How to assess a strategic option with AI

Freeze the decision frame

Write a short decision charter containing the decision to be made, accountable owner, time horizon, objective, current-state baseline, and constraints. Distinguish a constraint from a preference: a legal requirement or hard cash ceiling can disqualify an option, while implementation convenience may simply affect its score.

The UK Treasury Green Book treats appraisal as assessing costs, benefits, and risks of different options for achieving objectives. It also requires business as usual as a comparator and emphasizes uncertainty. Although written for public-sector appraisal, those disciplines are useful for structuring a private internal memo; they do not validate a particular commercial choice.

Set criteria before seeing the answer

Create a criteria register with definition, evidence, direction, and importance. Avoid overlapping criteria such as “speed” and “time to value” unless they measure different decisions. Do not let AI infer weights from persuasive prose.

For a synthetic question—whether to build an internal evidence-review tool, buy a specialist product, or continue the current process—the register might be:

| Criterion | Evidence | Importance | Failure boundary | | --- | --- | --- | --- | | Decision coverage | Tested workflow requirements | High | Cannot support required evidence links | | Twelve-month cost | Comparable cost model | Medium | Exceeds approved funding | | Time to controlled use | Delivery plan and dependencies | Medium | Misses operating deadline | | Reversibility | Exit and data-export plan | Medium | Creates unacceptable lock-in | | Operational risk | Security and support review | High | Unresolved critical control gap |

Only after this table is approved should AI normalize evidence and identify gaps.

Compare options without false precision

Write a short evidence card for each option: what it is, what must be true, cost range, capability fit, dependencies, major risks, and missing evidence. Use ranges or ordinal judgments where precise estimates are unavailable. A weighted score can summarize a discussion, but it should not hide decisive constraints or uncertainty.

In the example, “buy” might lead on delivery speed but depend on successful data export and contract review. “Build” may fit the workflow better but depend on scarce engineering capacity. “Continue” may cost least immediately but preserve manual review burden. Those are hypotheses until supported by direct evidence.

Run reversal tests

Ask AI to calculate or narrate at least four challenges:

Then perform a “preference leak” audit: remove option names and brand cues from the evidence cards and rescore them. If the result changes because the labels changed, the process is not decision-ready.

• Which single assumption most strongly drives the recommendation?
• At what cost or delay would the preferred option change?
• What happens in a low-adoption or supplier-exit scenario?
• Does changing a debatable criterion weight reverse the ranking?

Worked memo conclusion

Provisional recommendation: Take the buy option to a four-week verification stage; do not yet authorize full rollout.

Why: It currently meets the timing constraint and covers the core workflow, but export reliability and contract terms remain unresolved. The build option is retained as a fallback if either boundary fails.

Reversal conditions: Prefer build if verified twelve-month vendor cost crosses the agreed ceiling or export testing fails. Prefer current-state continuation if neither option clears the critical security control.

Next evidence: Run the export test, obtain the security review, and replace cost assumptions with comparable quotes.

This conclusion is useful because it exposes what would change the decision.

Sources and claim map

The tool choice, criteria, scores, and conclusion are synthetic. This educational method does not promise a superior strategy, financial return, or validated outcome.

• HM Treasury, The Green Book 2026: supports defining objectives, comparing options including business as usual, considering costs, benefits and risks, and communicating uncertainty.
• HM Treasury, Green Book guidance collection: provides current official appraisal resources and supporting guidance.

Sources: [1] [2]