software-tester-skills

Analytics & Tracking Testing

Validating analytics and tracking — GA4/GTM dataLayer, Segment/Amplitude/Mixpanel events, pixels (Meta, TikTok), server-side tagging, consent mode, UTM/attribution, event schema/tracking plan conformance, and data parity with the backend.

If tracking is wrong, every product decision built on it is wrong. Treat the tracking plan as a contract and test it like an API.

1. Tracking plan (source of truth)

event name · trigger · properties (name/type/enum/required) · owner · destinations. Naming: object_action snake_case (checkout_started). Version it in git; generate JSON Schema for validation (Segment Protocols, Avo, Amplitude Data).

2. Verify in the browser

// Console: watch the dataLayer
window.dataLayer.push = new Proxy(window.dataLayer.push, { apply(t, s, a){ console.log('DL', JSON.stringify(a[0])); return Reflect.apply(t, s, a);} });
  • GTM Preview / Tag Assistant; GA4 DebugView (?debug_mode=1); Segment Debugger; Meta Pixel Helper; chrome://net-export or DevTools Network filter collect|g/collect|track.
  • Automate: Playwright intercepts requests and asserts payloads.
const hits: any[] = [];
page.on('request', r => { if (/google-analytics\.com\/g\/collect|\/v1\/track/.test(r.url())) hits.push(r.postData() ?? r.url()); });
await page.getByRole('button',{name:'Add to cart'}).click();
await expect.poll(() => hits.some(h => h.includes('add_to_cart'))).toBeTruthy();

3. What to assert

AreaChecks
FiringFires once per action (no dupes on SPA re-render/route change); not on load unless intended; fires after success not on click
PayloadNames/types match plan; currency ISO code, value numeric, items[] complete; no PII (email, phone) unless hashed & allowed
Identityuser_id set on login, cleared on logout; anonymous→known stitching; cross-domain _gl linker
Page/routeSPA virtual pageviews, correct page_location, canonical vs UTM-cluttered URL
AttributionUTM params captured, persisted across pages, not overwritten by internal links; referrer exclusions (payment gateways)
ConsentNothing fires before opt-in (EU); Consent Mode v2 defaults denied; withdrawal stops tags; CMP banner accessible
Ecommerceview_item → add_to_cart → begin_checkout → purchase funnel with same transaction_id (dedupe)
Server-sideServer events dedupe with client via event_id; PII hashing SHA-256; CAPI/Measurement Protocol response OK
PerformanceTags async, not blocking LCP; total third-party weight
Ad blockers/ITPGraceful degradation; first-party endpoint

4. Parity (weekly automated)

Compare analytics revenue/orders/signups vs backend DB for the same window. Expect small drift (ad-blockers ~5–15%); alert when > threshold or trend changes after a release.

5. Regression net

Contract test each event schema in CI (JSON Schema + example payload from E2E); block deploys that rename events or drop required props; release checklist step "tracking verified".

e2e-testing, payment-ecommerce-testing, compliance-testing, seo-testing, observability-testing