Clients are named by industry unless they have agreed in writing to be named. A result appears only once the client has confirmed it.
FMCG company
10 weeks
Problem
Leadership questions over sales data spread across spreadsheets and a relational database.
What we shipped
We built a knowledge graph from the data and an LLM agent on top.
Result
Leadership understood the analysis sooner and made key business decisions on it. The client needed a smaller analytics team, and margins rose 3–8% year on year across its regions.As reported by the client.
Spreadsheetssales dataDatabaserelationalKnowledge GraphLLM agentLeadershipquestionsYear on year, across regionsMargins up 3–8%Faster leadership decisionsSmaller analytics team needed
Simplified from the engagement. Results as reported by the client.
Public-sector client
14-week pilot
Problem
Answering questions about frequently changing policy documents, with no cloud models allowed.
What we shipped
Local models for common questions, a larger model for complex ones, and escalation to an officer when confidence was low. Everything ran on their own infrastructure.
Result
Not yet published.
Their infrastructureQuestionRoutingLocal modelcommon questionsLarger modelcomplex questionsOfficerlow confidence
Their infrastructureQuestionRoutingLocal modelcommon questionsLarger modelcomplex questionsOfficerlow confidence
Simplified from the pilot. Everything ran on the client's own infrastructure.