We have read about sixty data and AI strategy documents from mid-market and mid-public-sector organisations. Most of them are interchangeable. Many of them are from reputable firms. Almost none of them are what a strategy is supposed to be.
A strategy is a set of choices. "We will do these things and not those things, because of this logic, and here is what we expect to change as a result." Most strategy documents are a tour of current-state plus an inventory of future possibility, with the choice part missing.
The twelve symptoms of a strategy that will not land
- A vision statement longer than two sentences
- A maturity model with five levels and no named current level
- A "digital transformation journey" diagram with curves
- A capability map that lists all capabilities without ranking any
- A target operating model with thirty roles and no named incumbents
- A technology section that says "we will adopt cloud" in 2026
- A roadmap shaped like three horizons that lists everything in all three
- A business case with a TAM for the organisation's own internal use
- "Data-driven" more than twice per page
- No mention of the things the organisation will stop doing
- No mention of the specific decisions it will support
- No named owner and no named reviewer
If a document has more than half of these, it is a procurement artifact, not a strategy. It was written to justify a spend, not to change behaviour.
What a strategy is supposed to say
Three things, as a minimum:
1. The handful of decisions this strategy will improve
Not "better decisions". A specific list. "We will be able to decide which properties to prioritise for damp-and-mould inspection." "We will be able to defend our pricing mix by broker." "We will be able to produce our TCFD disclosure without a 12-week manual exercise." Three to eight decisions. Measurable improvement.
2. The things we will stop doing or not do
Strategy requires a no. "We will stop maintaining the QlikView estate by Q3." "We will not build an internal ML platform; we will use the cloud provider's." "We will decline business cases for AI without an evaluation harness." A strategy with no "no"s is a wishlist.
3. The logic connecting the two
Why those decisions, why not others, what capability investment is needed, who carries it, and what is the measurable change six and twelve months in. This should fit on two pages with honest caveats. If it takes forty pages, it is hiding from the choice.
Why the genre has drifted
Two commercial reasons. The consultancy that writes it is paid by the slide, not by the change in behaviour. And the exec who commissioned it wants evidence of thoroughness, not a two-page provocation. The incentives on both sides pull the document toward bulk.
The organisations that get better strategy are the ones that insist, up front: we will pay for a short strategy and a detailed implementation plan, not the other way around.
What we write instead
Our strategy deliverable is three documents:
- The strategy note (8–15 pages). Choices, logic, named decisions, named no's, measurable outcomes at 6 and 12 months.
- The implementation plan (20–40 pages). Dependencies, sequencing, named owners, named risks, named controls.
- The dashboard. What will be visible to the exec team monthly, for the duration.
Everything else — capability maps, architecture diagrams, vendor comparisons — is an appendix. Useful, not essential.
How to read a strategy someone else wrote
If it is handed to you, ask three questions:
- Name three decisions this strategy will change. If the author cannot, the strategy is decoration.
- Name three things we will stop doing because of this. If the answer is none, it is a wishlist.
- Six months from now, what will be measurably different? If no one can answer, do not fund the programme on the strength of this document.
The shorter version
If you have written a strategy and nobody quotes it six months later, it was not a strategy. It was a document. Write fewer of those.