An AI idea raised in a workshop is a sentence. A use case the board can fund is the same sentence with ten questions answered. The ten are chosen so that any two opportunities, in any function, can be compared like for like, and so that the answers drop into whatever investment or technology-assessment process the business already runs without rework.

Here they are, with what each is really asking.

1. Priority. High, medium or low, as the people who raised the idea judge it, calibrated against the other ideas in the room. Not an external verdict.

2. Strategic fit. How directly does the opportunity serve what the business has said it is trying to do this year? An idea can be clever and irrelevant.

3. Technical feasibility. Does the capability exist and is it proven, and can it be made to work in this environment? Most things are feasible in the abstract; the question is feasibility here.

4. Complexity. The integration, change and dependency load, from “switch on a licence” to “multi-system programme touching three functions”. Complexity is the honest counterweight to enthusiasm.

5. Estimated cost. A three-band scale, £ / ££ / £££, relative and business-calibrated. Precise figures at this stage are false precision; the band is enough to prioritise, and it is firmed up at business-case stage.

6. Estimated benefit. The scale and reach of the value, including the “one licence, many benefits” effect when a capability is consolidated across functions. A productivity assistant raised by five functions is one use case with five times the benefit, not five thin ones.

7. Risks. Safety, compliance, data protection, reputational and delivery risks, with the mitigations each implies. A customer-facing assistant carries a transparency duty; a safety-related model needs validation before it is relied on.

8. Data position. Low dependency, in hand, siloed, partial or absent. The element that most often decides sequencing, and the one a workshop assesses from what participants report and a data audit then measures.

9. Vendor options. Candidate products from a landscape of over a hundred vendors across twelve functions. Candidates, not selections; vendor choice belongs in the business’s own assessment, and the shortlist is the start of a market scan, not the end.

10. Recommendation. Do now, prepare, research, or park, with the adoption pathway and the Opportunity Pyramid layer attached, so the recommendation says not just whether but how and when.

Why a straw man

The scoring is done before the convergent workshop, on every cluster of ideas, and deliberately offered as a straw man for the room to challenge, correct and force-rank. This is the step that makes the output credible. A blank sheet produces polite, uncalibrated scores; a straw man produces argument, and the scores that survive argument are ones the business believes, because it set them.

Mapping to your process

Most businesses of any size already have a template: a general description, strategic fit, a value tree, a total-cost-of-ownership breakdown, an evaluation framework and a prioritisation matrix. The ten elements map onto that structure so each consolidated use case can be lifted into it without rework. Where the template asks for detail not fixed in the workshop, a full cost breakdown or named accountable owners, the field is flagged for the use-case owner rather than guessed.

We also recommend four governance fields that most templates lack and the EU AI Act makes useful: risk category, data protection, human oversight and transparency.

The scorecard has its own page under our approach, and it is applied in every discovery programme.