Context
O2's contact center upgraded to a Genesys Infomart data warehouse: a large star schema capturing calls, queues, agent interactions, and routing activity.
Afiniti's AI routing models depended on that data. The models needed a reliable feed of call information to determine which agent should handle each caller. The existing warehouse, however, had not been designed specifically for that use case.
The problem
The routing feed was built by joining data across multiple tables in the Genesys star schema, each with its own timing characteristics, quirks, and failure modes.
Every fragile join created another opportunity for the feed to silently drift from reality. Since Afiniti's revenue was tied to measured routing performance, this was more than a data-quality problem. A broken feed could have a direct financial impact.
What I built
I redesigned the consumption layer by consolidating the underlying Genesys star-schema tables into a single, unified calls feed built specifically for the routing models.
One feed. One contract. One place to enforce quality.
The key constraint was that the routing engine could not notice the change. Before cutover, I reconciled the new feed against the existing process line by line and validated the results through automated checks.
The transition completed with 100% of client revenue preserved.
During the same period, I also led the integration of Cisco PQ-based routing with Afiniti's systems for Virgin Media. The account's monthly revenue was heavily dependent on the integration working correctly.
What made it hard
The work required understanding and translating a complex Genesys data model into a structure that could integrate with Afiniti's existing design.
At the time, LLMs were still highly experimental and were not a practical substitute for the engineering work involved. I had to work through extensive Genesys documentation, map the underlying data model, understand the relationships between calls, queues, agents, and routing events, and determine how that model could reliably feed Afiniti's systems.
The other challenge was rebuilding a live system while operating under a revenue guarantee. Any unexplained anomaly could quickly become a client or billing concern.
That meant the engineering approach had to be disciplined: reconcile everything, automate the checks, and make sure the team found problems before the client did.
Outcome
The result was a single, reliable calls feed powering AI routing on a major telecom account, with revenue fully preserved through the transition.
The unified-feed pattern ultimately outlasted the project itself. It provided a better way to shape complex warehouse data for downstream models and became the template for future deployments on similar platforms.