Hospitality · Enterprise AI · Boutique hotel group
An AI copilot that halved average handle time
Agents got a real-time copilot that drafts replies, pulls the exact policy, and writes the call summary the moment they hang up.
ReactWebRTCLLMRAG over policy
-48%
handle time
Timeline
4 months
Team
4 engineers + 1 ML engineer
The challenge
What was breaking
- 01Long holds while agents hunted for the right policy.
- 02Inconsistent answers across shifts and properties.
- 03High agent churn from repetitive, draining work.
The approach
How we tackled it
- 01Built a RAG index over every policy, rate, and property fact.
- 02Surfaced live suggestions and drafted replies in the agent's console.
- 03Auto-summarized each call and nudged on negative sentiment.
The solution
What we shipped
- 01A copilot that drafts and cites, never hallucinates a rate.
- 02Automatic call summaries straight into the CRM.
- 03Sentiment nudges that catch an escalation before it happens.
In pictures
A closer look
The results
-48%
handle time
+0.6
CSAT lift
-35%
escalations
90%
auto-summaries
"New agents sound like veterans on day one. The copilot carries the policy knowledge so people can carry the empathy."
Related work