Track Product Events and Analyze User Retention Funnels

This skill covers product analytics work: PostHog and Mixpanel event tracking, conversion funnels, cohort retention, DAU/MAU, feature flags, A/B testing, North Star metrics, OKRs, and product dashboards. It opens with a line attributed to W. Edwards Deming: "In God we trust. All others must bring data." Event naming follows an object-past_tense_verb convention - user_signed_up, conversation_started, and upgrade_completed are correct; generic names like signup, click, or conversion are not.

Use it to define an activation event, investigate a funnel drop-off, or calculate retention with an explicit denominator and time window. Before instrumenting anything, it has you record the product decision behind the metric, the data source, applicable consent, the timezone, and the unit of analysis. It explicitly does not apply when the task is unrelated to product analytics, when a simpler and more specific tool already handles the request, or when the user just needs general-purpose assistance without domain expertise.

The skill works through a set of illustrative examples rather than a single fixed pipeline. An example event taxonomy (AURI_EVENTS) groups events by lifecycle stage - acquisition (user_signed_up, onboarding_started, onboarding_completed), activation (first_conversation, aha_moment_reached, feature_discovered), retention (conversation_started, conversation_completed, session_started), and revenue (upgrade_viewed, upgrade_started, upgrade_completed, subscription_canceled, payment_failed) - each with named properties. A hypothetical activation funnel (landing page visit down to Pro conversion) pairs each step with a target percentage, and a six-step optimization loop (identify the exact drop-off point, understand why via session recordings or surveys, form a hypothesis, run a statistically significant A/B test, measure against a pre-set window and guardrails without stopping early on a favorable p-value, then learn regardless of outcome) drives improvement. A cohort-retention function computes a weekly retention matrix from a user_id/event_date/event_name dataframe, and illustrative retention bands are explicitly flagged as unsourced format examples to be replaced with real observed baselines, not market benchmarks. A North Star framework works from "what creates real user value" to a named, measurable metric - the worked example defines "Weekly Active Conversationalists" (users with 3+ conversations of at least 2 minutes in a week) with year-one and year-two numeric targets, plus a dashboard function that counts qualifying users in a single windowed, timezone-consistent aggregate rather than one row per user. An A/B significance calculator uses scipy.stats.chi2_contingency, validates inputs, and returns an insufficient-variation or sparse-counts status rather than a false read when the data can't support a verdict, always recommending review of pre-specified effect, uncertainty, and guardrails over automatic action. Seven suggested slash-command prompts (/event-taxonomy, /funnel-analysis, /cohort-retention, /north-star, /ab-test, /dashboard-setup, /okr-template) organize these workflows, though the skill notes these do not install client commands.

from posthog import Posthog
import os

posthog = Posthog(
    project_api_key=os.environ["POSTHOG_API_KEY"],
    host=os.environ.get("POSTHOG_HOST", "https://app.posthog.com")
)

def track(user_id: str, event: str, properties: dict = None):
    posthog.capture(
        distinct_id=user_id,
        event=event,
        properties=properties or {}
    )

def identify(user_id: str, traits: dict):
    posthog.identify(
        distinct_id=user_id,
        properties=traits
    )

PostHog is the primary integration, for both event capture/identify calls and feature-flag evaluation - the skill notes to match the API to the installed SDK version (current SDKs expose evaluate_flags; older ones ordered feature_enabled as (feature, user_id)), to fail safe on error or missing values, and to never send events or identify calls before the project's consent and authorization rules clear. It links to PostHog's official Python documentation for the current API. Mixpanel is named as an alternative product-analytics tool.

Product and growth teams who need to define activation events, diagnose funnel drop-off, size retention by real cohorts, pick and track a North Star metric, evaluate feature flags safely, or judge A/B test significance without over-trusting a single p-value.