Construct an initial investment case and decision gate

**The decision.** Before an initiative competes for portfolio funding, the committee needs to know what the first tranche will buy in knowledge or operating capability, and when spending will stop if the evidence fails to emerge. An initial investment case is neither a complete project plan nor a promised return. It is a disciplined comparison of continued practice and specific change options under resource, time, and dependency constraints, with assumptions visible. Put the question the board can answer now on the first page: “Should we fund a six-week, 50,000-unit trial of reserving spare parts before a visit, or continue current practice?” A request to “approve digital transformation” without a bounded scope or stop rule is not a usable portfolio decision.

**Construct a reviewable case.** Define the baseline option first. What is likely to happen if current arrangements continue during the comparison period? Who experiences its costs, and what remains unknown? Then describe a minimum change option and, where warranted, a broader option. Do not give “do nothing” zero cost and zero lost value just because it needs no new budget: customer waiting and repeat visits may matter. Equally, do not count all current losses as benefits of the proposed technology without evidence that it will prevent them. Separate one-time setup, recurring operation, transition and training, and the opportunity cost of scarce staff taken from another initiative. Where one department pays for a platform and another expects fewer returns, identify a benefits owner and a measurement route. Otherwise the return has no accountable witness.

For each important number, record its unit, time period, origin, scope, and confidence. An accounting average cost per repeat visit may not equal the marginal cash saved if an individual visit is avoided. Technicians may remain employed while their time becomes available for other requests. State whether the benefit is released cash, capacity, customer time, quality, or something else. Where customer time has not been credibly monetized, show it as a separate measured outcome instead of inventing currency value. Avoid double counting: “fewer visits” and “technician hours released” may be two views of one effect. A primary outcome, a harm guardrail, and actual cost can be enough for the first pilot gate.

Make the needed change tangible. If the pilot again receives 300 critical requests, a 60% result requires 180 on-time resolutions, 45 more than the baseline count. Do not label those 45 a causal effect without checking changes in request volume, fault mix, season, and technician capacity. If the observed outcome is 58% and a supplier issue is documented, the team must not redefine 58% as a pass after seeing it. It can consider a tightly bounded follow-up test of the dependency if the extra cost and learning justify it, or stop. If speed reaches 62% but reopening rises to 12%, the quality guardrail fails despite the attractive speed headline. Even if both thresholds pass, missing annual operating cost or an unresolved data-export term means the trial supports a learning decision, not an unrestricted purchase. Record what remains unknown and the specific step that will resolve it.

**What staged funding purchases.** The first tranche buys a better test of the causal mechanism, an integration surface the team can examine, and evidence for a later decision. Name the decision owner, date, evidence pack, and action if evidence is late. The trial leader keeps a dated record of changes to indicator definitions, sampled cases, and district scope. Before paying, the committee decides whether protocol changes during the trial are allowed and when a new baseline would be required. Include an orderly stop route: close the pilot contract, export the data, preserve the learning, and redeploy staff. Stopping after a weak result can be good governance when that possibility was agreed in advance. Continuing automatically because the first tranche has already been spent is not a valid investment rationale.

**Source boundary.** HM Treasury's Green Book 2026 describes comparing options, including continued practice and minimum change, and connecting appraisal with monitoring and evaluation. We use it as a methodological example, not as a worldwide legal rule for this firm. The case's amounts and thresholds are authored training assumptions, not claims drawn from the source. [HM Treasury, The Green Book 2026](https://www.gov.uk/government/publications/the-green-book-appraisal-and-evaluation-in-central-government/the-green-book-2026).

**A gate that also describes failure.** When approving 48,000 for the trial, state what happens if four-hour restoration stays at 45% even though messages and reservations arrive on time. That result weakens the hypothesis that part availability is the binding cause within the chosen scope. Scheduling priority or customer approval may matter more. Do not rewrite the success measure as “the platform was used.” Integration may remain a reusable asset, but a new use requires a separate value claim and maintenance estimate. The report should say what was learned against the current proposal and which new opportunity, if any, remains conditional.

Check payment timing as well. If the supplier contract collects an entire annual license at trial start, do not advertise a stoppable 48,000 tranche when most scale spending has already been committed. Seek contract terms that align commitment with the gate or disclose unavoidable payment. A 48,000 approval also does not eliminate measurement effort. If another two specialist weeks are required to clean and analyze data, include them in capacity and schedule. A minimum-change option might improve fault-code entry for 5,000 before integration; place it fairly in the learning decision tree.

At the decision meeting, show one page with the current request, baseline, options, cash cap, capacity, outcome and harm measures, stop limits, and missing evidence. An annex carries calculations, sources, and approval dates. If the board declines, record what evidence could reopen the decision instead of making the team infer a permanent rejection. If it approves subject to a condition, name who verifies that condition **before** spending rather than afterward. This makes “conditional approval” an auditable operating path rather than a political phrase.