Rebuilding the analytics layer for a national retailer

The challenge

Three teams reported three different revenue numbers, and every leadership meeting started with an argument about whose figure was right. Nobody trusted the dashboards, so nobody used them.

What we did

• Rebuilt the warehouse on a tested dimensional model with lineage in dbt
• Agreed a single, documented definition for every core metric — the hardest and most valuable part of the project
• Added pipeline monitoring so breakages were caught before they reached a report
• Layered propensity and demand models on top, once the foundation was trustworthy

Outcome

• One number for revenue, agreed and documented
• Merchandising began acting on demand forecasts rather than instinct
• Dashboard usage rose sharply once the data could be trusted

“They spent the first six weeks on definitions, not models. We were impatient. They were right.”

— Multi-channel retail chain

Client names are withheld under our confidentiality agreements. Outcomes are described qualitatively rather than with precise figures we are not permitted to publish — we would rather say less than overstate.

← All case studies