App Analytics

App Analytics

Measure the few things that drive decisions — not everything you can. Over-instrumenting creates noise, privacy risk, and a dashboard nobody reads. This skill is about what to measure; performance telemetry is performance-instruments (MetricKit).

> Test for every metric: if this number moved, would we do something differently? If not, don't track it.

---

Start with one north-star metric

A north-star is the single number that best captures delivered user value — leading, not lagging.

• Good: "weekly active note editors", "sessions completed per user/week", "tracked workouts/week."
• Weak: raw downloads or DAU alone (vanity; doesn't capture value).
• The north-star should predict retention and revenue. Everything else is a supporting or guardrail metric.

---

The funnel (AARRR, pragmatically)

| Stage | Question | Typical metric | |-------|----------|----------------| | Acquisition | Are people arriving? | Impressions → installs (see asc-aso) | | Activation | Did they reach first value? | % completing the "aha" action in session 1 | | Retention | Do they come back? | D1 / D7 / D30 cohort retention | | Revenue | Do they pay? | Trial→paid, conversion, ARPU (see paywall-design, app-store-pricing) | | Referral | Do they bring others? | Invites sent/accepted, rating prompts |

Activation and retention are where most apps actually win or lose. Define your activation event precisely ("created and saved 1 note", "logged 3 days") and measure the % who hit it on day 0 — it's the highest-leverage number for a young app.

Retention is a cohort, not a percentage

Always read retention as cohorts (users who installed in week N, % active in weeks N+1, N+7…). A single "retention %" hides whether you're improving. A flattening retention curve (not decaying to zero) is the signal of product-market fit.

---

Event taxonomy — design it before you instrument

A messy event log is worse than none. Conventions:

• Name consistently: object_action in one tense — note_created, paywall_viewed, subscription_started. Pick a casing and keep it.
• Properties over event explosion: one paywall_viewed event with a source property beats paywall_viewed_onboarding / paywall_viewed_settings / … Same event, structured dimensions.
• Track the funnel-defining events first (activation, key actions, purchase steps), then add detail only when a question demands it.
• Version the schema and write it down — undocumented events rot.
• Capture enough context to segment (platform, plan, locale) without capturing PII.

---

Apple's built-in sources (often enough)

• App Store Connect → App Analytics: impressions, product page views, conversion rate, installs, retention, and sessions — segmented by Source Type (Search / Browse / Referrer). Free, privacy-safe, no SDK. Start here.
• StoreKit / Subscription analytics: trials, conversions, churn, proceeds, by product — the source of truth for monetization (pairs with storekit-purchases).
• MetricKit: performance/diagnostics from the field (launch, hangs, hitches) — route to your dashboards via performance-instruments.

Reach for a third-party analytics SDK only when you need event-level product funnels the built-ins don't give you.

---

Privacy is part of analytics, not a footnote

• Minimize: collect the least that answers your question. Don't log PII or precise location "just in case."
• Anything you collect must be reflected in the privacy nutrition label and, for required-reason/tracking APIs, the privacy manifest — see privacy-manifest. Tracking across apps requires App Tracking Transparency consent.
• Prefer aggregate/on-device where possible. Truthful labels are an App Review and trust requirement.

---

From metric to action

1. Pick the one decision you're trying to inform.
2. Define the metric + target + window (ties to product-spec success metrics).
3. Instrument the minimum events to compute it, with clean names/properties.
4. Read it as cohorts/segments, not a single global number.
5. Change one thing, measure the move, keep or revert. Repeat.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: markdavidgan
• Source: markdavidgan/apple-dev-skills
• License: MIT

Install and usage instructions live in the source repository linked above.