Agentic analytics across 20+ markets
From weeks to under 15 seconds for a business answer
- Sector
- Consumer goods (FMCG)
- Market
- Asia-Pacific
- Timeline
- 9 months, team of 11
<15 sfrom question to validated answer
99.8%retrieval accuracy against trusted metrics
20+markets on one data model
The challenge
Data sat in many systems across operating companies in more than 20 countries, each market defined its KPIs differently, and every non-standard question needed an analyst and hand-written SQL. Decisions waited days or weeks for answers.
What was built
- A harmonised data foundation on Databricks and Delta Lake
- An enterprise semantic layer that fixes KPI definitions across every market
- Six cooperating agents: orchestrator, planner, context, query, analysis, and an LLM-as-judge that validates every answer
- Conversational analytics with automatic SQL, narrated insights and charts
How it was built
- A team of four data engineers, three AI engineers, two data stewards, a solution architect and a PMO
- Private networking and single sign-on, with the judge agent as a guardrail on reasoning