Revenue analytics platform
One governed warehouse behind every executive number, with the definitions owned in one place instead of re-derived per dashboard.
Context
- Leadership asks the same question three ways and gets three answers, each defensible, none reconcilable.
- Every team has built its own extract. The finance number, the sales number and the board number diverge, and reconciling them is a monthly manual ritual.
- Nobody disputes the data is wrong. What is disputed is whose version is wrong.
What was actually wrong
- The failure is definitional, not technical. Terms like net new, active customer, qualified, churned and booked are each used with several meanings, and no artifact records which meaning a given report used.
- Dashboards are built directly on source extracts, so business logic lives inside a dozen visualisation layers where it cannot be tested, reviewed or reused.
- Because the logic is invisible, a change to it looks like a change in the business.
Approach
- 01Start from the decision, not the data: inventory the decisions leadership actually makes on a cadence, and work backwards to the smallest set of metrics that move those decisions.
- 02Write the definitions down as a metric contract - grain, filters, inclusion rules, effective dates, owner - and get them ratified by the function that owns the number rather than by the analytics team.
- 03Model in layers: raw landing kept immutable, a conformed core with slowly-changing dimensions handled explicitly, then presentation models shaped to the questions.
- 04Serve every surface from the same semantic layer, so a dashboard, an export and an API answer identically by construction.
- 05Close the loop with a governance cadence: definition changes are reviewed, versioned and announced, because a silent definition change is indistinguishable from a business event.
Architecture
- Cloud data warehouse with partitioned, clustered fact tables sized to the real query patterns
- Transformation as version-controlled, tested models with lineage from source to serving
- Slowly-changing dimensions handled explicitly, so historical questions get historical answers
- A semantic layer as the single definition of every metric
- Containerised web application for the serving surface, deployed through CI with identity-aware access
What this demonstrates
- Turning an argument about numbers into a governed system with owners
- Warehouse modelling that survives contact with real history rather than only the current snapshot
- Designing for reconciliation up front, so the finance view and the board view cannot silently drift
Related
Demand and funnel diagnostics
Establishing whether a trend actually moved, before anyone spends a quarter reacting to it.
Analytics as an operating capability
Wiring analytics into how each department actually decides, instead of delivering reports at them.
AI agent orchestration, governed
Multi-agent systems that do real work under explicit human authority, with verification as a gate rather than a hope.
Client names, sector detail and figures are deliberately omitted throughout. These pages describe the shape of problems and the method applied to them - a metric attached to a real engagement does not belong on a public site, and an invented one is worth nothing.