Good to Great: analytics integration for 'what to build' insights

[maxfindel]

Context

Shreyas Doshi (ex-Stripe, Twitter, Google) defines the Good→Great PM gap:

Good PM: Fixes UI issues found in usability tests
Great PM: Uses diverse research to inform what to build, not just fix

FlowChad currently sits on the "Good" side — it walks flows, finds bugs, categorizes friction. That's valuable but reactive. The "Great" leap is connecting walk results to real user behavior data to answer: "Is this flow even the right thing to optimize?"

The Gap

A flow walk tells you what's broken. Analytics tell you:

• How many users actually hit this flow (is it worth fixing?)
• Where they drop off (which step loses people?)
• What they do instead (do they rage-click? go back? leave?)
• Whether fixes worked (did the drop-off rate change after your PR?)

Without this, FlowChad is a QA tool. With it, it's a product decision tool.

Proposal

1. Analytics MCP integration

FlowChad setup already detects Mixpanel/PostHog (Phase 1b in flowchad-setup). Wire it up:

# config.yml
analytics:
  provider: posthog
  mcp: true
  funnel_id: "signup-flow-123"  # optional: link flow to existing funnel

When walking a flow, if analytics MCP is available:

• Pull funnel data for the matching flow
• Annotate each step with real drop-off rates
• Flag steps where >20% of users abandon

2. Enriched friction reports

### Step 3: click submit
- **Status:** FRICTION (2.8s, threshold 2s)
- **Drop-off:** 34% of users abandon at this step (PostHog, last 30 days)
- **Rage clicks:** 12% of sessions show repeated clicks on this button
- **Impact:** ~1,200 users/month lost at this step

This changes the conversation from "the button is slow" to "the slow button loses 1,200 users/month."

3. LNO-informed priority

Shreyas's LNO Framework (Leverage / Neutral / Overhead) applied to flows:

Flow Priority	Analytics Signal	LNO Category
P0 (critical)	High traffic + high drop-off	Leverage — 10-100x return
P1 (important)	Medium traffic or low drop-off	Neutral — 1x return
P2 (nice-to-have)	Low traffic	Overhead — do if time permits

/flow-suggest should use real traffic data to re-rank suggestions, not just severity. A Cosmetic issue on a P0 flow with 50k MAU matters more than a Critical issue on a settings page with 200 MAU.

4. Before/after measurement

When a fix PR merges, re-walk the flow and compare:

• Timing improvement (from /flow-diff)
• Drop-off rate change (from analytics, 1-2 weeks after deploy)
• Generate a "fix impact" summary: "Step 3 submit time: 2.8s → 0.9s. Drop-off: 34% → 18%. Estimated 960 users/month recovered."

This closes the loop: find → fix → measure → prove.

Scope

• Wire up Mixpanel MCP in /flowchad-setup (config + .mcp.json)
• Wire up PostHog MCP in /flowchad-setup
• flow-walk: query funnel data per step if analytics available
• flow-report: annotate findings with drop-off rates and traffic volume
• flow-suggest: re-rank by LNO using real traffic data
• flow-diff: include before/after analytics comparison
• Add analytics section to knowledge/metrics-primer.md

References

• Shreyas Doshi: Good vs Great PMs
• Shreyas Doshi: LNO Framework
• Lenny Rachitsky: What is PM
• Teresa Torres: Continuous Discovery