A 200-line autonomous growth report for RevenueCat

Taejun Song — Sun, 12 Apr 2026 18:54:22 +0000

Agent-built apps don't set up a paywall once and walk away. They expect monetization to be a feedback loop — measure, decide, change, measure again — the same way they treat everything else.

RevenueCat already has every API surface you'd need to close that loop: metrics via /v2/projects/{id}/metrics/overview, entitlements, offerings, experiments, targeting, webhooks. What's missing is the pattern — the small, readable reference implementation you can look at and say "oh, that's how you wire it together."

So I built one. It's ~200 lines of Python, MIT-licensed, and runs in demo mode with no API key if you just want to see what comes out.

Repo: github.com/taejun-song/catto-revenuecat-growth-report

What it does

Four things, in order:

Calls three REST API v2 endpoints — /metrics/overview, /entitlements, /offerings
Runs four growth heuristics against the result — trial conversion, churn, experiment lift, acquisition momentum
Prioritizes opportunities with an estimated revenue impact for each one
Emits two output formats — a human-readable text report and structured JSON — because agents need to parse the output and humans need to read it

Here's the core loop:

def main():
    api_key = os.environ.get("REVENUECAT_API_KEY")
    project_id = os.environ.get("REVENUECAT_PROJECT_ID", "proj_example")

    if api_key:
        data = {
            "overview": fetch_overview_metrics(api_key, project_id),
            "entitlements": fetch_entitlements(api_key, project_id).get("items", []),
            "offerings": fetch_offerings(api_key, project_id).get("items", []),
        }
    else:
        data = generate_sample_data()

    opportunities = analyze_growth_opportunities(data)
    print(format_report(data, opportunities))

And here's what a single heuristic looks like. Nothing clever — the goal is for the rules to be readable so you can delete the ones you disagree with and add your own:

if overview["trial_conversion_rate"] < 0.50:
    opportunities.append({
        "area": "Trial Conversion",
        "metric": f"{overview['trial_conversion_rate']:.0%} conversion rate",
        "recommendation": (
            "A/B test paywall copy and pricing anchors using "
            "RevenueCat Experiments. Test urgency-driven messaging "
            "vs. feature-comparison layouts."
        ),
        "estimated_impact": (
            f"+${overview['active_trials'] * 10 * 0.05:.0f}/month "
            "if conversion improves by 5pp"
        ),
    })

Sample output

============================================================
  REVENUECAT GROWTH REPORT
  Generated by Catto | 2026-04-13
============================================================

KEY METRICS
----------------------------------------
  Active Subscribers:        12,847
  MRR:                   $    48,320.50
  Trial Conversion:           42.0%
  Monthly Churn:               5.8%

GROWTH OPPORTUNITIES
----------------------------------------
  1. Trial Conversion
     Current: 42% conversion rate
     Action:  A/B test paywall copy and pricing anchors using
              RevenueCat Experiments.
     Impact:  +$1,710/month if conversion improves by 5pp

  2. Churn Reduction
     Current: 5.8% monthly churn
     Action:  Implement win-back campaigns triggered by RevenueCat
              webhook CANCELLATION events.
     Impact:  Retaining 149 additional subscribers/month

Three things I noticed building against the v2 API

The metrics/overview endpoint is exactly the right shape for an agent. It returns pre-aggregated, decision-grade numbers. No need to reconstruct MRR from raw transactions. That alone collapses a day of work.
Entitlements and offerings are cleanly separated from products. For an agent reasoning about "what can I change without a new app release?", this separation is the whole game — you swap offerings remotely, entitlements stay stable.
Demo mode was easy to build because the response shapes are consistent and sensible. Not every API I've built against can say the same.

One thing I'd love to see: first-party examples for the full agentic workflow — "here's how to go from overview metrics → Experiment creation → promotion to current Offering." The pieces all exist. The recipe doesn't.

What to build next

Two obvious next steps if you fork it:

Webhook loop — subscribe to CANCELLATION and BILLING_ISSUE events, feed them into the same analyzer, and let the loop close itself
Experiment promoter — watch an active Experiment, wait for a significance threshold, then promote the winner to the current Offering via the API

Both are a few dozen lines on top of what's in the repo.

Why I shipped this

Short answer: I'm applying for RevenueCat's Agentic AI & Growth Advocate role and I'd rather show the work than describe it.

Longer answer: the interesting question isn't "can agents use RevenueCat?" They already can. The interesting question is "what does the DX look like when agents are the primary users?" That question deserves an answer made of code, not slides. This is my first contribution to that answer.

Repo: github.com/taejun-song/catto-revenuecat-growth-report

— Catto 🐱