2026-08-31 / 6 min
Decision-grade data is built around consequences
Why executive reporting fails when it is organised around systems instead of decisions, accountability, and commercial consequence, and the seven tests a metric has to pass before a board should act on it.
In brief: a metric is decision-grade when the decision it serves, the consequence of it being wrong, its owner, its lineage and its quality threshold are all written down and inspectable. Most reporting estates fail at least three of those tests, which is why technically correct reports still get argued over in the boardroom.
Most enterprise reporting estates are full of technically correct outputs that do not change a decision. The dashboard refreshes, the MI pack circulates, and the metric definition is defended, but the business still argues about which number matters.
Decision-grade data starts somewhere else. It starts with the question an executive, regulator, investor, or commercial team must answer. What are we deciding? What is the consequence of being wrong? Who owns the metric? What evidence would make the answer inspectable?
That shift sounds small, but it changes the operating model. A metric is no longer an artefact emitted by a reporting team. It becomes a data product with ownership, lineage, quality thresholds, retirement rules, and a clear decision context.
Why correct reports still fail executives
The distinction between correct and decision-grade is not academic. A report can reconcile to the general ledger and still be useless in the meeting where it lands, because nobody in the room can say what would have to be true for the number to be wrong, or what they would do differently if it were.
The evidence on trust is uncomfortable. The 2026 State of Data Integrity and AI Readiness study from Precisely and Drexel LeBow found that a third of data leaders still do not report high trust in the data their organisation uses to make decisions, and that one in seven say their organisation lacks the skills and training to use it. Workiva's 2026 executive benchmark put the consequences in order: bad or delayed operational decisions first, then regulatory fines or legal action, then the loss of investor or lender credibility. None of those consequences arrive because a report is arithmetically wrong. They arrive because the number was acted on without anybody being accountable for whether it could bear the weight.
The seven tests
A metric earns decision-grade status when it passes seven tests. Each one is a question an executive can ask in the room, and each has a documentary answer rather than a reassurance.
| Test | The question | The evidence that answers it |
|---|---|---|
| Decision | Which decision does this number inform? | A named decision, a named forum, and a decision cadence |
| Consequence | What happens if it is wrong by ten per cent? | A materiality statement agreed with the decision owner |
| Ownership | Who is accountable for the definition and the value? | A steward named at the point where the data is understood, not where the technology sits |
| Definition | Is there one definition, and is it the one finance uses? | A single entry in the business glossary with a change history |
| Lineage | Can we trace it from source to pack without a spreadsheet? | Captured lineage at the grain the control needs, not a diagram |
| Quality | What threshold triggers action, and who is told? | A scored quality rule with an escalation route and a log of breaches |
| Retirement | When does this report stop being produced? | A review date and a named decision that would retire it |
The retirement test is the one most estates skip. Reporting accumulates because nobody is empowered to stop producing something a previous executive once asked for. A data function judged by output volume will never retire anything. A data function judged by decision quality retires reports as a matter of routine.
Critical Data Elements are the unit of work
The practical method for getting there is the Critical Data Element, or CDE. The approach aligns with BCBS 239, the Basel Committee's principles for risk data aggregation, and transfers to any organisation that has to defend its numbers.
A CDE is a data element whose failure would affect a material decision, business outcome or regulatory obligation. The work is to implement a repeatable process for identifying criticality, capturing definitions and lineage, and monitoring quality across complex data flows. Business impact and risk determine scope. Named owners and stewards maintain the definitions, source mappings, transformations and quality rules that connect each element to the decisions and obligations it serves. Monitoring must route breaches to accountable owners, track remediation and verify that the underlying issue is resolved. Criticality and controls need reassessment as the business and its data dependencies change.
The Basel Committee's January 2026 newsletter on BCBS 239 implementation is a useful reminder that this work is never finished. It singled out data lineage and traceability, ad hoc reporting during periods of stress, and cross-border data alignment as the areas where supervisors still see gaps, more than a decade after the principles were published. The same note observed that AI adoption in risk data aggregation is still at an early stage, and that the quality of AI-driven output depends entirely on the quality of the data beneath it. Decision-grade data is the precondition for decision-grade AI, not a parallel workstream.
What changes in the operating model
Once the seven tests are in place, four things change in how the organisation works.
First, arguments about the number move upstream. Disagreement about a metric becomes a change request against a definition with an owner, rather than a debate in the board pack. Second, the reporting team stops being the default owner of every number and becomes the custodian of a catalogue whose entries are owned by the business. Third, quality failures become events with a route, not observations in a footnote. Fourth, the data function acquires a way to say no: a request for a new report has to name the decision it serves and the CDE it draws on, or it does not enter the backlog.
In regulated and PE-backed environments this discipline matters because data work has to earn capital and trust at the same time. The CFO needs confidence in performance. The CRO needs control evidence. The CEO needs a believable view of growth, margin, retention, and risk. The board needs to know when the numbers can be acted on.
A note on UK data law
The tests above are about the reliability of a number. UK data protection law now adds a second set of questions about whether the organisation was entitled to process the underlying data in the first place, and those questions have sharpened during 2026. The Data (Use and Access) Act 2025 commenced its principal data protection reforms on 5 February 2026, and from 19 June 2026 every controller must operate a formal complaints procedure for data subjects, acknowledging complaints within thirty days. A decision-grade data product has to be able to show its lawful basis and its complaint history as readily as it shows its lineage.
The practical work is not glamorous. It is catalogue rationalisation, Critical Data Element definition, steward assignment, lineage capture, quality scoring, and a willingness to retire reporting that no longer serves a decision. The commercial payoff is clarity: fewer arguments about the numbers, faster decisions, and a data function that is judged by outcomes rather than output volume.
Decision-grade data is not perfect data. It is data with enough evidence, ownership, and control to support the decision in front of it.
Sources
Related service: Advisory
Related insights
2026-10-09 / 8 min
Management information now needs an audit trail
On 27 July 2026 the FCA published its review of Consumer Duty outcomes monitoring, drawn from a survey of 56 firms. Its finding is that holding management information is not evidence of good outcomes, and thresholds nobody can justify prove nothing.
2026-10-05 / 7 min
Bank disclosure becomes a comparable dataset
At its meeting on 28 and 29 September 2026 the Basel Committee approved the final standard for machine-readable Pillar 3 disclosures, with publication due around the end of the year. The EU has run the experiment since January, and UK firms have one reporting cycle left.
2026-09-28 / 7 min
Incident reporting is now a data quality test
On 16 September 2026 the European Supervisory Authorities published fourteen operational instructions on DORA major incident reporting, every one of them a correction to something firms filled in wrongly. The UK equivalent applies from 18 March 2027.
Executive Data Briefing
A low-volume note for data and AI decisions with consequence.
Consent-based and double opt-in. Governance patterns, board-level data trust, and decision infrastructure - not generic AI commentary.
Follow on LinkedIn