Analytics & Tracking
Understand how users interact with your product through event tracking, funnels, and retention analysis.
Options
Tracking Scope*
Analytics Provider*
Decision Points
Do you need to track individual user behavior (not just aggregate)?
If yes
Enable user events and choose a third-party provider with identity stitching.
If no
Page views with a self-hosted provider (Plausible) is sufficient and privacy-friendly.
Are you in a privacy-sensitive market (healthcare, finance, EU users)?
If yes
Use self-hosted analytics to avoid third-party data processors and simplify GDPR compliance.
If no
Third-party providers give better tooling with minimal compliance overhead.
Product analytics (behavior) vs business analytics (revenue/funnels)?
If yes
Product analytics: Amplitude/Mixpanel/PostHog for events, funnels, retention. Business: pair with a warehouse (BigQuery) + BI (Metabase).
If no
If you only need one, pick product analytics first — behavior insights drive most decisions.
Self-serve exploration (SQL/dashboards) or fixed reports?
If yes
Pipe events into BigQuery/ClickHouse and layer Metabase or Hex — lets non-engineers write ad-hoc queries.
If no
Canned dashboards in Amplitude/Mixpanel are faster to ship and cover the 80% case.
Do data scientists need raw warehouse access?
If yes
Use Segment/RudderStack or a reverse-ETL path to land raw events in BigQuery/Snowflake with full fidelity.
If no
Stay with a hosted analytics tool — raw-access is dead weight for product teams.
Is real-time dashboarding required or is 24h acceptable?
If yes
Choose a streaming stack (ClickHouse + Kafka, or PostHog) — most warehouse-based setups have 1–24h latency.
If no
Daily batch (warehouse + dbt) is cheaper and simpler for most reporting.
Do you need server-side tracking in addition to client-side?
If yes
Add server-side events for revenue, subscriptions, and auth — resistant to ad-blockers and more reliable than JS.
If no
Client-side only is fine for early-stage product analytics on consumer apps.
Will you use a CDP (Segment, RudderStack) for fan-out?
If yes
A CDP lets you send events to multiple destinations (analytics, CRM, warehouse) from a single pipe — worth it once you have 3+ destinations.
If no
Direct SDK integrations are cheaper and simpler with one or two destinations.
Do you need cohorts/funnels/retention (Amplitude, Mixpanel)?
If yes
Mixpanel/Amplitude/PostHog are purpose-built for this — skip the warehouse-BI detour.
If no
Plausible/Umami are sufficient for pageview and top-line numbers.
Are you subject to GDPR/CCPA opt-out requirements?
If yes
Ship a consent banner (Iubenda, OneTrust) and gate all trackers behind opt-in — self-hosted analytics (Plausible) skips the banner in most EU cases.
If no
Still honor Do Not Track, but a global banner is not required.
Must PII be redacted from event streams automatically?
If yes
Add a redaction middleware at the SDK or CDP layer — allowlist fields, block email/phone/tokens before they leave the client.
If no
Still advisable as defense-in-depth, but not a hard gate.
Do you need session replay (FullStory, LogRocket)?
If yes
PostHog bundles replay with analytics; FullStory/LogRocket are purpose-built — budget for storage and mandatory PII redaction.
If no
Heatmaps or funnels cover most UX debugging without the privacy surface.
Will events be used for usage-based billing?
If yes
Use a metering-specific store (Orb, Metronome) or a dedicated events table with idempotency keys — not your analytics tool.
If no
Analytics tools are not billing-grade — never bill from them directly.
Do you need attribution tracking (multi-touch, last-touch)?
If yes
Add UTM capture + a session/touchpoint model — Amplitude and Mixpanel both offer multi-touch attribution views.
If no
Skip it until marketing asks — attribution is hard to get right and easy to mislead with.
Should events be sampled at high volume?
If yes
Sample non-revenue events (scroll, hover) at 1–10% — keep revenue and conversion events at 100%.
If no
Full-fidelity is fine until event volume hits provider-pricing tiers.
Do you need to join analytics with production data regularly?
If yes
Warehouse-based stack (BigQuery + dbt) lets you join events with subscriptions, users, and revenue — hosted tools can not.
If no
Hosted analytics is enough; skip the warehouse overhead.
Do customers see their own analytics (customer-facing reporting)?
If yes
Use an embedded-analytics tool (Metabase embedded, Cube.dev, Explo) — do not expose your internal dashboards.
If no
Internal analytics stays internal — simpler security model.
Do you need A/B test instrumentation?
If yes
PostHog, Statsig, or LaunchDarkly Experiments wire flags + analytics together — do not roll your own significance testing.
If no
Skip until you have enough traffic for tests to reach significance.
Tradeoffs
User data is shared with vendor; may require GDPR consent flow
Significant storage cost; must redact sensitive fields (passwords, PII)
Full data ownership and unlimited retention, but requires infrastructure expertise