Data strategy
Make data investment answerable to business decisions.
A data strategy should specify which decisions need better evidence, what information those decisions require and which capabilities are worth funding first.
The actual problem
The roadmap is busy, but its business logic is unclear.
Many data programmes are organised around systems, migrations and requests. That can produce useful infrastructure while leaving the central question unanswered: which decisions will become faster, more reliable or newly possible?
A decision-led strategy reverses the sequence. It begins with material recurring decisions, traces the evidence they require and then identifies the smallest set of changes to data, ownership and operating routines that will improve those decisions.
- Competing numbersFinance, sales and operations report different values for the same headline metric.
- Request-driven backlogPriorities reflect the loudest stakeholder rather than the value or risk of the decision supported.
- Technology-first roadmapProjects name platforms and pipelines but not the business behaviour expected to change.
- Slow evidenceSimple executive questions require manual reconciliation across teams and tools.
- Unowned definitionsA dashboard has a technical owner, but nobody is accountable for the meaning of its metrics.
- No retirement ruleReports and datasets accumulate because usefulness is never reviewed against a decision.
Method
Build the strategy from decisions backwards.
The method treats data capability as an operating system for decisions. It does not assume that every problem needs a new platform, a central team or a larger reporting estate.
- Decision inventory
- Identify recurring and consequential decisions, their owners, frequency, current evidence, delay cost and reversibility. Separate decisions from general questions and reporting preferences.
- Evidence map
- For each priority decision, document the required measures, dimensions, external context, level of freshness and acceptable uncertainty.
- Capability diagnosis
- Assess where definitions, capture, modelling, access, skills or governance prevent the evidence from being used reliably.
- Portfolio choices
- Compare initiatives by decision value, risk reduction, dependency, implementation effort and time to first usable outcome.
- Operating model
- Assign owners for metric meaning, technical production, quality response and the forums where evidence is reviewed.
- Learning cycle
- Define how the organisation will test whether a new data capability changed decision speed, consistency or outcome quality.
Evidence required
What must be examined before recommending a roadmap.
The useful evidence is not limited to database diagrams. Decision forums, spreadsheet workarounds and unresolved arguments often reveal more about the operating problem than the formal architecture.
| Input | What it reveals | Typical question |
|---|---|---|
| Leadership and operating meeting packs | The measures currently used to allocate attention and resources. | Which figure actually changes a decision? |
| Metric catalogue and dashboard inventory | Duplicate definitions, unused reporting and ownership gaps. | Where do two teams use the same name for different logic? |
| Representative source extracts | Grain, history, missingness and whether key dimensions are available. | Can the decision be analysed at the required level? |
| Transformation and lineage documentation | How raw events become reported measures. | Can a material number be traced and reproduced? |
| Data and technology roadmap | Current commitments, dependencies and sunk-cost assumptions. | Which project has a named decision outcome? |
| Roles and decision rights | Who defines, produces, challenges and acts on information. | Who can approve a definition or retire a report? |
Outputs
What a usable data strategy contains.
The result should be specific enough to govern investment and simple enough for business and technical leaders to use together.
- Decision and evidence mapPriority decisions linked to owners, measures, dimensions, sources, required freshness and known gaps.
- Metric architectureA concise hierarchy connecting strategic outcomes, operating drivers and diagnostic measures without treating every KPI as equally important.
- Prioritised capability portfolioInitiatives compared on decision value, risk, dependency, effort and time to usable evidence.
- Ownership modelAccountability for semantic definitions, technical production, access, quality incidents and decision use.
- Delivery sequenceNear-term proofs, enabling work and explicit conditions for scaling, pausing or retiring an initiative.
- Outcome measuresEvidence that the strategy improved decision latency, reconciliation effort, adoption or the quality of an operating outcome.
Worked example
Growth can conceal a retention decision.
A subscription business is considering additional sales capacity for its small-business segment. The dashboard reports that segment revenue grew by 18%, which appears to support acquisition investment. A decision-led review separates new revenue from the existing-customer base.
| Existing-base component | Value | Treatment in NRR |
|---|---|---|
| Opening recurring revenue | $5.0m | Denominator |
| Churn and contraction | -$0.9m | Subtract |
| Expansion from existing customers | +$0.2m | Add |
| New-customer revenue | +$1.8m | Excluded from NRR |
Closing revenue including new customers = 5.0 - 0.9 + 0.2 + 1.8 = $6.1m
The segment grows because acquisition more than offsets losses from the installed base. The relevant strategic question is no longer simply whether to add sales capacity. Leadership must compare the return from further acquisition with the return from correcting onboarding, product fit or service issues behind the 14% net loss in the existing base.
The data strategy implication: retain both growth and retention views, assign ownership for cohort logic and require investment decisions to state which economic mechanism they intend to change.
Limits
What this work does not replace.
A strategy creates choices and an operating frame. It does not remove the need for technical discovery, delivery leadership or accountable business owners.
- Not a platform selection exerciseTechnology recommendations require separate functional, architectural, security and commercial evaluation.
- Not a complete data auditRepresentative evidence can identify material risks, but it does not certify every source, pipeline or control.
- Not a substitute for ownershipA metric contract has little value if no business owner can resolve meaning or act on its movement.
- Not a fixed multi-year forecastThe roadmap should be reviewed as decisions, constraints and evidence change.