Sensitivity analysis examines how a decision changes when its assumptions change. A switching value identifies the point at which an option loses its advantage or another conclusion becomes preferred under the stated model. Together they make uncertainty actionable by connecting it to a decision boundary.

A modernization recommendation can depend more on one uncertain condition than on many documented details. Identifying that condition helps distinguish useful discovery from collecting information that would leave the choice unchanged.

Begin with comparable alternatives

Retention, stabilization, upgrade, relocation, replacement, outsourcing, and retirement need a common capability scope and time horizon where they are credible options. An alternative that fails an essential requirement should identify that failure explicitly.

Business as usual provides a comparison benchmark, including its costs and consequences. HM Treasury’s 2026 Green Book retains that benchmark in appraisal even when it fails the objectives. Including it does not endorse continued operation under unacceptable conditions.

The comparison should also retain unmonetized effects. An outcome does not cease to matter because the available evidence cannot assign it a defensible price.

Expose the assumption that carries the result

Take a deliberately simplified cost example. Suppose replacement costs 300 units initially, then 40 units per year, while retention costs 100 units per year. Ignoring discounting and all other differences, replacement recovers its initial cost difference after five years because its annual saving is 60 units.

If the capability is needed for only three more years, the same assumptions produce a different cost conclusion. Transition costs, risk, service quality, and timing could change it again. These constructed numbers explain the switching mechanism; they are not application estimates or an investment recommendation.

The important question is which input requires evidence: remaining useful life, actual operating saving, or the cost of making the transition complete.

Vary coherent scenarios

Changing one influential assumption at a time can reveal its direct effect. Combined scenarios can examine conditions that move together, such as a longer transition with a longer coexistence period and delayed savings.

A scenario should remain plausible and internally consistent. Combining every favorable assumption for one option while assigning every adverse assumption to another produces a biased comparison rather than useful sensitivity analysis.

Record the ranges and why they were selected. More decimal places cannot compensate for missing evidence about the underlying effort or benefit.

Scores need the same scrutiny

Multi-criteria comparison can represent factors that do not share a monetary unit. Treasury’s MCDA guidance examines whether rankings change under different scores and weights and calls for transparent evidence and assumptions.

A qualitative high-medium-low scale is ordinal. Converting its labels into numbers does not make them measured probabilities or guarantee meaningful arithmetic between categories.

Test whether reasonable changes in interpretation or weighting reverse the ranking. If they do, the recommendation should explain the value judgment or unresolved evidence on which it rests instead of hiding behind the weighted total.

Turn uncertainty into a next action

A useful verification action targets a fragile assumption: demonstrate a restore, inspect a critical interface, reconcile license usage, or validate the hardest workflow. Its result should have a stated effect on the recommendation.

Some uncertainties can remain without changing the preferred route. Others justify withholding a commitment until the evidence arrives. The decision record should distinguish those cases and name its reversal conditions.

This approach does not claim formal Green Book or MCDA compliance. It supplies an inspectable recommendation whose rationale survives challenge—or changes for a documented reason when a consequential assumption is resolved.