In Short: Fewer Slides, More Sequence
Most data strategies fail identically: a consultancy-grade vision deck, a maturity heat map, no named owners, no quarterly deliverables - shelfware by Easter. The version that survives is smaller and harder: name the decisions the business needs to make better, assess the foundations honestly, commit to a platform, define the operating model, and sequence the work quarter by quarter with owners. Ten pages, enforced, beats fifty slides admired.
Here is that structure, as we build it with clients.
Start From Decisions, Not Data
Inventory the decisions that matter and are currently made badly: pricing reviews on stale spreadsheets, capacity planning by anecdote, churn interventions after the fact. Each becomes a use case with a named beneficiary and a value hypothesis.
This inversion does two jobs: it gives the strategy a business language executives will fund, and it scopes everything downstream - you build the pipelines, models and reports those decisions need, not a platform in the abstract. The strongest strategies we have written fit on one page per decision.
Assess Foundations Without Flattery
Four questions, answered with evidence rather than workshop optimism.
- Quality: do the systems agree, and would anyone notice if they stopped? (Our data quality guide covers the diagnostic)
- Architecture: one governed platform or a decade of exports? Where does the warehouse-lake-lakehouse reality sit versus the diagram?
- Governance: owners, definitions, access - decided and enforced, or aspirational?
- People: who actually builds and stewards, and what does the team need versus have?
The assessment's output is not a maturity score; it is the ranked list of what blocks the named decisions.
Make the Platform Bet, Once
Platform ambiguity taxes every project, so the strategy commits. For Microsoft-centric organisations in 2026 the default is Fabric: one platform, OneLake, medallion layers, Power BI and the AI stack on top - with Azure Databricks alongside where engineering depth warrants it. Name the exceptions explicitly (the ERP that stays, the vendor lake you shortcut) and stop relitigating.
The bet includes money: capacity sizing, licensing and the cost measurement habit belong in the strategy, because platforms fail politically when the bill surprises.
Sequence Ruthlessly, Own Everything
The roadmap that ships shares a shape: quarter one, foundations for the first decision (one domain's pipeline, model, governed access) plus the first visible win; each following quarter, one more decision served end to end, plus one foundation debt retired. AI enters when its domain passes the readiness checklist, not before - AI on unfixed foundations is how pilots fail.
Every line has an owner and a review: quarterly for delivery, annually for direction. And the operating model says who does this work - the internal team you grow, the partner you use deliberately, the managed capacity in between - because strategies without hands are wishes.
That is the whole method: decisions, honest foundations, one platform bet, named owners, quarterly proof. Everything else is formatting.
Sources and Further Reading
Frequently asked
Five things, briefly: the business decisions the strategy exists to improve; an honest assessment of current data foundations; the platform and architecture bet; the operating model (who owns what, build versus buy versus partner); and a sequenced roadmap with quarterly deliverables someone is accountable for.
Short enough to be remembered and enforced: ten pages beats fifty slides. The length test: can every workstream owner say what they ship this quarter and why the business cares? If not, it is a vision document, not a strategy.
Yes - platform ambiguity is a hidden tax on every project. For Microsoft-centric organisations the modern default is Fabric (with Databricks alongside for heavy engineering where warranted), and the strategy's job is to commit, name the exceptions, and stop relitigating quarterly.
As a consumer of the same foundations, not a separate track: every AI ambition - Copilot, agents, forecasting - runs on data quality, semantic models and governance. Strategies that fund an AI workstream while starving foundations buy demos, not capability.
Quarterly for delivery (did we ship, what changed), annually for direction. The failure mode is the opposite cadence: annual theatre about direction, no quarterly accountability for shipping.
