AI ROI
AI business case: the structure that gets approved
A section-by-section AI business case structure: problem baseline, proposed change, value range, cost model, risk, governance, measurement plan and decision request.
Miguel Torres, Founder, Merjora · Published 14 August 2026 · Updated 3 September 2026
In short
A strong AI business case is eight sections long: the quantified problem baseline, the specific workflow change proposed, the value range with confidence, the full first-year and run-rate cost model, the risk and control position, the governance classification, the measurement plan with a pre-agreed baseline, and a single explicit decision request. Approval usually fails on the baseline and the measurement plan, not on the technology.
Key takeaways
- Lead with the current cost of doing nothing, expressed in euros.
- One decision request per document. Ambiguity gets deferred.
- Include the do-nothing and the non-AI alternative — it strengthens the case rather than weakening it.
- Pre-agree how the result will be measured; that is what makes the case auditable.
Section by section
| Section | Contains | Common failure |
|---|---|---|
| 1. Baseline | Volume, handling time, loaded cost, error rate today | Estimated rather than measured |
| 2. Proposed change | The specific step, the human checkpoint, the boundary | Described at process level, so scope drifts |
| 3. Value | Range, drivers, what converts hours into cash | Single point estimate with no driver breakdown |
| 4. Cost | Build, integration, evaluation, change, run-rate | Review and maintenance omitted |
| 5. Risk | Failure modes, residual error, reversibility | Only technical risk, no operational risk |
| 6. Governance | EU AI Act classification, data protection position | Deferred to 'later', which blocks go-live |
| 7. Measurement | Baseline capture, metrics, review cadence | Added after launch, when the baseline is gone |
| 8. Decision | The exact approval requested and by when | Multiple asks bundled together |
Include the alternatives
Every AI business case should carry two comparators: doing nothing, and solving the same problem without AI — process redesign, a template library, a rules engine or a different system configuration.
If the non-AI option is cheaper and sufficient, that is a good outcome. Documenting it also removes the most common board objection, which is that nobody checked.
Expressing uncertainty without losing the room
Use a low, expected and high case, and say in one sentence what drives the spread — usually exception rate, adoption rate or volume growth. Attach a confidence level to the expected case and state what would raise it.
Frequently asked questions
- How long should an AI business case be?
- Four to six pages plus an appendix. If the baseline and measurement plan are solid, length does not help.
- Who should own the AI business case?
- The operational owner of the workflow, supported by finance for the cost model and by IT for data and integration. Cases owned solely by an innovation function tend not to survive delivery.
Where would this apply in your business?
Merjora maps your workflows, scores each opportunity and quantifies the likely value range before you spend budget.
Discover your AI opportunitiesRelated reading
Editorial standard. Merjora publishes analysis, frameworks and publicly documented examples. We do not publish invented statistics, unattributed benchmarks or unverified customer stories.