[Root cause source doc](https://docs.google.com/document/d/e/2PACX-1vQW6hUO73E_r49xl5BenN7DjzZUDiSa-OZgraim5gO8uUjV-D6x1VGCBHD2VC5eNIVyMmQiu7C3zUn0/pub?embedded=true) ## Executive Summary - Escalations rose from **9% to 19%** one month after the workflow change. - Primary contributors: workflow visibility change, resident traffic misrouting, and new volume from RentsRUs.com. - Some handoffs are expected (negotiations, hyper-specific appliance questions), but this baseline should be stable. - Next steps focus on routing resident inquiries correctly and reinforcing "Chime In" usage. --- Client X (3,000 units, 12 buildings) reports that "every conversation" is escalating to their team. Their escalation rate jumped from **9% to 19%** month-over-month—more than doubling. This occurred one month after a workflow change to how handoffs enter their CRM. ## Guest Card Analysis **Classification:** Current Resident (not a prospect) This person has an active lease and tenant ID. Their message is a maintenance complaint about a broken sink. The leasing AI cannot handle resident service requests—it's designed for prospects only. This is automatic escalation. **Root cause indicated:** Resident traffic is being routed to the leasing AI channel. > Reminder: Client X is using our leasing AI only. **Classification:** Prospect | **Source:** RentsRUs.com The prospect asked if rent is negotiable. Per EliseAI's training materials, rent negotiation questions are explicitly flagged for human review. This lead came from RentsRUs.com, the new marketing source. The AI successfully ingested the lead but escalated due to the flagged topic. **Root cause indicated:** Expected behavior for the question type. **Classification:** Prospect | **Source:** Apartments.com The prospect asked what year the fridge was installed in a specific unit and whether it has maintenance history. This level of appliance detail isn't in the AI's knowledge base. **Root cause indicated:** None, this is the system working as designed. Hyper-specific questions get flagged. ## Root Causes The Former and Current workflows handle flagged questions very differently: Former Workflow Current Workflow Flagged questions split into Urgent vs Non-Urgent All flagged questions treated the same Urgent → Unreviewed Queue (real-time) All → Unreviewed Queue (real-time) Non-Urgent → Daily Report (batched email) No more Daily Report for action items AI continues managing non-urgent leads Agent must Chime In or Takeover for all

**What changed:** Previously, non-urgent handoffs (like "What's the view from apt 302?") went to a daily email digest while the AI continued managing the lead. Now ALL handoffs hit the Unreviewed Queue immediately. The team isn't necessarily receiving more escalations, **they're seeing more**. That statement seems contradictory but the emphasis lies in the distinction between generally receiving and the amount of escalations that 'get to their desk'. Items that used to arrive summarized once daily now appear continuously in their queue. This explains why the client perceives that "every conversation" is hitting them when the actual rate is 19%. Guest Card 1 proves current residents are contacting channels monitored by the leasing AI. Since the AI handles prospects only, every resident inquiry automatically escalates. **If even 10% of escalated traffic is residents, this alone could explain much of the increase from 9% to 19%.** We investigated your concern regarding the new RentsRUs.com integration. Guest card data confirms that these leads are successfully routing to the AI. **Status:** Working as designed. **Impact:** This source contributes 3% of your total traffic. While the integration is healthy, this new volume contributes slightly to the overall increase in activity your team is seeing. Some escalations are the system correctly recognizing its limits: rent negotiations, specific appliance questions, off-topic messages. These aren't bugs. However, this baseline should be stable, not doubling. ## Traffic Source Breakdown Source Share of Traffic Status Property Website 39% Established Zillow 31% Established Apartments.com 16% Established apartmentlist.com 11% Established RentsRUs.com 3% New - Monitor Quality

## Recommended Next Steps - Segment escalations by customer type (resident vs. prospect) - Compare urgent vs. non-urgent handoff breakdown before and after the workflow change - Monitor RentsRUs.com lead quality to ensure continued alignment with AI training - Route resident inquiries away from leasing AI channels - Reinforce team training on the new workflow, specifically the "Chime In" feature which lets them answer the flagged question and return control to the AI (vs. full Takeover which requires them to manage the entire conversation) ## Summary Table Root Cause Impact Level Evidence Workflow visibility change High (perception) Training docs show daily report → real-time queue Resident traffic misrouting High (volume) Guest Card 1 is a current resident RentsRUs.com Integration Low (baseline) Verified active and working as designed Legitimate AI handoffs Low (baseline) Guest Cards 2 and 3 show expected flags

## Conclusion The client's frustration is valid, but "every conversation" is hyperbolic; it's 19%, not 100%. **The fix requires both technical corrections and change management.**