The first deposit is not the final point.
FTD is convenient as an understandable funnel event, but it does not show further user value. Two sources can give the same number of FTDs at the same price, and after a few weeks give a completely different result.
Example of two sources
| Indicator | Source A | Source B |
|---|---|---|
| FTD | 100 | 100 |
| Cost per FTD | $70 | $70 |
| Re-active users | 18 | 42 |
| Conditional NGR for the period | $4 500 | $8 200 |
At the first deposit level, the sources are the same. On the long horizon, B creates a stronger base.
What is retention in the working sense
Retention answers the question of how much of the attracted audience remains active after a given period of time. The period is chosen for the task: D7, D30 or longer window. It is important to compare the same cohorts, rather than users drawn in different seasons and under different conditions.
Why this is especially important for RevShare
With RevShare, the quality of the audience directly affects the long tail. A source with an expensive FTD can be more profitable than a cheap one if users stay active longer. Therefore, cost per FTD optimization can lead to incorrect scaling.
How to build a simple cohort view
Group users by week of the first FTD and source. Then compare the results after the same number of days. This removes the skew when the old cohort has managed to accumulate the result, and the new one has not yet.
What questions to ask
- Which source gives a stronger D30 retention?
- Does the quality differ according to GEO?
- Whether retention worsens after scaling;
- What creativity leads not only to clicks, but also to a more sustainable audience?
Don’t turn LTV into a magic number.
LTV is a model, not a truth. If data is scarce, the long-range forecast can be very inaccurate. It is better to use conservative windows and update the estimate as the evidence accumulates.
Linking retention to the income model helps the material RevShare and NGR.
Do not mix the retention of different cohorts.
Users who arrive during a major promotion may behave differently than regular traffic. Therefore, it is useful to note not only the source and date, but also the significant context of the attraction. Otherwise, a weak or strong stock will be mistakenly attributed to the channel as a whole.
How to Use Retention Optimization
You don't have to wait six months. You can choose an early indicator that is historically associated with further value: for example, D7 or D30. First, check with the old data to see if this signal really explains something, and then use it as an auxiliary metric.
Cost of quality
Source B may have a cost per FTD 20% higher, but noticeably better retention. If you optimize only the first metric, the budget algorithm will continually supplant B, even though it creates more value over the long horizon.
When to Predict LTV Is Dangerous
New GEOs, new sources, and dramatic scaling are changing audience profiles. A model trained on old, quiet traffic can overstate expectations. In such cases, reduce the forecast horizon and increase the share of evidence in decision making.
Look not only at share, but also at value.
Two sources may have the same D30 retention, but different NGR per active user. Therefore, retention is useful to read along with the monetary indicator, and not as a standalone quality rating.
Scale changes retention
When a campaign moves from a narrow, strong audience to a wider audience, retention often becomes different. After each major targeting expansion, create a new cohort and don’t mix it with the original.
Simple quality map
Place sources on the plane: horizontally cost per FTD, vertically D30 value. The low-cost, high-value segments are obvious, and the disputed areas become visible: cheap but weak retention or expensive but very high-quality source.
