Real Estate Lead Qualification

Phased rollout timeline

1. 01

   Week 1 · Lead-stage mapping

   We began by mapping the team's existing lead stages against conversation signals. Their CRM had 7 lead statuses, but we found only 3 were meaningful for chatbot qualification: intent type (buy/rent/invest), budget range, and location preference. We designed qualification fields that mapped directly to these stages so sales agents received context they could act on immediately.

2. 02

   Weeks 2-3 · Dynamic branching

   We implemented dynamic question branches based on budget and location constraints. A challenge we encountered early was that property terminology varied significantly by region -- “HDB” vs “condo” vs “landed” confused the classifier initially. We resolved this by adding region-specific synonym mappings to the intent model, which improved classification accuracy from ~78% to ~91% for property-type queries.

3. 03

   Weeks 4-5 · CRM sync and scheduling

   When we tested the CRM sync, we discovered that duplicate leads were being created when users revisited the chatbot within the same session. We implemented a deduplication layer using session fingerprinting and email matching. We also hardened the booking workflow to handle timezone edge cases that caused missed meeting slots during the first pilot week.

4. 04

   Week 6 · Conversion tuning

   In our final sprint, we optimized low-confidence handoffs. We found that leads with ambiguous intent (e.g., “just browsing”) converted at 4% when auto-routed but at 18% when the chatbot asked one clarifying follow-up question. This single change drove the largest week-over-week improvement in qualified conversion rate.

What we observed in production

• In our experience, qualification quality improves most when intent and urgency signals are captured in the first 2-3 conversation turns. Late-stage qualification questions had lower completion rates.
• We tracked sales acceptance rate as our primary optimization signal rather than raw lead volume. This prevented the chatbot from over-qualifying leads that agents wouldn't follow up on.
• Speed-to-contact dropped from 4h 06m to 17m, but only after we fixed CRM ownership assignment rules. Without clear routing, leads sat unassigned for 30+ minutes even with chatbot handoff.

What we'd do differently next time

• Launch stricter duplicate-lead controls before week 1 pilot traffic. We spent 2 days cleaning up ~120 duplicate records created during the first test run.
• Add localized property-term mappings from day one. The regional terminology gap cost us a week of classifier retraining.
• Define escalation SLAs upfront with the sales operations team. Without pre-agreed response windows, high-intent leads occasionally waited longer than the chatbot's promised callback time.