Engineering docs

Product analytics on ourselves (PostHog)

Last updated Oct 20, 2018 · 5 sections

On only when NEXT_PUBLIC_POSTHOG_KEY is set on Vercel; web/lib/analytics.ts is a no-op otherwise (dev, CI, previews). Host: NEXT_PUBLIC_POSTHOG_HOST=eu (default, data stays in the EU) or us, or a full URL for a self-hosted instance.

Privacy posture (CLAUDE.md rule 3)

No identify(), ever. No email, phone, business name or IP (ip: false, person_profiles: "never"). PostHog sees an anonymous device id per browser and the events below.
Autocapture, pageviews and session recording are off. Only the seven explicit events fire.
Properties are counts and enum values only. Never add free-text fields.
The SDK is loaded lazily, so a build without the key ships zero analytics bytes.

Events

Event	Fired from	Properties
`signup_completed`	`/signup` after the account is created	—
`creative_uploaded`	`/creatives` after a file upload succeeds	`kind`: `image` \| `video`
`game_ad_created`	game editor, first save of a new game ad	`template`: `cricket` \| `spin` \| `wheel` \| `catch`
`booking_created`	`/book` after "Confirm & book" succeeds	`slots` (total), `days`, `screens`
`approval_decided`	`/ops/approvals` after approve/reject succeeds	`decision`: `approved` \| `rejected`, `ops: true`
`report_viewed`	campaign report opened (`/campaigns/[id]`)	—
`wall_paired`	`/ops/pair` when the player confirms the bind	—

Every event also carries app: "web" and build (the Vercel commit SHA) via register().

The funnel to watch

Build one PostHog Funnel insight, conversion window 14 days, strict order:

signup_completed
creative_uploaded or game_ad_created ("first creative")
booking_created ("first booking")
approval_decided where decision = approved ("first approved")

Read it weekly. Pre-launch targets to argue about: ≥ 60 % of sign-ups upload something the same day; ≥ 40 % of uploaders book within 14 days; approval within one working day (ops SLA, measured on the backend, not here).

"Days to first booking"

Because we do not identify users, this is per anonymous browser, which is fine for a self-serve funnel. In PostHog: Trends → booking_created → breakdown "first time event" is not enough; use the funnel above and read time to convert between step 1 and step 3 (median and p90). Watch the median: if it drifts past 3 days the booking flow has a friction problem, not a demand problem.

Also worth a dashboard tile

creative_uploaded by kind vs game_ad_created by template — which ad formats sell.
booking_created → sum of slots and avg days — order size over time.
approval_decided rejected share — if > 15 %, the upload-time checks need to catch more.
wall_paired count per week — screens coming online.

Not here

Audience numbers, plays, revenue — those come from the ledger (/v1/metrics/*, /ops/screens Metrics tab), never from the analytics tool.