Anonymised under NDA
Market intelligence agents for sports and media operators
Sports & Media · Multiple sports and media operators in the UK
Multi-agent system deployed across several UK sports and media operators, monitoring news, rights announcements, competitor activity, and fixture and audience data to generate explainable signals for commercial and editorial teams.
The problem
- Commercial and editorial teams at each operator were spending hours per day triaging news, rights announcements, competitor pricing, and social activity by hand.
- Earlier vendor tooling produced unexplained scores that nobody would act on: a signal with no evidence was ignored, however accurate it was.
- Each operator had its own sources, rights, and priorities, so a single shared model was not acceptable; the system had to be deployed per company with strict data separation.
The architecture
- Agent graph: ingestion agents normalise sources; analyst agents reason and cite; reviewer agents evaluate before a signal is published to the team.
- Hybrid retrieval (BM25 + vector) over each operator's curated corpus, with re-ranking and source-grounded citations on every signal.
- Model-agnostic orchestration deployed per operator, with tenant-isolated corpora, credentials, and evaluation sets so no company's data or signals were shared with another.
LangGraphClaudeLlamaQdrantPostgreSQLTemporal
The outcome
- Teams moved from triage to interpretation, with every signal explainable down to its source documents.
- The same architecture rolled out across several operators with configuration rather than re-engineering, each keeping its own sources and priorities.
Services drawn on
A similar problem on your roadmap?
A Discovery Sprint is the fastest path to a scoped, costed proposal.