Scaling an AI leasing assistant from 63% to 93% automation

Background

The numbers told a clear story: only 63% of messages were automated, leaving the rest to human Operators and creating inconsistent experiences. The product was running at -5% profitability. Less than 1% of customers were enterprise clients, which meant Lisa was running out of customers to sell to. On the user side, leasing agents didn't trust Lisa, found her too rigid, and showings weren't optimized for how their teams actually worked. The original classification-based ML model couldn't handle the nuance needed to fix any of this.

Demo: Lisa working inside the AppFolio CRM. Property managers can see how Lisa responded, review a summary and key tenant details, and step in to take over the conversation at any point.

My role

I led the product strategy for transitioning Lisa from a classification-based ML model to an LLM architecture. I designed a Skill Router that classified incoming messages and routed them to purpose-built LLM skills, which gave Lisa the flexibility and personalization agents needed to actually trust her. I coordinated across Engineering, Machine Learning, Design, Marketing, Operators, the CRM team, and business stakeholders to execute a phased deployment timed to the AppFolio Customer Conference. I also redesigned onboarding to cut time-to-value. One of the biggest bets I made was the CRM integration: rather than relying on background Operators to monitor and correct Lisa's responses, I built the tooling directly into the AppFolio CRM so property managers could see exactly how Lisa had responded, get a high-level summary of the conversation, review key details about the prospective tenant, and step in to take over the conversation whenever they wanted. This shifted control from a separate Operator layer to the property managers themselves, which reduced costs and made the product far more trustworthy.

Outcomes

• ·Improved automation from 63% to 93%, eliminating the need for most background Operator intervention
• ·Turned product profitability from -5% to 30% within the first 2 months
• ·Cut churn from 20% to 3%
• ·Reduced onboarding time by 2 weeks
• ·Expanded enterprise client acquisition with packaged pricing across 3 leasing products
• ·CRM integration generated 456 paying customers in the first 3 months ($50/month upsell)