Churn Auditor

Dashboard Overview
Ai Reasoning Trace
Action-Executor
Evidence-Panel
Before-After
Solution-section

Inspiration

What it does

How we built it Inspiration

Every SaaS founder knows the pain: a customer cancels, and you're left wondering why. The manual investigation takes 2+ hours—digging through support tickets, usage data, exit surveys—only to end up with a vague diagnosis and no clear next step.

We asked: what if AI could do this entire investigation in 60 seconds AND execute the recovery plan automatically?

What it does

Churn Auditor is an Action Era AI tool that doesn't just analyze churn—it acts on it.

When a customer cancels on Stripe, Churn Auditor:

Collects evidence — pulls support tickets, usage metrics, and exit survey responses
Reasons through the diagnosis — uses Gemini 3 Pro to triangulate the real cause (not just the stated reason)
Outputs a structured dossier — with confidence scores, evidence quotes, and save probability
Executes recovery actions — automatically creates Linear tickets, drafts personalized winback emails, and alerts the team on Slack

Total time: 47 seconds. Manual process: ~2 hours.

How we built it

We built Churn Auditor with a modern, lean stack:

Next.js 14 (App Router) with TypeScript for type safety
Gemini 3 API as the AI backbone:
- Flash for quick triage (is this worth deep analysis?)
- Pro for multi-step reasoning and diagnosis
- Structured outputs for typed JSON dossiers
- Tool calling for the Action Era — executing real actions
Tailwind CSS + shadcn/ui for a polished dashboard
Server-Sent Events (SSE) for real-time streaming of AI reasoning
Zod for runtime validation of all external data

The key innovation is making the AI's reasoning visible. Users watch the analysis unfold step-by-step, see evidence being evaluated, and watch actions execute in real-time.

Challenges we ran into

Making tool calling feel magical — The technical implementation was straightforward, but making it visible and impressive required careful UI choreography. We added progress bars, animations, and a live timer to make the "Action Era" tangible.
Balancing speed vs. depth — We use model routing (Flash → Pro) to keep analysis fast while maintaining quality. Flash handles triage in milliseconds; Pro does the deep reasoning.
Streaming structured output — Gemini's streaming API + JSON mode required careful handling to parse partial responses and update the UI progressively.

Accomplishments that we're proud of

60-second end-to-end flow — From Stripe cancellation to executed recovery actions
Visible AI reasoning — Users see the chain of thought, not just the answer
Action Era showcase — Tool calling that actually does things, not just recommends

What we learned

The biggest insight: AI tools that analyze are table stakes. AI tools that ACT are the future. Gemini 3's tool calling capability unlocks a new category of applications where AI doesn't just inform—it executes.

What's next for Churn Auditor

Real integrations — Connect to actual Linear, Slack, and email APIs (currently mocked for demo stability)
Predictive churn — Flag at-risk customers before they cancel
Playbook library — "Companies like you save 23% of churned customers with this approach"
Stripe App Marketplace — Native distribution to millions of SaaS companies