Chapter 5 / 9
Analysis logic: which chart for which question?
In Amplitude you pick the question first and the chart type second. Each chart counts the same event rows with a different logic. The examples in this chapter are interactive: change the settings and watch the same data give a different answer.
- You will know which question each of the seven core chart types answers.
- You will see how the counting unit, segment and time window change the result.
- You can turn a client question into the right chart settings.
Summary table
| Chart | Question it answers | How it counts | Closest in Adjust |
|---|---|---|---|
| Segmentation | How many people did X, and how did that change? | Event count or unique users, broken down over time | Event reports |
| Funnel | Where do they drop off between steps? | Users who complete ordered steps within a time limit | None |
| Retention | Do they come back? | Share of users who do a return event N days after a start event | Cohort reports |
| Journeys / Pathfinder | What do they do before and after X? | Shows event sequences as paths | None |
| Cohorts | Who is the audience showing a behaviour? | A list of users matching a condition | Audiences (partly) |
| Stickiness | How often do they come? | Number of active days in a week | None |
| Lifecycle | What is the mix of the active audience? | New, current, resurrected, dormant | None |
Chart names and their place in the menu can change with Amplitude UI updates (especially Journeys and Pathfinder). The concepts stay the same; check the current names in the UI.
Three settings that change every chart
- Counting unit: total events, unique users or average per user? On the same chart this choice can change the number several times over.
- Segment and filter: the "for whom" question. User properties, event properties, cohorts and attribution fields from Adjust come in here.
- Time window: time between steps in a funnel, the day definition in retention, and the project timezone in every chart.
Segmentation: how many, how much?
The most basic chart. You pick one or more events, count them over time and optionally split by a property. Below, the same Product Viewed event is shown with three different counting units.
Product Viewed by platform, 14 days. sample data Unique users answers "how many people looked", total events "how many times was something viewed", and average "how many products did each viewer see".When a client says "daily product views", ask which one they mean. If one dashboard uses uniques and another uses totals, two teams end up discussing numbers that differ several times over for the same metric.
Funnel: where do they drop off?
A funnel shows the share of users who complete ordered steps within a given time. The critical setting is the conversion window: how quickly must a user get from the first step to the last to count as "converted"?
Other funnel settings
| Setting | Options | When it matters |
|---|---|---|
| Step order | This order; any order; exactly this order with no other events in between | Flows where users go back and browse between steps |
| Hold property constant | For example, product_id is the same across all steps | "Did they add the product they viewed?" |
| Counting | Unique users or totals | Repeat purchases |
| Breakdown | Platform, country, acquisition channel | "Which channel's users convert better?" |
In Adjust, a "Checkout Completed / Install" ratio looks like a funnel, but there is no concept of order or window: it is the ratio of two counts in the same period. An Amplitude funnel checks that the same person did A first and then B.
Retention: do they come back?
Retention is the share of users who did a start event and then do a return event on later days. You choose both events: "did people who signed up do anything on day 7?" and "did first-time buyers buy again within 30 days?" are very different questions.
Amplitude retention also offers a definition based on day ranges (bracket). Check the current names of these options in the UI, and always state which definition you used when giving a client a retention number.
Cohort table view
The table view of retention shows a start week in each row and the return rate in the following weeks in the columns. It is the view closest to Adjust cohort reports.
Signup Completed. Return: any active event. sample data Empty cells are weeks that are not complete yet; for newer cohorts only the first weeks are visible. Reading the Week 1 column top to bottom shows how the quality of new users changes over time.Journeys / Pathfinder: what happens before and after?
Path analysis shows the most frequent event sequences leading to or from a given event. A funnel measures "the path you planned"; path analysis shows "the path users actually take".
Cohorts: audiences defined by behaviour
A cohort is a list of users who match a condition. It is not a chart on its own; it is used as a filter or breakdown in other charts and can be synced to other tools. Its strength is that conditions can be based on behaviour.
| Cohort | Definition | Use |
|---|---|---|
| Cart abandoners | Did Checkout Started in the last 7 days and did not do Checkout Completed | Push campaign, retargeting audience |
| Power users | Active on at least 10 days in the last 30 | Compare their behaviour with new users |
| Dormant premium | plan = premium and not active in the last 14 days | Churn prevention |
| Campaign buyers | Acquisition channel Meta and at least one Checkout Completed | Channel quality analysis |
Adjust Audiences build device lists from campaign and event conditions. Amplitude cohorts are more flexible with ordered and frequency-based conditions ("at least 3 times", "did A then didn't do B"). Both produce an audience list.
Stickiness: how often do they come?
Stickiness shows, for users active in a week, on how many different days they were active that week. A product meant for daily use and one used once a week have very different healthy distributions.
Lifecycle: the mix of the active audience
Lifecycle splits each period's active users into four groups: new (first seen), current (also active in the previous period), resurrected (not active in the previous period but seen before) and dormant (active in the previous period, not in this one).
Breaking the "new" group in lifecycle down by campaign shows how much UA spend contributes to the active audience. This requires the acquisition field from Adjust to exist as a user property in Amplitude (Chapter 7).
Turning a question into chart settings
When you turn a client question into a chart, write down five things explicitly. If any one is ambiguous, the result is open to debate.
| Question: "Are users from the Ramadan campaign more loyal?" | Decision |
|---|---|
| Chart type | Retention |
| Start event | First open (or Signup Completed) |
| Return event | Any active event (or stricter: Checkout Completed) |
| Segment / breakdown | Acquisition campaign = ramadan_sa vs all other users |
| Time and definition | Users acquired during the campaign, D1/D7/D30, "on day N" definition, project timezone |
Write a question and guess the answer. Build the chart. If the result differs from your guess, find which of counting unit, segment and time window caused the difference. Repeat this loop once for every chart type.