Kepner-Tregoe Decision Analysis

Make a consequential choice traceable by defining what every acceptable option must satisfy, what outcomes are preferred and what could go wrong.

In one minute

Kepner-Tregoe Decision Analysis separates three questions that teams often mix:

1. Musts: Which conditions are mandatory? An option that fails a Must is not eligible.
2. Wants: Which desirable outcomes distinguish the eligible options? Wants receive explicit weights and evidence-based ratings.
3. Adverse consequences: What material risk or disadvantage follows each leading option?

The method does not calculate an automatic answer. It creates a transparent comparison and exposes where the decision depends on weak evidence, judgement or risk appetite.

Best for: choices among a small set of defined alternatives with meaningful constraints and consequences.
Avoid when: the real problem is still unclear, no credible alternatives exist or an emergency requires an immediate reversible action.

The problem it addresses

Decision meetings commonly begin with a preferred solution. Requirements are invented to support it, attractive features compensate for non-negotiable defects and a weighted score creates false precision.

Kepner-Tregoe Decision Analysis forces the team to define the decision statement, Musts and Wants before scoring. Risk is considered after the benefit comparison so a high score cannot conceal an unacceptable consequence.

When to use it

Use the method when:

• several credible alternatives compete for one decision;
• mandatory legal, security, cost or timing conditions exist;
• stakeholders value different desirable outcomes;
• an AI vendor, model or autonomy level must be selected;
• the choice needs an auditable rationale;
• adverse consequences could overturn the apparent favourite.

Keep the alternatives at the same level. Do not compare “build an internal assistant” with “improve customer satisfaction.”

When not to use it

Do not use it:

• to diagnose why a failure occurred;
• when the decision statement hides several separate choices;
• to turn uncertain estimates into precise-looking facts;
• after leaders have made an irreversible choice but want retrospective justification;
• when one Must already determines the answer;
• without naming who owns the final decision.

Use Root Cause Analysis for causal diagnosis. Use Delphi Method when important estimates require structured expert judgement.

Inputs required

• a decision statement with action, object and scope;
• decision owner and affected stakeholders;
• a feasible alternative set, including the status quo where relevant;
• mandatory constraints and their verification evidence;
• desirable outcomes and relative importance;
• comparable evidence for rating alternatives;
• material risks, assumptions and uncertainty;
• a decision date and review trigger.

Step-by-step process

1. Write the decision statement

Use a specific form such as: “Select the operating model for triaging standard support requests in the European service team for the next twelve months.”

Do not include the preferred solution in the statement.

2. Confirm the decision owner and participants

Record who decides, who supplies evidence, who must be consulted and who implements. Decision ownership should remain visible even when the comparison is collaborative.

3. Generate credible alternatives

Define options at a comparable level. Include “do nothing for now” when delay is genuinely feasible. Remove duplicates and describe each option sufficiently for evidence collection.

4. Define and test Musts

Write each Must as a measurable pass/fail condition. Examples:

• customer data remains in an approved region;
• the workflow supports human reversal within fifteen minutes;
• twelve-month cost does not exceed the approved ceiling;
• the option can be piloted before the regulatory deadline.

State the evidence required for a pass. “Secure” is not a testable Must.

5. Define and weight Wants

Wants describe preferred outcomes among eligible options. Remove overlaps, define the measurement direction and assign relative importance before rating alternatives.

Examples include reduction in handling time, implementation effort, exception visibility, maintainability and evidence quality.

6. Eliminate options that fail a Must

Record the failed condition and evidence. Do not rescue an ineligible option by giving it a high Want score. If a Must turns out to be negotiable, relabel it openly and obtain the decision owner’s approval.

7. Rate eligible alternatives

Use one consistent scale. Attach an evidence note and confidence level to each important rating. Calculate weighted totals as a comparison aid, not a verdict.

8. Examine adverse consequences

For the leading options, identify plausible harm or disadvantage, probability evidence, impact and available mitigation. Consider concentration risk, exit cost, data exposure, user contestability and downstream workload.

9. Make and document the decision

Record the selected option, reasons, rejected alternatives, decisive assumptions, dissent and the conditions that would trigger review.

10. Verify after implementation

Compare realised outcomes and consequences with the original ratings. Use the result to improve future criteria and evidence practices.

AI automation lens

For an AI-enabled workflow, alternatives should often include different levels of autonomy, not only different models or vendors:

• human work with better information;
• AI draft with human approval;
• automated action for a narrow valid case;
• automation with abstention and an exception queue;
• no deployment until evidence improves.

Useful Musts cover data permission, auditability, reversal, accountable approval and prohibited outcomes. Wants may cover task quality, cycle time, reviewer effort, maintainability and cost.

AI may prepare evidence, check arithmetic and flag inconsistent ratings. It should not invent missing evidence, redefine Musts after seeing the scores or make the final accountable choice.

Visual model

Text alternative: a decision statement leads to comparable alternatives. Mandatory requirements filter out ineligible options. Eligible options are compared on weighted desirable outcomes, then leading choices are checked for adverse consequences before an accountable decision and review trigger.

Interactive example

Scenario

Northstar Support must choose a triage operating model for 3,000 weekly tickets:

• A: retain manual triage and improve forms;
• B: use an AI classifier in draft mode with agent confirmation;
• C: enable autonomous routing for all tickets;
• D: enable autonomous routing only for three standard categories and route uncertainty to agents.

Mandatory conditions are approved data residency, reversal within five minutes and urgent-case sensitivity of at least 98% in a controlled test.

Your move

Define four Wants, compare the eligible options and identify one adverse consequence that could change the decision.

Worked answer

Option C fails the urgent-case Must and is removed. The team weights correct first routing, queue-time reduction, reviewer effort and maintainability. Option D leads on benefit, while B has a lower benefit score but simpler operational recovery.

The adverse-consequence review shows that D could create a hidden queue of abstained cases during demand peaks. The decision is to pilot D for the three standard categories only, with an exception-age limit and automatic return to manual routing. Scaling depends on urgent-case sensitivity, exception age, agent correction and customer impact—not model accuracy alone.

Facilitation notes

• Agree the decision statement before discussing criteria.
• Ask whether every Must is truly non-negotiable and testable.
• Define weights before alternatives are rated.
• Show evidence and confidence beside important scores.
• Keep benefit comparison separate from adverse consequences.
• Record dissent instead of forcing artificial consensus.
• Set a review trigger for assumptions likely to change.

Expected output

A sound application produces:

• a specific decision statement and owner;
• comparable alternatives;
• testable Musts with pass/fail evidence;
• weighted Wants with clear definitions;
• an evidence-linked comparison;
• adverse consequences and mitigations;
• a documented choice and review trigger.

Common mistakes

1. Writing the preferred option into the decision statement. The analysis begins biased.
2. Using vague Musts. “Easy” or “safe” cannot be tested consistently.
3. Letting Wants compensate for failed Musts. Mandatory means ineligible if failed.
4. Weighting after seeing scores. Importance is adjusted to favour an option.
5. Treating the total as the decision. Evidence quality and adverse consequences still require judgement.
6. Scoring unknowns as average. Missing evidence becomes a hidden advantage.
7. Forgetting the status quo. The cost and risk of delay remain invisible.

Quality checklist

• The decision statement names one choice and scope.
• The decision owner is explicit.
• Alternatives are credible and comparable.
• Musts are measurable pass/fail conditions.
• Wants are distinct and weighted before scoring.
• Important ratings cite evidence and confidence.
• Failed Musts cannot be offset by Want scores.
• Adverse consequences are reviewed separately.
• The decision includes assumptions and a review trigger.

Template

Decision statement:

Decision owner and date:

Decision, rationale and review trigger:

Knowledge check

An option fails a mandatory data-residency requirement but has the highest weighted Want score. What should the team do?

A. Select it because the total score is highest.
B. Increase the weight of data residency.
C. Remove it from the eligible set unless the decision owner explicitly changes the requirement.
D. Replace the failed Must with an average rating.

Answer: C. A failed Must cannot be compensated by desirable performance elsewhere.

Related tools

• Inputs from: Delphi Method, Pairwise Comparison
• Risk-checked with: FMEA, Process Decision Program Chart
• Implementation through: Objectives and Key Results, PDCA/PDSA Cycle
• Not to be confused with: a generic weighted spreadsheet or an automated recommendation engine

References

1. Kepner, C. H., and Tregoe, B. B. The Rational Manager. McGraw-Hill, 1965.
2. Kepner-Tregoe. “Introduction to Decision Analysis.” Official training overview (opens in a new tab).
3. Kepner-Tregoe. “The Consequences of Choice: The Final Step in Decision Making.” Official practitioner article (opens in a new tab).
4. National Institute of Standards and Technology. Artificial Intelligence Risk Management Framework (AI RMF 1.0). NIST AI 100-1, 2023. Official publication (opens in a new tab).
5. National Institute of Standards and Technology. Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile. NIST AI 600-1, 2024. Official publication (opens in a new tab).

Sources reviewed 12 August 2026.