As easy contacts move to self‑service, the remaining ones get harder. Agents need better tools.

Agent Success is an agent-assist platform that works alongside the people answering customer calls and chats: knowledge, guidance, writing help, and after-call summaries, configured for each client. I joined as product manager, took on design when that role opened, and carried both as the team grew. That meant researching inside client operations, designing and specifying much of the product, and leading client engagements.

The problems worth solving

When we started, transcription and summarization were the hard part. Now they’re table stakes. The interesting problems were the ones a model couldn’t solve on its own:

• Making a new agent sound like a veteran. Burst events meant onboarding a wave of temporary agents for a few weeks, many of whom didn’t know the domain.
• Helping agents listen. In personalized sales, the opportunity lives in rapport and building relationships, not the product list.
• Answering what no single system knows. Some answers meant pulling several live, dynamic data sources together in the middle of a conversation.

Then making it scale

Each solution started bespoke: the integration, the data, the AI, and the problem. The harder product problem was turning them into a platform that could reach 40,000 agents across hundreds of lines of business, each with its own systems, rules, and knowledge. Throughout, the core decision was where the AI should act and where the agent’s judgment should win.

Outcomes