FITPASS · Agentic AI Prototype
Partner Churn - Early Warning System
FITPASS today only learns a partner is unhappy when they reach out, or when they've already gone silent. This project designs an agentic system that watches supply, demand, and payout signals across 12,000+ gym, studio, and sports facility partners in 150 cities, and flags who's at risk before they leave.
The problem
FITPASS's partner network is the supply side of the marketplace. When a gym or studio quietly reduces the slots it lists, or stops getting bookings, that's an early signal of disengagement, long before the partner is picked up by an SLA or a complaint. Today, that signal goes unwatched. Partners either escalate themselves, or they churn silently, and the org finds out after the fact.
The goal of this system is to move FITPASS from reactive to proactive: continuously score every partner's health, tier them by commercial value and market context, and route the ones losing health to the right city head before churn happens, not after.
How it works
Sensing
The agent monitors four signals per partner, refreshed monthly:
| Signal | What it tells us |
|---|---|
| Slots listed | Supply the partner is choosing to offer, often the first sign of pulling back |
| Slots booked | Utilization, demand actually meeting that supply |
| Attendance confirmed | Real usage versus a booking that never converts |
| Payout per cycle | The commercial reason a partner has to stay |
Reasoning
Detection isn't a single threshold. It's a two-step check: a partner is provisionally flagged against their own trailing 3-month baseline, then verified against city-wide demand context before a human ever sees it, so a citywide soft month doesn't get mistaken for one partner's problem.
Tiering
Every partner is scored A through E using payout per slot and city demand density together, since a premium partner in a quiet market carries different risk than the same partner in a saturated one.
Action & routing
Flagged partners route to the relevant city head: Tier A and B inline in a monthly email with contact details and a suggested next step, Tier C through E one click away on an internal dashboard. Claiming a partner and logging outreach syncs status to CRM automatically, so managers get visibility without a separate report.
The artifacts
Four pieces make up the full case study: the agent architecture, the internal dashboard, the monthly email, and the functional requirements handed to engineering.