One AI agent, four stores, four different refund policies — the customer found all of them

Description and use

One AI receptionist, four stores, four different cancellation policies — the customer discovered all of them

Problem

A shared AI receptionist returned different cancellation policies depending on which location page it retrieved, causing contradictory answers and customer distrust.

Outcome

Zero cross‑location contradictions, improved customer trust, and streamlined policy updates via a single table edit.

Workflow

• Audit existing policy pages for inconsistencies
• Create a canonical policy table with scope and last_reviewed
• Replace prose pages with table rows
• Update system prompt to ask for location when rule is location‑scoped
• Run monthly contradiction audit on 40 random transcripts
• Monitor audit metrics on weekly scorecard
• Edit table rows for policy updates instead of emailing managers

Evidence

The AI was quoting four different cancellation policies, depending on which page of the knowledge base it happened to retrieve from.

Source

The fork rate — transcripts where two policy answers in the same week contradicted each other — went from 4.1% of all transcripts to zero in the six weeks I tracked.

Source

The dental group rebuilt 43 policy facts into 28 brand rows and 15 location rows in one afternoon.

Source