Product Analytics
Product Analytics
The analytical discipline of measuring what users do inside a product, why, and whether it creates value — then feeding that back into product decisions. Distinct from generic web analytics (page-level) and from instrumentation (the plumbing). Product analytics is event-centric, user-centric, and decision-oriented.
This skill covers the analysis layer. Leave to adjacent skills:
- Cohort/retention curves, N-day/unbounded retention, retention math →
da-34-cohort-retention-analytics - SDK wiring, autocapture vs manual, identity stitching plumbing →
da-3-2-7-web-app-analytics-instrumentation - Experiment statistics, p-values, CUPED, sequential testing theory →
da-12-ab-testing-causal-inference
Overview
A product-analytics practice answers four recurring questions: (1) Are users reaching value (activation)? (2) Do they keep coming back and going deeper (engagement/adoption)? (3) Where do they drop off (funnels/paths)? (4) Is the whole thing growing toward a single meaningful outcome (North Star)? The analytical quality of every answer is capped by the quality of the event taxonomy underneath it — so taxonomy and governance come first, not last.
Core Concepts
1. Event taxonomy & tracking plans
A taxonomy is the hierarchical naming + classification scheme for events and properties so a platform can produce comparable insights. Design it deliberately before instrumenting.
- Object-Action naming: pick objects (
Song), define actions (Played,Paused), agree a tense (past tense recommended), produceSong Played. Alternativelyverb_nounsnake_case (checkout_completed). Pick one and enforce it. (Amplitude event taxonomy, 2024; Avo naming conventions, 2025; Heap naming conventions, 2024) - Parameterize, don’t proliferate: one
Add to Cartevent with acampaignproperty — neverAdd to Cart Summer Saleas a separate event. (Amplitude data planning playbook, 2025) - Tracking plan = the central contract: every event, its properties, data types, owner, trigger, and examples. The spreadsheet is the legacy form; dedicated tools (Avo) version-control it. (Amplitude tracking practices, 2024)
- Goldilocks granularity: too few events = blind spots; too many = noise and maintenance debt. Track events that map to decisions.
2. North Star metric (NSM) framework
A single metric that best captures the value customers get, that product/marketing can influence, and that leads revenue.
- The NSM is an output/outcome — you should not be able to move it directly. You move it through 3–5 inputs that teams influence day-to-day. (Amplitude North Star Playbook, 2024)
- A metric tree decomposes NSM → inputs → initiatives, so every team sees how their work ladders up. (Amplitude NSM & inputs, 2024)
- Good NSM = leading indicator of value (e.g. “weekly active collaborators”), not a vanity output (e.g. raw signups or revenue itself). (Amplitude good vs bad NSM, 2024)
3. Funnel & conversion analysis
Map an ordered multi-step flow, measure step-to-step conversion, diagnose the biggest drops.
- Three parts: define the ordered steps → measure conversion between steps → diagnose the leakiest step (session replay, segmentation, qual). (Statsig funnel analysis, 2025; UXCam conversion funnel guide, 2026)
- Keep funnels to 4–7 ordered steps; longer funnels hide where the real loss is. (Count funnel conversion, 2025)
- Choose a conversion window deliberately (e.g. 7-day signup→activation). The window changes the number — state it.
- Segment the funnel (source, device, plan, cohort) — an aggregate funnel almost always masks a segment-specific cliff. (Userpilot conversion funnel, 2025)
4. Activation & the “aha moment”
Activation = the set of early actions that correlate with later retention. The “aha moment” is when the user internalizes core value; the activation metric is its measurable proxy.
- Find it empirically: test event groups and frequencies (e.g. “watched ≥5 replays” beat “watched 1”). PostHog’s activation metric was “set a replay filter ≥1 and watched ≥5 replays” because it maximized retention. (PostHog activation metrics, 2024)
- The magic number is a frequency threshold within a time window (Facebook’s “7 friends in 10 days” archetype). Validate with odds-ratio/correlation against retention, not eyeballing. (Amplitude aha moment, 2024; Statsig spot aha moment, 2024)
- Correlation ≠ cause: a high-retention behavior may be a symptom of an already-engaged user. Treat the activation metric as a hypothesis to test via experiment, not a law.
5. Feature adoption (breadth / depth / time / duration)
- Adoption rate = users who used the feature ÷ active users × 100. ~24–28% is a healthy core-feature band. (Userpilot feature adoption metrics, 2025; Artisan benchmarks, 2025)
- Breadth = how many users reach it (reach). Depth = how intensively they use it once there (value delivery). Low depth = value problem, not discovery problem.
- Time to adopt = speed to value after first exposure; adoption typically builds over 30–90 days — don’t kill a feature on week-one numbers. Duration = whether usage persisted into a habit. (Plane measuring feature adoption, 2025; Appcues adoption metrics, 2024)
6. Engagement & stickiness (DAU/WAU/MAU)
- DAU/WAU/MAU = unique users in 1/7/30-day windows. Stickiness = DAU/MAU (≈ days used per month / 30). (Gainsight DAU/MAU, 2024)
- Benchmarks are product-shape dependent: social/messaging 50–80%, productivity 40–60%, fintech/e-commerce 10–30%. Don’t compare across categories. (Mixpanel MAU benchmarks, 2026; Statsig DAU/MAU, 2025)
- Use WAU/MAU for async products (content, newsletters, docs) where daily use isn’t the natural cadence. (Userpilot DAU/WAU/MAU, 2025)
7. Session & path analysis
- Session = a visit; ends after an inactivity timeout (commonly 30 min on web). Session duration is a depth metric that complements frequency metrics. (Amplitude session duration, 2024)
- Direction of “good” is context-dependent: long sessions = engagement for content; long sessions = friction for transactional/banking apps. GA4 now favors engaged time per session over raw duration (handles background tabs). (GA4BigQuery sessions deep dive, 2025; PostHog session metrics, 2024)
- Path analysis = aggregated flows of the actual sequences users take (not a predefined funnel). Use it for discovery (“what do users do before converting / before churning?”), then formalize findings into funnels. Amplitude folds Pathfinder into Journeys, which adds drop-off and per-user paths that raw path charts lack. (Amplitude Journeys, 2025)
8. Metric frameworks: AARRR vs HEART
- AARRR (Pirate Metrics) — Acquisition, Activation, Retention, Revenue, Referral. Lifecycle/growth lens: “is the business growing?” (Amplitude pirate metrics, 2024; PostHog AARRR funnel, 2024)
- HEART (Google, 2010) — Happiness, Engagement, Adoption, Retention, Task success. UX-quality lens: “is the experience good?” Each dimension pairs with Goals-Signals-Metrics. (Productcompass analytics playbook, 2025; Ideaplan HEART vs AARRR, 2025)
- Pick by question: AARRR for growth orgs/PLG; HEART for UX-led orgs. NSM sits above both as the single rallying output. (Hyperact product metrics frameworks, 2025)
9. Experimentation operations (the ops layer, not the stats)
- Running product experiments at scale needs operational glue: feature flags ↔ analytics ↔ experiment readouts sharing one event pipeline so metrics auto-populate. Modular stacks (separate flag/analytics/experiment products) add setup friction. (PostHog vs Statsig, 2025; Statsig vs PostHog, 2025)
- Operational essentials: pre-registered primary metric tied to the tracking plan, automated power/sample-size estimation, guardrail metrics, and a defined readout cadence. (Statistical validity itself →
da-12-ab-testing-causal-inference.) (ProductQuant PostHog experiments, 2025)
10. Governance & data quality
- Bad data silently corrupts every metric above. Govern with: a versioned tracking plan, schema validation before events hit production, and ongoing observability comparing live events vs the plan to catch schema drift. (Avo data observability, 2025)
- Assign owners per event/property; route changes through review (branch reviews, Slack notifications). (Avo actionable ownership, 2025)
- Schedule quarterly taxonomy reviews with PM + analytics + marketing to retire dead events and absorb new needs. (Amplitude tracking practices, 2024)
Tools / Frameworks
| Tool | Strength | Notes (2025–26) |
|---|---|---|
| Amplitude | Behavioral cohorts, governance, metric trees, Journeys | Best for analysis depth + cross-functional scale; MTU pricing. (Amplitude best tools, 2026) |
| Mixpanel | Fast event analytics, clean funnels/reports | Relaunched experimentation + added flags/replay late 2025; event-based pricing. (Cotera comparison, 2026) |
| PostHog | All-in-one for dev-led teams (analytics + flags + experiments + replay + warehouse) | Modular, open-source, self-host option. (PostHog alternatives, 2025) |
| Heap | Autocapture — records everything, define events retroactively | Fast start, less upfront taxonomy discipline. (PostHog Heap alternatives, 2025) |
| June | Pre-built SaaS company-level reports on top of Segment | Lightweight, B2B-oriented; thinner than the majors for deep analysis. |
| Avo | Tracking-plan governance + Inspector observability | Sits upstream of the analytics tool to guarantee data quality. (Avo, 2025) |
| Statsig | Experimentation-first unified pipeline | Strong ops/experiment scale; CUPED/sequential built in. (Statsig vs PostHog, 2025) |
Frameworks summary: Taxonomy (Object-Action) → NSM + inputs (metric tree) → lifecycle lens (AARRR) or UX lens (HEART) → activation/aha → adoption (breadth/depth/time) → funnels & paths → governance loop.
Methodology (end-to-end)
- Frame the decision — what product question are we answering? (Don’t start from “what can we track?”)
- Define the taxonomy — Object-Action events + properties, owners, in a versioned tracking plan.
- Pick the NSM + 3–5 inputs; draw the metric tree.
- Instrument & validate (hand to instrumentation/governance) — verify live events match the plan.
- Establish activation — find the aha-moment metric empirically; validate against retention.
- Build core funnels (4–7 steps, stated window) and run path analysis to discover real journeys.
- Measure adoption & engagement — adoption rate, breadth/depth, DAU-WAU-MAU/stickiness, segmented.
- Experiment to move inputs; read out against pre-registered metrics.
- Review quarterly — prune the taxonomy, re-validate the NSM, refresh benchmarks.
Practical Patterns
- Segment before you conclude. Every aggregate metric (funnel, stickiness, adoption) hides a segment story. Break by source, plan, platform, cohort.
- Tie each metric to a decision. If no decision changes based on a metric, stop tracking it.
- Activation metric = retention’s leading indicator. Optimize activation to move retention upstream of churn.
- Discover with paths, confirm with funnels. Paths surface the unknown; funnels measure the known.
- State your windows. Conversion window, active-user window, adoption window — all change the number.
- NSM is an output you steer via inputs, never a dial you turn directly.
Anti-Patterns
- Vanity NSM. Picking raw revenue or total signups as the North Star — not a leading value indicator. (Amplitude good vs bad NSM, 2024)
- Event sprawl / inconsistent names.
Song PlayedvsSong_Playedfrom different teams destroys comparability. Parameterize and govern. (Heap naming, 2024) - Funnel theater. Reporting drop-off % without diagnosing why the leak happens.
- Cross-category benchmark abuse. Comparing a fintech app’s 22% DAU/MAU to a social app’s 60% as if underperforming.
- Killing features on week-one adoption before the 30–90 day adoption curve plays out.
- Treating session duration as universally “more is better.” Wrong for transactional products.
- Correlation-as-causation on the aha moment. Shipping a forced onboarding step because a behavior correlated with retention, without an experiment.
Troubleshooting
- Numbers differ between two tools/dashboards → almost always different windows, dedup logic, or event definitions. Reconcile against the tracking plan first.
- Stickiness dropped overnight → check for a taxonomy/SDK change (broken event) before concluding behavior changed; use observability (Avo Inspector) to spot schema drift. (Avo data observability, 2025)
- Funnel conversion looks impossibly high/low → check the conversion window and whether steps are strictly ordered vs “any order”.
- NSM flat while inputs move → inputs may be mis-chosen (don’t actually drive the output) — re-derive the metric tree.
- Feature “failing” → separate breadth from depth: low reach is a discovery/onboarding fix; low depth is a value/UX fix.
- Path analysis is unreadable → too many distinct events; collapse to a smaller event set or anchor on a start/end event.
References
- Amplitude — Event taxonomy (https://amplitude.com/explore/data/event-taxonomy), 2024
- Amplitude — Data planning playbook (https://amplitude.com/docs/data/data-planning-playbook), 2025
- Amplitude — Analytics tracking practices (https://amplitude.com/blog/analytics-tracking-practices), 2024
- Avo — Naming conventions (https://www.avo.app/docs/data-design/best-practices/naming-conventions), 2025
- Heap — Naming conventions (https://www.heap.io/blog/naming-conventions-and-their-place-in-analytics), 2024
- Amplitude — North Star framework (https://amplitude.com/books/north-star/about-north-star-framework), 2024
- Amplitude — NSM & inputs (https://amplitude.com/books/north-star/amplitudes-north-star-metric-and-inputs), 2024
- Amplitude — Good vs bad NSM (https://amplitude.com/blog/good-bad-north-star-metric), 2024
- Statsig — Funnel analysis in product analytics (https://www.statsig.com/perspectives/funnel-analysis-product-analytics), 2025
- UXCam — Conversion funnel analysis guide (https://uxcam.com/blog/conversion-funnel-analysis/), 2026
- Count — Funnel conversion analysis (https://count.co/metric/funnel-conversion-analysis), 2025
- Userpilot — Conversion funnel analysis (https://userpilot.com/blog/conversion-funnel-analysis/), 2025
- PostHog — How we found our activation metric (https://posthog.com/product-engineers/activation-metrics), 2024
- Amplitude — The aha moment (https://amplitude.com/blog/aha-moment), 2024
- Statsig — Spot your product’s aha moment (https://www.statsig.com/perspectives/spot-product-aha-moment-analytics), 2024
- Userpilot — Feature adoption metrics (https://userpilot.com/blog/feature-adoption-metrics/), 2025
- Artisan — Feature adoption benchmarks 2025 (https://www.artisangrowthstrategies.com/blog/feature-adoption-metrics-top-benchmarks-2025), 2025
- Plane — Measuring feature adoption (https://plane.so/blog/measuring-feature-adoption-and-usage-metrics-funnels-and-examples), 2025
- Appcues — Product adoption metrics (https://www.appcues.com/blog/success-with-product-adoption-metrics), 2024
- Gainsight — DAU/MAU guide (https://www.gainsight.com/essential-guide/product-management-metrics/dau-mau/), 2024
- Mixpanel — MAU definition & 2026 benchmarks (https://mixpanel.com/blog/mau/), 2026
- Statsig — Understanding DAU/MAU (https://www.statsig.com/perspectives/understanding-daumau-key-metrics-for-product-success), 2025
- Userpilot — DAU/WAU/MAU explained (https://userpilot.com/blog/dau-wau-mau/), 2025
- Amplitude — Session duration glossary (https://amplitude.com/glossary/terms/session-duration), 2024
- GA4BigQuery — Understanding sessions in GA4 (https://ga4bigqueryblog.com/2025/08/25/understanding-sessions-in-google-analytics-4-ga4-a-deep-dive/), 2025
- PostHog — Session metrics tutorial (https://posthog.com/tutorials/session-metrics), 2024
- Amplitude — Journeys / paths (https://amplitude.com/docs/analytics/charts/journeys/journeys-understand-paths), 2025
- Amplitude — Pirate metrics (AARRR) (https://amplitude.com/blog/pirate-metrics-framework), 2024
- PostHog — AARRR pirate funnel (https://posthog.com/product-engineers/aarrr-pirate-funnel), 2024
- Productcompass — Product analytics playbook (https://www.productcompass.pm/p/the-product-analytics-playbook-aarrr), 2025
- Ideaplan — HEART vs AARRR (https://www.ideaplan.io/compare/heart-vs-aarrr), 2025
- Hyperact — Product metrics frameworks (https://www.hyperact.co.uk/blog/product-metrics-frameworks), 2025
- PostHog — PostHog vs Statsig (https://posthog.com/blog/posthog-vs-statsig), 2025
- Statsig — Statsig vs PostHog (https://www.statsig.com/vs/posthog), 2025
- ProductQuant — PostHog A/B experiments setup (https://productquant.dev/blog/setup-posthog-ab-experiments/), 2025
- Avo — Data observability (https://www.avo.app/data-observability), 2025
- Avo — Actionable data ownership (https://www.avo.app/blog/introducing-actionable-data-ownership), 2025
- Amplitude — Best product analytics tools 2026 (https://amplitude.com/compare/best-product-analytics-tools), 2026
- Cotera — Product analytics platform comparison (https://cotera.co/articles/product-analytics-platform-comparison), 2026
- PostHog — Best Heap alternatives (https://posthog.com/blog/best-heap-alternatives), 2025
- PostHog — PostHog alternatives (https://posthog.com/blog/posthog-alternatives), 2025