How a 90-agent brokerage cut lead response from 41 minutes to 2
The before state was familiar. Inbound leads landed in a shared inbox, an ops coordinator triaged them by hand, and the average wait was 41 minutes. Two thirds of the lost deals were lost in that window.
What we built
A scoring agent that reads each inbound lead, enriches it against the firm’s historical close data, and pages the agent most likely to win the deal in Slack. Tools used: HubSpot for the CRM of record, Clay for enrichment, OpenAI for scoring, n8n to glue it together.
The four signals we started with
- Buyer geography vs. agent farm area
- Property type vs. agent recent closings
- Price band vs. agent median deal size
- Inbound channel vs. agent historical win rate by channel
By week three we cut geography and channel. They added latency without changing routing decisions in more than a handful of cases. The agent got faster and the team trusted it more.
What this is not
This is not an AI SDR. It does not message the lead. A human still owns the first reply. The system buys the human time to be useful.
What we measured
| Metric | Before | After |
|---|---|---|
| Avg first response | 41 min | 2 min |
| Lead-to-meeting rate | 8.4% | 14.1% |
| Coordinator hours / week | 22 | 6 |
The coordinator did not lose her job. She owns escalations and quality review on the routing model now, which is the work that should always have been hers.