The trust layer for an agentic sales system

A large financial services company replaced manual outbound sales with AI agents that send follow-ups, draft briefs, and recommend pricing in the background. Sellers went from doing the work to supervising it. I was embedded with product and engineering for a quarter as the AI experience flow designer, designing the interfaces that made that supervision possible.

For sellers, I designed an exception queue instead of a feed. The agent works on its own, and the screen surfaces only the accounts that need a person, with the agent's reasoning in plain language. Sellers set how much the agent can do without asking, by action type and lead type. Every action lands in an activity trail they can audit, and a visible feedback loop shows their corrections shaping the agent over time.

For leadership, I designed an outbound performance dashboard that traces the whole funnel, from nurture engagement to pipeline handed to sales. Every metric carries a status for whether the data behind it can be trusted yet, so gaps became visible, assignable work instead of silent errors.

The system shipped over the quarter, and seller overrides became their own signal for where the agents still needed work.

If your team is putting AI agents in front of users, I can design the layer that makes them trustworthy.

The client is confidential. Screens use sample data with identifying details removed. Full case study: https://jonhaines.design/uxcalibur/work/agentic-sales