How to Use the Retention Report
This document explains how to set filters and read results in the 리텐션 리포트 in the left-hand menu. The key is distinguishing between the cohort criterion event (who to group) and the return event (what to measure as coming back).
1. Cohort Criterion Event vs. Return Event
Retention is measured as a combination of "who to group together, and what to count as coming back." The two events play different roles.
| Item | Role | Default |
|---|---|---|
코호트 기준 이벤트 | The starting event that groups users into a cohort. Users who performed this event become the subject of analysis. | install |
리턴 이벤트 | The activity event used to judge that a user "came back." | 모든 이벤트 (all in-app activity) |
For example, if you set the cohort criterion event to install and the return event to purchase, you measure "the proportion of users who installed and came back with a purchase on D7." If you set the return event to 모든 이벤트, you do not specify a particular behavior and instead use all in-app activity that occurred in the app as the criterion for coming back.
For both events, you can choose from the recommended (install, first_open, sign_up, login, purchase), standard, and custom event lists via the search input. Event names not in the list can also be typed directly into the search box and used as-is.
2. Selecting the Cohort Period
In 코호트 기간, you specify the reference period for the cohort being analyzed.
| Item | Description |
|---|---|
| Presets | You can quickly choose from 어제, 최근 7일, 최근 14일, 최근 28일, 최근 30일, 최근 90일. |
| Custom range | You select the start and end dates directly from the calendar. You can specify up to 366 days. |
The default is the 30-day span ending yesterday. To view long-term retention such as D30, you need data from at least 30 days after the cohort reference date, so if you only look at very recent periods, the later D-day values may appear empty.
3. Selecting Grouping Dimensions
In 그룹화, you choose the criteria by which to break down the results. You can select up to 6, and the defaults are 채널 and 캠페인.
| Dimension | Description |
|---|---|
| Cohort date | Groups cohorts by install date. |
| Channel | The channel name (e.g., tiktok.ads, meta.ads, google.ads). |
| Campaign | The campaign name. |
| Ad group | The ad group / placement name. |
| Creative | The ad creative name. |
| Platform | The device platform (iOS / Android / web). |
If you select multiple dimensions, rows are split by combination units such as channel × campaign × creative.
4. Selecting Retention Days
In 리텐션 일자, you pick the D-days to measure. The selectable values are D1, D3, D7, D14, D30, and all are selected by default. Only the selected days are shown as D-day columns in the result table.
5. Filters and Platform
| Item | Description |
|---|---|
필터 | Adds conditions that narrow the result scope by dimension value (e.g., viewing only a specific channel or campaign). You can add multiple conditions. |
플랫폼 | Choose from 앱 (iOS + Android), Web, or 전체. The default is App. |
6. Minimum Cohort Size
최소 코호트 사이즈 is the minimum number of unique users a cohort must have to be included in the results. The default is 30.
Cohorts with too small a population produce retention rates that are statistically meaningless. For example, a cohort with one user can only show a retention rate of 100% or 0%, causing distortion. Setting this value appropriately excludes such small cohorts from the results. If you see almost no result rows because there is little data, try lowering this value; if there is a lot of noise, raise it to compare only stable cohorts.
7. Running the Query and Reading Results
After setting the filters, click the 쿼리 실행 button at the top left to retrieve the results. The 초기화 button returns all filters to their default values.
7.1 Weighted Average Card
When results are available, a 전체 코호트 가중평균 리텐션 card appears at the top right. It sums all cohorts and shows the average retention rate per D-day, so you can grasp the overall retention trend at a glance rather than individual cohorts. It is a weighted average method in which cohorts with a larger cohort size are reflected more heavily in the average.
7.2 Result Table
Columns are arranged in the order of dimension columns → cohort (size) → D-day (user count # and rate).
| How to read | Description |
|---|---|
| Horizontal (row) | Look at how the retention rate of the same cohort declines as it moves from D1 → D30. Usually D1 is the highest and it gradually decreases thereafter. |
| Vertical (column) | At the same D-day, compare retention rates by dimension (channel, campaign, etc.) to identify which acquisition source stays longer. |
| Heatmap | Rate cells are shaded more darkly the higher the retention rate, so you can quickly find strong/weak segments just by scanning the table. |
| Sorting | Clicking a column header sorts by that column. Clicking again toggles between ascending and descending order. |
7.3 Cautions
- Do not look only at the retention rate (
D{n}); check it together with the adjacent코호트size and theD{n}#(the number of returned users). If the population is small, a high rate is hard to trust even if it looks high. - Cohorts filtered out by the minimum cohort size do not appear in the results, so if the results are empty, widen the period or lower the minimum cohort size and run again.
- When viewing recent periods, the later columns whose D-day has not yet arrived (e.g., D30) may have empty values.