Direct answer
Enough evidence means enough for this next commitment
Do not wait for certainty. Ask whether the evidence is strong enough for the consequence and reversibility of the action you are about to fund. A small, reversible learning test can be reasonable while uncertainty remains. A large, difficult-to-reverse rollout needs stronger, more independent and more decision-relevant evidence.
Hard stop: if one material unresolved condition could reverse a large or hard-to-undo decision, do not authorize that commitment. Do not average the blocker away.
Approval consequence: Budget turns an analytical judgment into an organisational commitment. The evidence does not need to remove every uncertainty; it needs to justify this tranche, show what remains unresolved, and make the hold, reversal or next-evidence condition clear enough that the next stakeholder can defend why the commitment was proportionate.
One evidence set, three different decisions
A €5k pilot can be justified when a €150k rollout is not
Consider a fictional company evaluating a search and content expansion. It has credible search-demand evidence, competitor evidence, technical feasibility work, some customer and commercial evidence, one unresolved buyer assumption and one contradictory signal. The amounts below are illustrative only; they are not financial recommendations.
€5k research / pilot
Potentially supportable. The action is bounded and reversible if the pilot is designed to resolve the buyer assumption or contradiction that controls the next decision.
€30k production rollout
Conditional. The pilot must produce decision-relevant evidence, and the material contradiction must be resolved or shown not to control this tranche.
€150k broader expansion
Not yet supportable. The same exploratory evidence is insufficient if a buyer assumption or material contradiction could still reverse the scale decision.
Memorable rule: the evidence set does not become “good” or “bad” in the abstract. Its sufficiency changes when the commitment changes.
Why evidence sufficiency is decision-relative
A universal rule such as “three sources are enough,” “five interviews are enough,” or “ten competitors are enough” fails because it ignores what the decision can cost, how difficult it is to undo, and what uncertainty still matters. Evidence sufficiency is a permission boundary: what action does this evidence support now, and what larger action does it not support?
Formal appraisal and evaluation guidance uses the same underlying discipline in different language: analytical effort and scrutiny should be proportionate to cost, risk, uncertainty and the decision being made; staged evaluation can create learning before wider rollout; and additional information has value only when it can improve the decision. Those principles do not create a universal commercial budget rule. They support a decision-relative one.
Reversible decision
- Exposure is bounded.
- Rollback is practical.
- The action can generate evidence.
- Residual uncertainty may be acceptable.
Hard-to-reverse decision
- Downside or disruption is material.
- Rollback is costly, slow or incomplete.
- Errors can lock in resources or reputation.
- Material unknowns require stronger resolution.
The conditions that actually move the evidence threshold
You do not need a weighted score. You need to test a small set of conditions in a strict order. One material blocker should beat a pile of favorable but non-binding signals.
- Define the next commitment.Name the action being funded now—not the whole strategy. If the decision is not defined, use the decision-definition method first.
- Test consequence and reversibility.Ask what happens if you are wrong and how difficult, expensive or disruptive reversal would be.
- Check decision scope.Evidence must apply to the right buyer, offer, market and geography. If those boundaries are unclear, resolve them using the four-boundary method.
- Inventory relevant evidence.Use evidence that actually bears on the decision. Reliability, provenance and independence are handled in the reliable-evidence guide.
- Test independence, freshness and first-party needs.Five documents can still be one signal if they trace back to the same dataset. Where a commercial claim requires your own customer or sales evidence, proxy evidence does not silently replace it.
- Find material contradiction and material unknowns.A material unresolved conflict is different from missing evidence. Use the contradictory-evidence method to resolve the conflict; use the evidence-state method to label what remains unresolved.
- Ask whether more information could change the action.If the answer cannot change what you are allowed to do next, more research has little decision value for this gate.
- Choose the maximum supported action.Commit, run a smaller test, gather specific evidence, resolve scope/conflict, hold, or reopen a prior decision.
- Attach review and reversal conditions.Every funded tranche should say when evidence will be reviewed and what condition would stop, reverse or reopen it.
Non-negotiable blocker rule
Do not let an average hide the one thing that can reverse the decision
If one material unresolved condition could reverse a large or difficult-to-reverse commitment, do not commit yet. A readiness score, weighted average or confidence percentage can make ten favorable signals look stronger than one fatal dependency. The correct treatment is precedence, not averaging.
This does not mean every unknown blocks action. A non-material unknown can remain. The distinction is whether that unknown could reasonably change the permitted action.
When to stop researching—and when to keep going
Stop gathering more evidence when…
- The next action is sufficiently bounded.
- Remaining uncertainty cannot change the permitted action.
- A safe real-world test can resolve the uncertainty more directly.
- The expected decision value of more research is lower than the cost or delay it creates.
Continue research when…
- A material unknown could reverse the commitment.
- A material contradiction remains unresolved.
- Evidence is outside the buyer/offer/market/geography scope.
- Commercially necessary first-party validation is absent.
- The decision is hard to reverse and the gap is material.
Plain-language value-of-information question: “Would learning this actually change what we do?” If not, another report may improve comfort without improving the decision.
Evidence Sufficiency Commit Gate
How to read it: select a checkpoint to isolate that part of the method. The canvas shows the relationship; the inspector explains what the checkpoint changes in the decision.
Guided decision product
Evidence Sufficiency Check
Use the quick check to test the next commitment. It uses blocker logic rather than a score. Your inputs stay in this browser; the tool has no external API and does not send company names, budget values, evidence notes or customer data anywhere.
Quick evidence check
Decision boundary
- Why
- What the evidence supports
- What it does not support
- Maximum supported action
- Material unknown
- Next evidence
- Possible test
- Review trigger
- Reversal condition
- Record completeness
Worked example: what changes from pilot to scale?
| Decision gate | Evidence state | What it supports | What it does not support | Review / reversal control |
|---|---|---|---|---|
| €5k pilot | Search/competitive/technical evidence is credible; buyer assumption unresolved; contradictory signal exists. | A bounded test that directly resolves the controlling buyer assumption or conflict. | Production scale or multi-market rollout. | Predefine learning objective and stop condition; review before production. |
| €30k rollout | Pilot resolves the buyer assumption; contradiction is resolved or shown non-material; first-party evidence is present. | The defined production tranche. | Automatic permission for broader expansion. | Review transferability, commercial evidence and operational rollback before next tranche. |
| €150k expansion | Scope-specific evidence is strong and sufficiently independent; required first-party validation is present; no material blocker remains. | The defined expansion only, with explicit review/reversal conditions. | Any future geography/offer/buyer extension not covered by the evidence. | Reopen if scope changes, evidence decays, a new contradiction appears or reversal criteria are met. |
The important point is not the fictional amounts. It is the change in exposure. Each larger or harder-to-reverse tranche earns a new evidence threshold.
A correct result is not the same as a complete record
The logic can legitimately say that a small reversible test is supportable while the documentation record is incomplete. Keep three states separate:
- Logical decision result: what action the evidence supports.
- Record completeness: whether owner, evidence, review and reversal details are fully documented.
- Publication / governance readiness: an administrative state handled outside this public tool.
That separation prevents an incomplete worksheet from looking like a failed decision—and prevents a plausible decision result from looking fully governed when required documentation is missing.
Take the decision into the budget meeting
The workbook records the commitment, evidence, sufficiency review, decision record, worked example and canonical fixtures. The PDF is a shorter field guide for a meeting or handoff. They complement the article rather than duplicating it.
Six rules worth carrying forward
Evidence sufficiency is relative to the next decision.
A reversible test can require less evidence than a hard-to-undo rollout.
More sources do not necessarily mean more independent evidence.
More research has value only if it can change the next action.
One material unknown should not be averaged away.
The same evidence can support a pilot and still be insufficient for scale.
Frequently asked questions
How much evidence is enough before committing budget?
Enough evidence is the evidence that is sufficient for the consequence and reversibility of the next commitment. A small reversible test can be justified with more uncertainty than a large hard-to-reverse rollout, but one material unresolved condition that could reverse the decision should block the larger commitment.
When should you stop gathering evidence?
Stop when the next action is bounded and additional information cannot materially change the permitted action, or when a safe real-world test can answer the controlling uncertainty more effectively than more research.
How does reversibility affect evidence requirements?
The harder or more expensive a decision is to undo, the stronger and more decision-relevant the evidence should be before committing. Reversible actions preserve options and can tolerate more residual uncertainty.
Does a small test need the same evidence as a full rollout?
No. A small test can require less evidence if its downside is bounded and the test is designed to answer the uncertainty that controls the next decision.
How many sources are enough?
There is no universal source count. Several documents may repeat one underlying source, while one strong first-party observation can be highly relevant to a specific decision. Judge relevance, independence, provenance, scope and material unknowns.
What if the evidence conflicts?
If the conflict is material and unresolved, do not authorize a large or hard-to-reverse commitment. Resolve the contradiction or run a bounded test only if that test can address the conflict safely.
What if evidence is incomplete?
Incomplete evidence does not automatically mean stop. Ask whether the missing evidence is material to the next action. Non-material gaps may coexist with a reversible test; material gaps can block scaling.
When should you run a pilot?
Run a pilot when a smaller, bounded action can test the uncertainty that controls the next decision. Do not pilot reflexively if the pilot cannot answer the question or is itself too risky.
What evidence is needed before scaling?
Scaling requires evidence that applies to the intended scope, enough independent corroboration for the key claim, required first-party validation, no unresolved material contradiction, and no material unknown that could reverse the scale decision.
What is a staged commitment?
A staged commitment releases resources in bounded tranches. Each tranche has its own evidence threshold, review point and reversal or stop condition before the next tranche is authorized.
What is residual uncertainty?
Residual uncertainty is what important uncertainty remains after reviewing the available evidence. It does not need to disappear; it needs to be tolerable for the next bounded action.
When is more research a waste?
More research is wasteful for the current decision when it cannot change the permitted action, when it repeats existing provenance, or when a safe test can resolve the controlling uncertainty faster and more directly.
Research basis and evidence limits
This method is a SERPIFAI synthesis of decision, appraisal, evaluation, evidence, uncertainty, staged-investment and experimentation principles. The sources below support specific load-bearing principles; they do not endorse SERPIFAI or establish a universal commercial evidence threshold.
- HM Treasury: The Green Book (2026)
Supports: Evidence should inform appraisal; uncertainty should be explicit; scrutiny and analytical technique should be proportionate to risk/cost; decision trees and real options can help with sequential or hard-to-reverse decisions.
Does not establish: Does not define a universal commercial evidence threshold or endorse SERPIFAI.
- HM Treasury / Evaluation Task Force: Magenta Book: Central Government guidance on evaluation
Supports: Evaluation should be planned before, during and after implementation; pilots and early tests can investigate risks, uncertainty, implementation and outcomes.
Does not establish: Does not say every uncertain decision should be piloted.
- Evaluation Task Force: Test and Learn
Supports: Test and Learn can generate robust learning before wider rollout and should be prioritized where learning can improve decisions.
Does not establish: Does not justify a pilot whose design cannot address the controlling uncertainty.
- UK Government: Value for money study: mega-project governance and budgeting
Supports: Stage gates should be real decision points and additional funding may or may not be released; robust cost/risk evidence is required before final investment decisions.
Does not establish: Applies to mega-project governance, not routine marketing spend.
- U.S. GAO: Cost Estimating and Assessment Guide
Supports: Reliable estimates require defined scope, assumptions, data, sensitivity/risk analysis, documentation, cross-checking and updates with actuals.
Does not establish: Does not define marketing evidence sufficiency.
- NIST: Assessment of Conformity, Decision Rules and Risk Analysis
Supports: Decision rules should account for uncertainty and the asymmetric economic consequences of false acceptance/rejection.
Does not establish: Metrology context; not a commercial investment rule.
- World Bank: Investment Decision Making Under Deep Uncertainty
Supports: Under deep uncertainty, robust/flexible decision methods and real options can be preferable to pretending uncertainty can be eliminated.
Does not establish: Climate/infrastructure focus; transfer requires caution.
- National Academies / NCBI: Uncertainty and the Value of Information
Supports: Information has value when it improves decisions; collecting information should be judged against the value of improved decisions.
Does not establish: Older and context-specific; conceptual transfer only.
- ISPOR Task Force / PMC: Value of Information Analytical Methods
Supports: VOI addresses whether reducing uncertainty through new evidence can improve resource decisions.
Does not establish: Healthcare framework; this guide uses the concept qualitatively.
- AHRQ: Assessing Applicability of Studies
Supports: Applicability should be assessed explicitly relative to the target population/context.
Does not establish: Healthcare framing; maps conceptually to buyer/offer/market/geography.
- GRADEpro: Introduction to Evidence-to-Decision frameworks
Supports: Structured decisions should consider benefits/harms, resources, values, equity, acceptability and feasibility rather than one evidence metric.
Does not establish: Healthcare framework; not directly adopted.
- Microsoft Azure: Architecture strategies for safe deployment practices
Supports: Progressive exposure/canary rollout reduces impact by exposing changes to smaller groups before broad rollout.
Does not establish: Technical deployment context; not a marketing-budget rule.
Method boundary: the tool does not estimate returns, recommend a monetary budget, calculate confidence, or replace finance, legal, regulatory, technical or domain-specific approval requirements.