Ryan Biggs

Every agent conversation starts the same way: the agent knows nothing about the person it is talking to. It asks questions the user already answered elsewhere in the session, or it gives a generic answer that ignores what they were doing ten minutes earlier. That is the cold start problem. Even agents with memory usually pull from CRMs or warehouse tables that are hours or days stale, missing what is happening in the session right now. This session shows how connecting a real-time behavioural data pipeline to Google Gemini Enterprise closes that gap. Using Snowplow Signals to turn live event streams into user attributes, walk through a working e-commerce concierge agent built on Gemini Enterprise CX Agent Studio. The agent calls Signals for session context before its first reply, then again on every turn, so recommendations reflect what a shopper is currently doing. Attendees learn: - The architecture connecting Snowplow to Gemini agent tools. - How attribute retrieval stays under 10 milliseconds. - How teams configure agent instructions and attributes without redeploying the app. The benefits extend beyond e-commerce, to support and publishing, where the same context changes how an agent responds, plus a pattern for wiring real-time context into any customer-facing agent.