Agent Skills: Product Analytics

Product Analytics

A product analytics skill focused on decisions from data, not dashboards. Covers the metric tree, instrumentation patterns, funnel + retention + cohort analysis, and the operational rituals that turn measurement into product changes.

When to use this skill

• Designing the North Star metric and its tree of input metrics
• Auditing product instrumentation (events, properties, gaps)
• Building or refreshing an activation funnel for a new product or feature
• Designing or analyzing retention cohorts (D1/D7/D30/W1/W4/M1/M3)
• Building or refining the PM analytics dashboard
• Translating product data into decisions and roadmap inputs
• Auditing dashboards for actionability (kill the vanity)

Inputs the advisor expects

• Product type (B2B SaaS, consumer, marketplace, etc.)
• Current analytics stack (Amplitude / Mixpanel / GA4 / Segment / Snowflake + dbt + Looker)
• Existing North Star + input metrics
• Current event taxonomy + instrumentation gaps
• Top product questions you can't answer today
• Org expectations: who consumes analytics, at what cadence

Clarify First

Before designing the metric tree or audit, confirm these inputs. If any is unknown or vague, ASK — do not assume:

• [ ] Product type — B2B SaaS, consumer, marketplace, etc. (drives the North Star pattern and input metrics)
• [ ] The value moment — what "delivered value" looks like for a user (defines the North Star and activation event)
• [ ] Current analytics stack and event taxonomy — Amplitude/Mixpanel/GA4/Segment plus existing events (drives the instrumentation audit and gap list)

Stop rule: ask only the 2-3 that most change the output. If the user says "just draft it," proceed and list your assumptions at the top of the artifact.

Workflows

Workflow 1 — Design the metric tree

1. Define the North Star (one number that summarizes value delivered).
2. Decompose into inputs (drivers of the NS).
3. Add guardrails / counter-metrics that catch unintended consequences.
4. Run metric_tree_designer.py against your candidate tree to surface imbalance, missing layers, anti-patterns.

python3 product-analytics/scripts/metric_tree_designer.py \
  --input metric_tree.json --format markdown

Workflow 2 — Audit instrumentation

1. Pull the current event taxonomy + properties.
2. Run event_taxonomy_auditor.py to flag PII risk, schema drift, naming inconsistency, duplication, undocumented events, and gaps.
3. Generate the remediation backlog and assign owners.

python3 product-analytics/scripts/event_taxonomy_auditor.py \
  --input event_inventory.json --format markdown

Workflow 3 — Analyze retention cohorts

1. Pull cohort retention data (raw counts by cohort week and offset).
2. Run retention_cohort_analyzer.py to compute retention rates, identify patterns (smile curve, leaky bucket), and surface cohort-level alerts.

python3 product-analytics/scripts/retention_cohort_analyzer.py \
  --input retention.json --format markdown

Decision frameworks

North Star metric — what makes one good

A good North Star metric:

• Measures value delivered to the user (not just usage)
• Aligns to business outcome indirectly via clear chain
• Is a leading indicator of long-term success
• Can move week-over-week (so it can be acted on)
• Is hard to game without delivering real value

Common patterns by product type:

| Product type | Common North Star | |--------------|-------------------| | Communication / messaging | Messages sent per WAU | | Marketplace | Successful transactions per MAU | | Content | Hours of meaningful content consumed | | Productivity SaaS | Activated workspaces × engagement depth | | Consumer payments | Active payment senders per week | | Developer tool | Weekly active developers performing core action |

Don't pick "DAU" or "Revenue" as North Star — they're outputs, not value drivers.

Metric tree structure

A clean metric tree has three layers:

1. North Star (1 metric)
2. Input metrics (3–5 that combine to produce the NS)
3. Driver metrics (per input, 3–5 that move the input)

Plus a guardrails / counter-metrics sidebar (3–5 that catch unintended consequences).

If you have 30 KPIs at the top level, you have no top level.

The activation question

For any new product or feature, ask: "What does it look like when a user realizes value from this?"

That's the activation event. A clear definition makes:

• Onboarding design — clearer
• Funnel analysis — possible
• Eval of marketing channels — sharper
• Customer success interventions — better-timed

Common mistake: defining activation as "completed signup." Signup is table stakes; activation is the moment of value.

Retention curve shapes

| Shape | Diagnosis | Action | |-------|-----------|--------| | Power-law smile | Healthy product-market fit | Invest in scale | | Slow decay then flat | Product-market fit | Investigate the flatline cohort segment | | Steep then zero | Novelty product | Re-evaluate the value proposition | | Linear decline | Leaky bucket | Improve retention features | | Inverted (rising) | Network effects kicking in | Acquire harder |

Read shape before reading numbers.

Vanity vs actionable metrics

| Metric | Vanity if | Actionable if | |--------|-----------|---------------| | DAU / MAU | Tracked alone | Decomposed by segment, action | | Pageviews | Tracked alone | Tied to conversion funnel | | Total revenue | Tracked alone | Decomposed by cohort, channel, segment | | App downloads | Tracked alone | Paired with activation rate | | Total accounts | Tracked alone | Paired with active accounts |

The test: "If this metric goes up 10% next week, what do we change?" If you don't have an answer, it's vanity.

Common engagements

"Help me design our analytics for the launch"

1. Define activation event and 3–5 input metrics.
2. Spec event taxonomy (event names, properties, user/account context).
3. Pilot dashboards (one for the team, one for execs).
4. Set the review cadence; don't let dashboards rot.

"Our funnel rate is dropping. What's wrong?"

1. Decompose: which step's conversion dropped?
2. Segment: which user segment is driving it?
3. Cross-check: is the dropping segment newly acquired?
4. Test hypotheses against the data; don't guess.

"Help me audit our instrumentation"

1. Pull the event inventory (last 30 days, all events fired ≥10x).
2. Tag PII risk, naming inconsistency, gaps.
3. Identify the events that should be fired but aren't.
4. Build the remediation backlog with owners.

Anti-patterns to avoid

• More dashboards = more insight. Usually inverse. Cull aggressively.
• Confusing event volume for insight. Tracking everything badly is worse than tracking a few things well.
• PII in event properties. Privacy + compliance nightmare.
• Custom event names per developer. Naming convention or chaos.
• No event documentation. Future you and the next analyst will hate present you.
• One metric for the whole product. Different surfaces need different metrics.
• Vanity North Star. "Total signups" tells you nothing about value.

References

• references/metric-tree-and-north-star.md — patterns by product type, tree structure, anti-patterns
• references/instrumentation-and-event-design.md — event taxonomy, naming, PII, schema discipline
• references/cohort-retention-and-funnel-analysis.md — analysis techniques, segmentation, anti-patterns

Related skills

• product-team/ab-test-setup — experimentation (paired with metrics)
• product-team/product-strategist — strategy upstream of metrics
• data-analytics/ skills — for the data engineering side
• engineering/data-quality-auditor — for instrumentation data quality
• c-level-advisor/chief-data-officer-advisor — for platform decisions