Conversion analytics

Conversion Analytics

Conversion analytics answers: can we measure what users do on the path to value, reliably and without leaking PII? This skill covers detecting your stack, mapping funnels, designing events, implementing tracking, verifying instrumentation, and establishing baselines — the foundation before A/B tests or optimization.

This is a general skill for agent-built projects. Reuse existing analytics; do not bolt on a second system for one funnel unless the repo has none.

---

Quick Start

1. Detect — stack and existing analytics (Step 1)
2. Map funnel — surface → action → outcome (Step 2)
3. Design taxonomy — event names and properties (Step 3)
4. Output table — event plan; no file edits (Step 4)
5. Implement — wire tracking after user approval (Step 5)
6. Verify — events fire, properties present, baseline captured (Step 6)

---

Step 1 — Detect stack and existing analytics

Before designing events, read the repo:

1. Dependencies — package.json, mobile config, wrangler.toml, framework (Next.js, Vite, Expo, etc.).
2. Existing SDK — GA4 (gtag, @next/third-parties), Plausible, PostHog, Mixpanel, Amplitude, Segment, Cloudflare Web Analytics, Firebase Analytics.
3. Current events — search for track(, capture(, plausible(, gtag('event', custom analytics wrappers.
4. Consent / privacy — cookie banners, CMP, beforeSend hooks, DNT handling.
5. Identity — anonymous visitor ID, logged-in user ID, device ID (mobile).

If analytics exists, extend it. If none exists, recommend the lightest fit for the stack (see references/vendor-patterns.md).

---

Step 2 — Map the funnel

For each conversion goal the user cares about, trace:

Surface (page/screen) → User action → Business outcome

Examples:

Goal	Surface	Action event	Outcome event
Signup	Landing hero	cta_click	signup_complete
Trial start	Onboarding step 3	onboarding_step_complete	trial_started
Purchase	Checkout	checkout_started	purchase_complete

Tag priority — P0 events block experiment readiness; P1 nice-to-have; P2 defer.

For naming conventions and anti-patterns, see references/event-taxonomy.md.

---

Step 3 — Design event taxonomy

Every event needs:

• Stable name — object_action (e.g., signup_cta_click, not click)
• Trigger — exact UI moment (button click, form submit success, route change)
• Properties — surface, variant (slot for A/B), source, non-PII context
• No PII — no raw email, name, or free-text user input in properties

Reserve variant or experiment_id + variant for future A/B tests even if not running one yet.

Read references/privacy-compliance.md before implementing.

---

Step 4 — Output format (no edits on first pass)

Do not edit files on the first pass unless the user explicitly says "implement all" or "apply without review."

Executive summary

• Analytics vendor detected (or recommendation if none)
• Funnel goal(s)
• Event count by priority (P0 / P1 / P2)
• Gaps blocking A/B readiness (missing primary metric, no visitor ID, etc.)

Event plan table

Surface	Event name	Trigger	Properties	Priority

Priority: P0 — required for conversion/experiment readiness · P1 — useful context · P2 — defer

Baseline note

State how baseline rates will be captured once events are live (e.g., "7-day signup_complete / landing_view rate").

Partial approval

User may say "implement P0 only" or "implement rows 1 and 3" — follow exactly.

---

Step 5 — Implement (after approval)

1. Edit approved rows only
2. Reuse existing analytics SDK and naming patterns in the repo
3. Attach variant property at capture sites used in experiments
4. Do not duplicate tracking (one event per user action)
5. Respect consent — no tracking before consent when CMP is present

Stack-specific snippets: references/vendor-patterns.md.

---

Step 6 — Verify

Before calling instrumentation done:

• P0 events fire in dev/staging (document how to trigger each)
• Properties match the approved table (no missing surface)
• No PII in payload (spot-check network tab or vendor live view)
• Duplicate fires ruled out (double-click, strict mode, re-renders)
• Baseline window stated (e.g., collect 7–14 days before sample-size math)

Hand off to ab-testing when the user wants to run experiments — baseline rate + primary metric must exist first.

---

Related skills

Skill	When
ux-writer	CTA/label copy before naming events tied to UI text
ab-testing	After P0 events and baseline exist

npx skills add shahrulestar/skills --skill ux-writer
npx skills add shahrulestar/skills --skill ab-testing

---

---

Additional resources

File	Contents
references/event-taxonomy.md	Naming, properties, anti-patterns
references/vendor-patterns.md	GA4, Plausible, PostHog, Workers, React Native
references/privacy-compliance.md	Consent, PII, GDPR-conscious defaults

---

Rules

• Reuse existing analytics — never add a second vendor for one funnel without asking.
• Table first, implement after approval — same as ux-writer.
• No PII in event properties.
• Reserve variant for future A/B tests.
• State baseline collection window before sample-size math.
• If no analytics exists, recommend one lightweight option — don't install three.