SaaS Product Analytics — A Field Guide

Revenue and churn tell you what happened last month. Product analytics is supposed to tell you why — which onboarding step loses people, which feature correlates with retention, whether the thing you shipped last sprint actually changed behavior. Most teams install Amplitude or Mixpanel, fire a burst of events, and stop there, which produces a dashboard nobody opens rather than an instrumentation layer that actually informs product decisions.

Event tracking that survives contact with reality

The single biggest failure mode in product analytics is an event taxonomy that grows organically — every engineer names events however they like (button_click, Clicked_Signup, signup-cta-tapped) and six months later nobody can write a reliable query because the same user action is logged three different ways. A workable taxonomy fixes the grammar early: object_action in a consistent case (project_created, invite_sent, report_exported), a small fixed set of standard properties on every event (user_id, account_id, plan, timestamp, source), and a lightweight review step before a new event ships, similar to how you'd review a schema migration. Track actions that indicate value delivered, not just page views — a pricing page view means almost nothing, while "second project created" or "first integration connected" tells you the user is on a path to activation.

Activation funnels: finding where people actually drop

An activation funnel maps the sequence of steps between signup and the moment a user experiences the product's core value — signup, first login, key setup action, first meaningful use. The value of a funnel isn't the aggregate conversion rate, it's finding the single step with the steepest drop-off, because that's where a UX fix or a better default has the highest leverage. A funnel that goes 100% signed up, 82% completed onboarding, 71% invited a teammate, 24% created their first real project is telling you the gap between "invited a teammate" and "created a project" is where people are getting stuck or losing motivation, not the signup form.

SELECT
  DATE_TRUNC('week', u.signed_up_at) AS cohort_week,
  COUNT(DISTINCT u.user_id)                                             AS signed_up,
  COUNT(DISTINCT u.user_id) FILTER (WHERE e1.user_id IS NOT NULL)       AS onboarded,
  COUNT(DISTINCT u.user_id) FILTER (WHERE e2.user_id IS NOT NULL)       AS invited_teammate,
  COUNT(DISTINCT u.user_id) FILTER (WHERE e3.user_id IS NOT NULL)       AS created_first_project
FROM users u
LEFT JOIN events e1 ON e1.user_id = u.user_id AND e1.event_name = 'onboarding_completed'
LEFT JOIN events e2 ON e2.user_id = u.user_id AND e2.event_name = 'invite_sent'
LEFT JOIN events e3 ON e3.user_id = u.user_id AND e3.event_name = 'project_created'
GROUP BY 1
ORDER BY 1;

Retention curves, and why shape matters more than day-30

A retention curve plots the percentage of a signup cohort still active on each subsequent day or week. What matters is not any single point on the curve, it's whether the curve flattens (asymptotes above zero) or keeps declining toward zero. A curve that flattens at 35% by week 6 means you have a durable core of users who found lasting value — that's a product with genuine retention. A curve that keeps sliding toward zero, even slowly, means the product hasn't yet found the group of users for whom it's truly sticky, and growth is just refilling a leaky bucket. Segmenting the retention curve by acquisition channel or by whether a user hit an activation milestone in their first week is usually more revealing than the blended curve, because it often shows that users who complete a specific early action retain dramatically better — which tells you exactly what onboarding should be pushing people toward.

Tool choice: Amplitude, Mixpanel, and PostHog

The three dominant product analytics tools solve the same core problem with different tradeoffs. Amplitude has the deepest behavioral cohorting and predictive analytics (churn prediction, "users like this one") but gets expensive fast as event volume grows and has a steeper learning curve for non-technical team members. Mixpanel is generally more approachable for a smaller team and has strong funnel and retention reporting out of the box, though its data modeling is less flexible than Amplitude's at the high end. PostHog is open-source, self-hostable, and bundles session replay, feature flags, and A/B testing alongside analytics in one product — a strong default for a smaller SaaS team that wants to avoid paying for four separate tools, at the cost of a less mature analytics query experience than either Amplitude or Mixpanel for very large datasets. None of these tools fix a bad event taxonomy; picking the tool is a distant second-order decision after getting the instrumentation itself right.

Wrapping up

Product analytics only pays off when the event taxonomy is disciplined enough to query reliably, when funnels are used to find the single steepest drop-off rather than admired as a dashboard, and when retention curves are read for their shape and segment, not a single day-30 number. Get the instrumentation right first — the tool you layer on top (Amplitude, Mixpanel, or PostHog) matters far less than most teams assume.