The problem
A housing rights organization serving low-income tenants had built custom AI systems to support intake, triage, and referral workflows, but staffing changes left the systems undocumented and risky. Leadership needed independent assessment of which AI components were working, which needed replacement, and how to navigate vendor transitions without losing functionality. Data fragmentation across systems created gaps in service delivery. There was no oversight of automation failures (text failures, intermittent vendor failures).
What we did
- Conducted structured discovery interviews with intake and legal staff to understand actual workflows and pain points
- Mapped data flows across systems to identify fragmentation, security risks, and handoff failures
- Assessed custom AI systems: evaluated what was working, what was brittle, and what required replacement
- Migrated critical vector database from failing vendor to Microsoft Azure to restore stability
- Redesigned chatbot-to-human handoff so clients wouldn't be lost between automated triage and live staff
What changed
- Stabilized AI systems post-staffing transition without service disruption
- Improved text delivery rate from 50% to 98%
- Reduced technology costs by consolidating redundant systems
- Expanded self-service tools to new issue areas and jurisdictions
Built with
Microsoft Azure
vector databases
LLM integration
Chatbots