How to use base rates before you trust an ambitious forecast
Use base rates by examining what happened in comparable situations before you adjust a forecast for your company’s plan. Define the outcome, select a defensible comparison group, and show where your case differs. The comparison should discipline the forecast without pretending that historical cases determine your result.
A base rate is the observed frequency or distribution of an outcome within a defined comparison group. Kahneman and Lovallo’s 1993 discussion of the outside view contrasts reasoning from comparable outcomes with reasoning mainly from the details of the current plan. An attractive narrative deserves examination alongside the comparison, rather than permission to replace it.
For a CEO, the useful question is practical: what usually happened when organizations attempted something sufficiently similar? That question can improve a forecast discussion even when the available sample is imperfect. It also requires judgment about which similarities matter.
The method below, called the comparison-first forecast, is a proposed management practice. It is not a validated forecasting instrument, and it cannot manufacture reliable data where none exists.
Start with the decision the forecast will support
A forecast for staffing needs a different resolution from a forecast supporting an irreversible acquisition commitment. Identify the decision before requesting more analysis. Otherwise, a team may polish a number without knowing what degree of uncertainty the decision can tolerate.
In this explicitly composite scene, a founder reviews an expansion proposal at home after dinner. Their partner has started a film, but the founder keeps returning to the launch estimate. The forecast assumes a rapid launch, strong customer uptake, and little disruption to the existing operation. Each assumption has a reasonable explanation. Together, they support a commitment that would be difficult to unwind.
The founder asks for comparable expansions. The presentation contains market size and customer enthusiasm, but no record of how similar launches performed. The missing comparison does not establish that the forecast is wrong. It establishes that the meeting has evaluated a plan more thoroughly than its likely outcome.
The next useful request is a comparison memo, with a clear deadline and a decision it must inform. This belongs within a broader CEO decision system, where evidence requirements increase with the consequences of being wrong.
The comparison-first forecast
1. Define the outcome before choosing the cases
Write the forecast as an observable claim. Specify what will happen, by when, and under which operating conditions. Separate revenue, margin, adoption, implementation time, and cash requirements. A project can meet one forecast while missing another.
For a new service, distinguish signed interest from paid use. For an acquisition, distinguish closing the transaction from integrating the operation. For a product launch, distinguish initial demand from repeat purchase. These are proposed examples of measurement boundaries, not claims about typical business performance.
Then decide which outcome matters to the current commitment. If the cash requirement determines whether the company can proceed, a collection of revenue success stories cannot answer the central question.
Record the definition before searching for cases. This reduces the opportunity to select a favorable measurement after seeing the available results. If the definition changes, preserve the old version and explain why.
Also state the unit of comparison. A company, branch, sales territory, customer cohort, and individual contract are different units. Mixing them can produce an impressive-looking collection that answers no coherent question. Precision about the unit matters more than a polished chart.
2. Build a comparison group that can withstand challenge
List the characteristics that should affect the outcome: business model, starting scale, customer access, implementation demands, regulation, and relevant timing. Choose characteristics for a causal or operational reason, not because they make the forecast look attractive.
Include unsuccessful and incomplete attempts when they meet the criteria. A group built only from published successes cannot describe the full experience of comparable attempts. If failed cases are unavailable, state that limitation directly rather than silently treating visible cases as representative.
Use internal history where definitions are consistent and the records are accessible. External cases may widen perspective, but differences in reporting can restrict what they support. Identify the source and definition for each case that materially affects the conclusion.
A useful challenge is: would this case remain in the group if its outcome were disappointing? If the answer changes with the result, the inclusion rule needs work.
Do not insist on a perfect comparison group before making any decision. Herbert Simon’s work, discussed in the bounded rationality literature, starts from limits on information and computation. A defensible imperfect comparison can be useful when its limits remain visible.
3. Show the distribution and the missing information
Avoid reducing every comparison to one average. Where the data permits, show the spread of outcomes, the number of cases, and the conditions associated with different results. A central estimate alone can conceal a range that matters to the commitment.
If there are too few comparable cases for a meaningful statistical estimate, use them as structured examples. Call them examples. Do not dress a small, selectively available collection as a reliable probability distribution.
Keep unknowns separate from zeroes. An unreported implementation cost is not evidence that implementation was free. An unavailable failure outcome is not a successful outcome. Explain how missing observations could affect the interpretation.
The comparison memo should distinguish verified observations, reasonable estimates, and unresolved gaps. That distinction lets an executive decide whether additional research could change the action.
For example, incomplete timing data may be acceptable when authorizing a small discovery phase. The same gap may be unacceptable before promising a firm delivery date to a major customer. Data quality is a decision requirement, not merely an appendix topic.
4. Make company-specific adjustments earn their place
After presenting the comparison, invite the team to explain why the current case should differ. Require an observable mechanism for each adjustment. A stronger distribution agreement is a mechanism if its terms, capacity, and relevant customer access are confirmed. General confidence in the team is not a quantified adjustment.
Separate differences already reflected in the comparison group from new information. Otherwise, the same advantage may influence both case selection and the final adjustment.
Write each proposed adjustment in plain language: the difference, the evidence, the expected direction, and the uncertainty. Where the magnitude cannot be justified, describe the difference without attaching a precise numerical effect.
This is the important distinction between a base rate and a target. The base rate describes comparable experience. The target describes an intended result. Ambition can justify investment and effort without turning the desired outcome into the most likely forecast.
Connect this work to executive calibration. Confidence becomes more useful when people can explain what evidence would change it. Neither certainty nor habitual pessimism should receive special status in the review.
5. Translate the forecast into a commitment rule
Ask what the company will do under plausible favorable, central, and unfavorable conditions. Keep those scenarios tied to the comparison evidence. Do not create a favorable case merely to make the range look balanced.
Specify which commitments can wait and which must occur now. A forecast may support a bounded test while providing too little support for a full rollout. The decision can preserve opportunity without pretending the uncertainty has disappeared.
Assign an owner to the evidence that would justify the next commitment. State when the review happens and what result would change the action. An observation date alone is insufficient if nobody knows what the observation will decide.
The proposed meeting script is: “Show the comparable experience first. Then show which differences we can verify. Finally, tell us what we can commit before learning more.” These are suggested words, not a research quotation.
Use decision speed and accuracy as a separate design question. An expiring opportunity may require action with incomplete information. Record that constraint openly. Urgency does not improve the forecast’s evidentiary basis, but it may change the acceptable decision.
6. Review the forecast without rewriting its history
Preserve the original forecast, assumptions, comparison criteria, and approval date. When the outcome becomes observable, compare it with what was knowable at the time. Keep later discoveries out of the reconstruction of the original information set.
Baron and Hershey’s 1988 experiments examined how outcome knowledge affected judgments of decision quality. Their findings provide a reason to distinguish evaluation of the process from satisfaction with the result. They do not imply that results should be ignored.
Ask which assumptions failed, which comparisons proved useful, and which company-specific differences mattered. Then decide whether to change the forecasting process, the operating plan, or both.
Do not reward someone solely for a favorable miss. If a forecast was unsupported but the result happened to be good, the process still deserves examination. Likewise, an unfavorable result does not by itself establish that the forecast was unreasonable.
Use the review to improve the next comparison group. Preserve unsuccessful attempts and changed definitions. Without that record, the organization can repeatedly relearn the same lesson while retaining only its most convenient success stories.
Handle weak comparisons honestly
Some decisions involve a new market, unfamiliar regulation, or technology without a close precedent. The outside view then becomes a set of imperfect comparisons rather than a single authoritative answer. Show which component each comparison informs.
One set of cases might inform implementation timing, while another informs customer acquisition effort. Explain why combining them does not establish a joint probability for the entire business plan. Dependencies between assumptions remain a separate problem.
When evidence is thin, reduce the size of the first commitment where practical. Commission research that can resolve an important uncertainty. If the company must proceed at scale, make the exposure and reasoning explicit.
The goal is an accountable judgment under uncertainty. A comparison should make the reasoning easier to inspect, including its weaknesses. It should not become a technical phrase used to silence people who understand relevant differences.
A compact forecast review agenda
Bring the decision owner, operating owner, and someone able to challenge the comparison. Ask for the defined outcome, inclusion rules, observed cases, missing information, proposed adjustments, and next commitment.
End with a recorded decision and the evidence that could reopen it. Give unresolved questions an owner or state that the company accepts the uncertainty. Do not let an unanswered research request function as an indefinite veto.
A base rate earns its place when it changes the quality of the discussion. It makes the company explain why its plan deserves a different result, while preserving room for evidence that the difference is real.