AI strategy that names three things to build, not thirty to consider

Most AI strategy work produces a maturity score, a long list of use cases and a roadmap nobody can act on. This produces a short, ordered list of specific changes, each with an owner, an estimate and a way to tell afterwards whether it worked.

  • A ranked shortlist. Three to five candidates, ordered by value against what your organisation can actually absorb this quarter.
  • A number on each. What it would take to build, what it would cost to run, and what it would save or unlock — with the assumptions written down and challengeable.
  • A measurable definition of success. What will be different, stated so that in six months nobody has to argue about whether it worked.
  • The honest exclusions. What was considered and rejected, and why. Usually the most useful section, because it stops the same ideas coming back every quarter.
Talk to us about this

What is delivered

  • Decision map

    Where the decisions that matter are actually made, who makes them, what they use, and how long each takes today. This is the artefact everything else derives from.

  • Candidate assessment

    Each candidate scored on value, feasibility, data readiness and change cost — with the reasoning visible rather than a score in a table.

  • Sequenced plan

    What to do first and why, sized against the organisation's real capacity to absorb change rather than against an ideal roadmap.

  • Data and platform read

    What your data can currently support, what would need to change, and which platform choices follow from your existing commitments and constraints.

  • Risk and regulatory read

    What the EU AI Act and data protection obligations mean for each candidate, in specifics rather than as a compliance appendix.

  • The case against

    The strongest argument for doing nothing, or for solving the problem another way. If we cannot make it, the recommendation is not tested.

How it runs

  1. 01

    Interviews

    The people who make the decisions and the people who execute them, separately. The gap between those two accounts is usually where the opportunity is.

  2. 02

    Evidence

    Actual volumes, actual cycle times, actual error rates. Perception and measurement disagree more often than not, and it matters which one the plan is built on.

  3. 03

    Shortlist and scoring

    Candidates assessed and ranked in the open, with you in the room, so the reasoning is transferred rather than presented.

  4. 04

    Written up

    A document short enough to be read by everybody who has to act on it, and specific enough to be argued with.

A good fit when

  • You have a budget and a mandate but no agreement on where to start.
  • Previous AI work produced demonstrations that never became systems.
  • Somebody senior has asked for an AI plan and you want it to survive scrutiny.
  • You suspect the real problem may be process or data, and want that tested honestly.

Not the right service when

  • You already know exactly what to build — go straight to engineering.
  • You need a document to justify a decision that has already been made.
  • The organisation has no capacity to change anything for the next two quarters. Fix that first.

Frequently asked questions

How long does it take?
Typically two to four weeks depending on how many parts of the business are in scope. Longer than that and it stops being framing and starts being a programme, which is a different thing that should be bought differently.
What does it cost?
A fixed price agreed before it starts, scaled to scope rather than billed by the hour. Framing paid by the day has an incentive nobody should have to manage around.
Do we have to use you for the build?
No, and the document is written on that assumption. It should be specific enough for another supplier or your own team to execute. A strategy that only works if its author builds it is a sales document.
What if the conclusion is not to use AI?
Then that is the deliverable, and it is a good outcome. Several engagements have ended with a process fix and a data cleanup that were worth more than any model would have been.
Who needs to be involved from our side?
Whoever decides, whoever executes and whoever owns the data. Two to six people for a few hours each. If the people who execute are not interviewed, the plan will describe a company that does not exist.

Keep reading

Other services