An employee agent that gives field reps 20% of their day back

Concierge reps at this luxury resale marketplace carry several consignor appointments a day, and the administrative work around them competed directly with the relationship work the business runs on. Pulling up a consignor's history, taking appointment notes, entering forms and creating the opportunity all happened through a desktop CRM, while the rep was on site with a consignor and working from a phone.

Missed appointments added a second manual loop. When a consignment drop-off or van pick-up did not happen, someone had to notice it, then decide what to say and write the outreach by hand. Inside sales reps working high-volume ship-direct consignments faced the same data-entry load at a different scale.

The company wanted to grow consignment supply without diluting the high-touch service that differentiates it, which meant taking the administrative load off the rep rather than adding headcount to absorb it.

We delivered the advisory, solution architecture, prompt engineering and custom Apex and Flow work behind a Phase 1 Agentforce Employee Agent, designed for a phone first rather than adapted to one later.

The agent summarizes a consignor's profile and interaction history on request. It surfaces the rep's own assigned contacts first, but still allows lookup across every matching contact in the system, so a rep covering for a colleague is not blocked. Knowledge and SOP content was indexed into a custom Data Cloud data library and segmented by article type, so the agent links a rep straight to the training document when the answer is a documented process, and summarizes cleanly when it is not. Reps also create opportunities by talking to the agent, which removed the picklist API-versus-label friction that made mobile opportunity creation error-prone.

For missed drop-offs and pick-ups, we built detection logic on opportunity stage and a custom UI widget on the opportunity record. The widget calls an agent action that drafts the recommended follow-up text, which the rep sends through the client's existing Zoom SMS integration in one click.

Two things needed a second pass, and both are worth naming. Early responses treated procedural SOPs like any other knowledge article, summarizing steps a rep was meant to follow exactly - so we separated handling by article type and required direct attribution and hyperlinks rather than paraphrase. And prompt templates carrying hardcoded sandbox URLs would not have survived promotion between environments, so those were parameterized before production. We also stood up a pilot testing and feedback framework, a topic migration guide between service and employee agent architectures, and Data Cloud evaluation reporting for instruction adherence and token consumption, so the client's team could govern the agent after we left.