If a closure quota then encourages premature closure, a Goodhart effect can arise as well.
Neither concept requires deliberate fraud. Measures can become misleading through narrow definitions, changed behavior, or an incomplete understanding of the work. The Goodhart’s law guide and KPI-review worksheet examines those target-related mechanisms in more detail.
Worked example: a green dashboard, unresolved customer problems
This is a hypothetical example. The figures below illustrate a decision process; they are not research results or recommended universal thresholds.
A company is preparing to expand a new customer-service process from a pilot to three more teams. Its dashboard reports all planned milestones complete, 96% training completion, and a growing number of closed tickets. The expansion recommendation is green.
Two frontline employees report that customers with a particular account type cannot complete a routine change without manual intervention. Three customer interviews describe having to repeat information after a handoff. The sponsor responds that these reports cannot be used because they are not statistically representative.
The sponsor is right that a few interviews cannot establish prevalence across all customers. But that limitation does not justify ignoring a potentially important failure mode. The reports warrant investigation before the team treats the process as ready to scale.
Test the warning before changing the conclusion
- Specify the claim. Which account type, task, and handoff are involved?
- Observe the workflow. Follow cases through the process and inspect the associated records.
- Broaden the sample. Include different teams, account types, successful cases, and unsuccessful cases. Do not select only cases that support the concern.
- Record the limits. Distinguish a reproducible defect from an estimate of how many customers are affected.
- Make a proportionate decision. Depending on severity and scope, fix the issue before expansion, restrict the affected workflow, or continue a controlled pilot with explicit monitoring.
Suppose the checks reproduce the account-specific failure. The team now has grounds to repair that workflow and test it again before exposing more customers to it. Training and milestone data remain useful, but they no longer carry the entire readiness decision.
This is an important distinction in delivery assurance: confidence should depend on evidence about the outcome and its risks, alongside evidence that planned activities have been completed.
How to avoid the McNamara fallacy
1. Define the outcome before reviewing the score
State what should become possible or improve. “Employees attended training” describes an activity. “Employees can complete the new process accurately without unplanned help” describes a capability. Decide which claim the evidence needs to support.
2. Write down what each metric cannot establish
Give important measures a short limitation statement. For example: “Completion rate confirms participation; it does not demonstrate independent task performance.” This makes the boundary visible during the decision instead of burying it in a reporting methodology.
3. Gather evidence that addresses a specific gap
Use interviews, direct observation, case reviews, or demonstrations where they answer a relevant question. Record how participants or cases were selected, what was observed, and where accounts disagree. Separate observations from interpretations.
4. Check alternative explanations
A reported difficulty might reflect a system defect, unclear instructions, an unusual case, or a temporary learning period. Look for evidence that distinguishes those explanations. Do not promote the first plausible story into a conclusion.
5. Avoid arbitrary precision
Turning every concern into a score can conceal uncertainty rather than resolve it. If you introduce a rating, define its criteria and what a difference between ratings means. When a defensible score is unavailable, a bounded description with supporting evidence can be more informative.
6. Connect findings to a decision and an owner
Specify what happens if the concern is confirmed, who has authority to act, and when the evidence will be reviewed. A qualitative appendix that nobody uses does little to correct an otherwise narrow decision process.
Before a major commitment, a premortem analysis can help surface ways the plan might fail despite healthy indicators. Treat those scenarios as prompts for investigation, not proof that failure will occur.
Copyable evidence-review worksheet
Copy this worksheet into your decision record. Complete one record per material evidence gap. It is a practical aid developed for this guide, not a validated assessment instrument.
Decision and date: What are we deciding, and by when?
Desired outcome: What real-world result must the decision support?
Current metric: What does the reported number actually establish?
Omitted evidence: What relevant condition or experience is outside the measure?
Verification method: Which observations, records, or conversations will test the concern? How will we select cases?
Alternative explanation: What else could explain the observation? What would contradict our interpretation?
Uncertainty: What remains unknown, and does it affect this decision?
Action: What will we do if the concern is confirmed, rejected, or unresolved?
Owner and review date: Who will investigate, who can decide, and when will we review?
Completed example: rollout readiness
- Decision
- Whether to expand the pilot to three additional teams at the next release review.
- Outcome
- Customers can complete the intended account change without an avoidable manual workaround.
- Current evidence
- Milestones are complete and training attendance is high. Neither demonstrates that the affected customer journey works.
- Gap and verification
- Investigate reports about one account type by observing tasks and checking case histories across teams. Include cases that completed successfully.
- Alternative explanation and uncertainty
- The issue may be account-specific or related to instructions. The initial reports do not establish its prevalence.
- Action and ownership
- The service owner verifies scope with the process and support leads before the release review. The sponsor then approves expansion, restricts the affected workflow, or defers expansion with a recorded reason.
Frequently asked questions
What is the McNamara fallacy in simple terms?
It is assuming that only what you can readily measure matters. You use the available numbers to judge a situation while dismissing other relevant evidence because it is harder to quantify.
What is a simple example of the McNamara fallacy?
A company declares training successful because everyone attended, while ignoring demonstrations that employees cannot perform the task. Attendance may be correctly measured; the conclusion about capability is unsupported.
Is the McNamara fallacy the same as Goodhart’s law?
No. The McNamara fallacy concerns omitted or dismissed evidence. Goodhart’s law concerns a measure becoming less useful when it is made a target. Both can affect the same dashboard.
Does avoiding the fallacy mean trusting anecdotes?
No. An anecdote may identify a question worth investigating, but it does not necessarily establish a widespread pattern. Check records, examine a range of cases, consider contradictory evidence, and state the limits of your conclusion.
Can qualitative evidence be reliable?
Its usefulness depends on how it is gathered and interpreted. Clear questions, transparent case selection, documented observations, and checks against other evidence make conclusions easier to assess. Qualitative evidence is not automatically reliable, just as a numerical score is not automatically valid.
Should organizations stop using KPIs?
No. Use KPIs for the questions they can answer, acknowledge their limits, and seek additional evidence where the decision requires it. Adding more metrics helps only when they cover a meaningful gap without creating new distortions or unnecessary reporting work.
Use the next review to test what is missing
At your next performance or delivery review, choose one important green indicator. Ask: “What could still be going wrong even if this number is correct?” Identify one relevant gap, agree how to check it, and assign someone to bring the findings into the decision.
The aim is a better-supported judgment: one that explains what the measures show, what they leave out, and why the available evidence is sufficient—or insufficient—to proceed.
Sources and scope
- Daniel Yankelovich, The New Odds, speech delivered October 15, 1971; transcript reproduced with permission by Ryan Madden. Primary source for the term and the four-step warning.
- For the provenance of the transcript, see Ryan Madden’s account of locating the original source.
The workplace examples, decision process, and worksheet are the practical synthesis of this guide. They are not claimed as an empirically validated framework. The historical term is used to examine a reasoning error, not to reduce complex events to one managerial practice.