How to Build an Amplitude Retention Chart You Can Trust

October 31, 2026

An Amplitude retention chart measures how many users who did event A came back and did event B in a later interval. Most bad retention charts are built from a mislabelled return event, not from a bug in the tool.

What are the two events, and why does the choice matter?

Start event defines the cohort — usually signup or first core action. Return event defines what counts as retained.

Defaults matter here. If the return event is “any event”, you measure session habit. If it is the core value action — project created, invoice sent, report exported — you measure whether the product did its job. The second is what you want and almost nobody sets it.

A rule that works: the return event should be the thing a user has to get value out of the product. Not the page they land on.

How do you configure the chart?

  1. Open Charts → Retention in Amplitude.
  2. Set start event to Signed Up and return event to your core action.
  3. Choose the interval. Weekly for B2B, daily for consumer.
  4. Set the window — 8 to 12 intervals is readable; 30 is a wall of noise.
  5. Segment by acquisition channel or plan.
  6. Save the chart before you share it, so people compare like with like.

Step 5 is what turns the chart into a decision. An overall retention curve of 30% is unremarkable; a curve where one channel retains at 55% and another at 9% is a spending decision.

N-day, unbounded, or bracket — which one?

Type Counts a user at interval N if… Use for
N-day They were active exactly on day N Short-cycle consumer products
Unbounded They were active on day N or later Habitual, low-frequency products
Bracket They were active within days N to M Weekly-usage B2B tools

For most B2B SaaS, bracket retention at weekly intervals is the honest view. N-day retention on a tool people use twice a week will produce a falsely brutal curve, and someone will panic and fund a redesign nobody needs.

Pick one type and never mix. Two charts of different types compared in a slide is the most common way retention gets misread in a board deck.

What do the first four data points mean?

Day 1 shows whether the product delivered value immediately. A steep drop is an onboarding problem, not a retention problem.

Day 7 is habit formation for most tools. If the curve flattens here, the product has a real user base.

Day 30 tells you about the paid tier. This is where self-serve tools separate from tools people bought and abandoned.

Day 90 is about switching cost. A flat tail past day 90 is the strongest signal you have.

Flattening matters more than height. A curve that starts at 40% and settles at 20% is healthier than one that starts at 70% and falls to near zero.

What are the mistakes that make retention unreadable?

Three. Counting page views as the return event. Changing the start event mid-quarter and comparing to old charts. And segmenting by plan when the plan is assigned after onboarding, which introduces the outcome you are trying to measure into the cohort definition.

That third one is subtle and common. If you only upgrade after success, plan segments will always show paid users retaining better — it is circular.

Compare definitions against PostHog or Pendo if you want a second reading; if the two tools disagree, the definition is almost always the reason.

When is retention the wrong chart?

When your product is used quarterly (tax, annual compliance): cohort windows get too wide to compare. Use account-level engagement instead.

When you have fewer than a few hundred users per cohort. The noise will exceed any real difference, and segmenting makes it worse.

Key Takeaways

  • Set the return event to the core value action, not “any event” or a page view.
  • Weekly + bracket retention is the honest configuration for most B2B tools.
  • Flattening beats height: a 20% floor is better than a 70% start that decays to zero.
  • Never compare charts of different retention types in one slide.
  • Segmenting by plan after onboarding is circular and inflates paid retention.