Why did first-line tenant inquiries take a full working day?
Regional teams were clearing the same inbox every morning: rent statement questions, key handovers, repair status, contract copies. Roughly two thirds of the volume was repeatable and already documented somewhere — but finding the documentation took longer than writing the answer from memory, so answers varied with whoever was on shift.
How do you build an assistant over a tenancy handbook and ticket history?
We built a retrieval layer over the tenancy handbook, the maintenance archive and three years of resolved tickets, then an assistant that drafts a reply with its sources attached. Anything touching contracts, terminations or payment disputes is classified out of scope by design and routed to a named person rather than answered.
Where does the assistant run day to day?
The assistant runs inside the existing ticket system, so there is no new tool for the team to learn. It reads from SharePoint and the document archive through service accounts scoped per region, and writes the resolved thread back to the CRM so the tenant record stays the system of record.
How is a tenant-facing assistant kept compliant?
Classified as limited-risk, with a transparency notice on every outbound message. Draft replies were reviewed before sending for the first eight weeks; the review requirement was relaxed only for the categories where measured accuracy stayed above the agreed threshold. Every draft, source and edit is retained for audit.
What changed for the service desk?
The service-desk assistant moved first-line tenant responses from the next working day to minutes, and the operators kept the cases where a tenant actually needed a person. That result came from a retrieval layer over the tenancy handbook, the maintenance archive and three years of resolved tickets: the assistant drafts a reply with its sources attached, inside the ticket system the team already uses. Anything touching contracts, terminations or payment disputes is classified out of scope and routed to a named person. What made it stick was the governance, not the model. Draft replies were reviewed before sending for the first eight weeks, and the review requirement was relaxed only for categories where measured accuracy stayed above the agreed threshold. The escalation rate is still reviewed weekly, and it decides whether a category stays automated. That measure-then-relax discipline is what the Operate engagement continues after go-live, and the same insistence on traceability runs through the logistics KPI briefing.