Pilot business-case facilitation guide
Contents
This guide supports a 60–90 minute pilot business-case workshop. For the current audience presentations, use Executive strategic vision and Technical architecture. The Executive deck addresses strategic choices and shared capabilities; this workshop addresses pilot funding and measurement.
Structure for a 60-90 minute session
| Segment | Time | Content | Purpose |
|---|---|---|---|
| 1. The question and the concern | 5 min | The recommended framing sentence; the sponsor’s central question from the questionnaire; spending concerns ranked | Show the session is about their problem, not a product |
| 2. What we heard | 10 min | Questionnaire results in six charts: primary outcome, top concerns, top-three effort activities, automation and regression profile, trusted-metric grid, acceptable autonomy level | Current-state questions rather than conclusions; no savings number yet |
| 3. What the external evidence actually says | 10 min | One slide: task gains exist (vendor and lab), METR’s early-2025 study found a slowdown and its follow-up is inconclusive, surveys and telemetry show quality risks alongside some output gains, consultancy estimates do not establish audited QA savings; one slide: what could not be verified | Credibility; pre-empts “the analysts say” |
| 4. Automation, assistance and agency | 5 min | The autonomy ladder; what is proposed for the pilot (levels 1-5 with human approval) and what is explicitly out (autonomous merge or deploy) | Governance boundary in plain language |
| 5. Where the opportunity is | 10 min | QA: test design, generation, data, environments, execution, maintenance and evidence with the two defaults; the engineering gaps that are not AI problems | Separate foundational automation from AI |
| 6. Conservative savings hypothesis | 10 min | Scenario table in percentages; the base-case waterfall per $10M with assumptions visible; the four-level distinction | Discipline; no dollar forecast |
| 7. The phased model | 10 min | Phases 0-4 with duration, ceiling, deliverables and stop rules; applications proposed | Minimize upfront spend; show early exit |
| 8. Controls and regulatory alignment | 5 min | One-page control mapping | Turn caution into a regulatory strength |
| 9. Supporting architecture note | 3 min | Action authorization now; shared provider routing when scale requires it | Answer the question before it is asked |
| 10. Decisions requested | 5 min | Named sponsor; Finance partner and recognition rule; one or two applications; approved AI platform and pilot project; Phase 0-1 ceiling; second-line observer | Specific asks |
| Discussion | 15-30 min |
The recommended message
See the one-paragraph message in Slide language. The decision requested is whether to fund Phases 0 and 1, not whether to transform QA.
Supporting architecture discussion
Distinguish a model-provider router (model selection, cost allocation and portability) from an action authorization boundary (identity, resource scope, approved tools, expiry and human approval). A new shared model router may be deferred until scale. Action authorization must exist from the first tool-using pilot; existing IAM and CI controls can implement it. An approved enterprise model platform does not automatically provide authorization for downstream actions.
Visuals that support the discussion
Use the questionnaire charts when bank responses exist, the canonical scenario waterfall, the shared-assurance strategy diagram, the payment-API current-to-target workflow and the action-denial sequence. Choose each visual to explain a decision or relationship. Show assumptions and uncertainty alongside any quantity; no invented bank results.