How do you get partners to try an AI agent they already doubt?

The questions a text agent handled worst were about the unit itself: what floor, what view, and whether it takes pets. Text-only answers and links to the listing sent renters away to look, which was where they left the conversation. We chose to give the agent media already attached to the listing and surface it inline. That meant it could only be as good as a partner's uploads: four dark photos could make it confidently show four dark photos, shaping a renter's view of a unit they had decided against. There was no mitigation that fixed this. We accepted the trade because renters who leave to look do not return, and a mediocre answer in the conversation beats a good answer elsewhere. Partners could check the agent's work on their own listings before being asked to trust it with a lead.

The handoff note had been written by a person on paper. No handoff artefact left leasing agents without context, while a full transcript was technically complete but impractical to read mid-shift. We chose a generated summary of what had been established and where the lead moved from low to high intent. A wrong summary could lead an agent to confidently restate something the renter never said, so it sat alongside the conversation rather than replacing it and could be checked in one click. Leasing agents could pick up without re-asking. The summary also made lead quality legible: after Diffe.rent was acquired by Apartment List, it was counted in move-ins a property could trace to an Apartment List lead and verify against its own leases.