Enterprise AI strategy. A framework to prioritise use cases that pay back.

Why the long list of use cases is the wrong starting point.

When a leadership team asks us for help with their AI strategy, the document we are usually shown first is a workshop output with thirty or forty candidate use cases. The list is well intentioned. It was produced by gathering ideas from across the business, and it represents weeks of effort. The room is quietly proud of it. The problem is that it is not a strategy. It is an inventory of curiosity, and inventories of curiosity do not pay back.

In our work across Paris, Dubai, Singapore and Bali we have come to recognise a consistent failure pattern in these long lists. Budget is allocated thinly, three or four pilots run in parallel under-resourced, the team that should be giving each one its full attention is spread across all of them, and at the end of the year nothing has reached production. The leadership team concludes that AI is harder than expected and trims the budget. The conclusion is fair but the cause is usually not the technology. The cause is the lack of a prioritisation discipline.

The framing we tend to propose is uncomfortable on the way in and useful on the way out. The right number of active AI cases for most organisations we work with is three to five, not thirty. Everything else is on a watch list. Concentration beats coverage at this stage, and the framework below is what we use to defend that concentration when the room pushes back.

Four criteria that cover almost every question worth asking.

Business value at stake.

The first criterion is the size and certainty of the value. We avoid scoring this in points. We ask the case sponsor to defend a quantified prize in monetary terms, anchored to a real business line, defensible to a finance director who has seen many forecasts. A case where the value is large but the assumption chain is fragile scores below a case where the value is moderate and the chain is solid. We are not trying to maximise the headline number. We are trying to maximise the value that survives the first quarter of operation.

Feasibility on data and tooling.

The second criterion is whether the case can actually be built and operated with the data, the systems and the integrations available today. The honest answer is often no, or not yet. A case that requires a data foundation the organisation has not built scores low here, regardless of how attractive the value looks. We have learned to be strict on this. The most expensive AI projects we have seen were the ones that ignored feasibility on the basis that the data problem would resolve itself during delivery. It rarely does.

Risk and compliance exposure.

The third criterion is the regulatory, reputational and operational risk attached to the case. AI Act exposure, personal data handling, fairness considerations, model failure modes that touch customers directly. A case can be highly valuable and feasible and still belong on the watch list because the risk envelope is not yet matched by the governance capability. We tend to make this explicit rather than fudge it. The teams that bury risk in a footnote pay for it later.

Organisational readiness.

The fourth criterion is the readiness of the team that will actually run the case in production. Whether the operators have been consulted, whether the operations function has agreed to take ownership, whether there is a named sponsor with the authority to act when something goes wrong. This is the criterion most often scored too generously, and it is the one we have seen wreck the most pilots. Strong scores on the first three with a weak score here usually means the case ships and then withers.

Scoring as a workshop, not a spreadsheet.

The version of this exercise that works is conversational. The room is small. The case sponsors, the chief data officer or chief technology officer, the compliance lead, and the operator who will live with the result. Each case is presented in ten minutes, scored on the four criteria with a one-paragraph defence per criterion, and discussed.

A scoring spreadsheet on its own tends to produce false precision and political behaviour. The numbers get inflated. The cases sponsored by powerful executives end up high. The cases with thin defence on data quietly disappear into the average. The workshop format forces the defence into the open and makes the disagreements visible while there is still time to act on them.

The deliverable we recommend is short. A two-page summary per case that anchors the four scores in a single argument, plus a portfolio view that shows the three to five cases that will get full funding, the cases on the watch list, and the cases being retired. Anything longer tends to be a hedge against the harder conversation, which is who said no to what.

Funding the top three. Retiring the rest.

The output of the workshop should make the funding decision close to mechanical. The top three or four cases get a full team, a clear sponsor, a budget that runs to production rather than to a pilot, and the operational owner identified before the build begins. The watch list gets a quarterly review, not capacity. The retired cases get a one-page memo explaining why, which is useful both for the people who proposed them and for the next round of ideation.

The discipline of explicit retirement is the part of the framework we see skipped most often, and it is the part with the highest leverage. A use case that has been retired with a clear reason gives the organisation permission to say no the next time something similar appears. A use case that has merely been deprioritised tends to come back six months later with a different sponsor and consume budget again. We have seen organisations lose entire years to this pattern.

The discipline of saying no.

In the leadership teams we work with, the partner who can defend a strong no on a popular use case tends to be more valuable than the one who can defend a long list. Saying no without alienating the sponsor is a craft. It usually rests on the same elements. A clear value comparison, a transparent risk position, and a route back if the conditions change.

If your team is heading into the next planning cycle with a list of twenty AI use cases, the question worth asking before the budget meeting is not which three are most exciting. It is which three are most defensible against all four criteria, and what conversation needs to happen with the sponsors of the rest. The portfolios that pay back tend to be the ones that survived that conversation early rather than late.

The Consulting team in our Paris, Dubai, Singapore and Bali offices is reachable from the brief form below. We answer within one working day, with the partner who will sit on the file.