The challenge
Seven unreconciled business data sources, reports produced manually each month, and figures that differed between departments — fuelling arbitration rather than informing it.
The outcome
A unified data platform feeding both reporting and predictive models, shared and versioned business definitions, and eighty percent fewer manual extractions.
The context
Every department produced its own indicators from its own extract. Monthly revenue had three different values depending on which source you consulted, and nobody could explain the gap.
Our approach
The work was less technical than expected. Most of it consisted in converging business definitions before building anything.
- A shared business glossary — every indicator has a written definition, an owner and a calculation rule, versioned alongside the code.
- Lakehouse architecture — ingestion of all seven sources, historisation, and a single serving layer for both reporting and models.
- Tooled quality — consistency checks run at every load, alerting on trend breaks rather than fixed thresholds.
- First predictive models — demand forecasting per point of sale, built on the same layer as the reporting.
The outcome
Manual extractions fell by eighty percent. The discrepancies between departments disappeared — not because the tool corrected them, but because the definitions were finally settled.