Revenue Forecast
General methodology of the model
Machine Learning algorithm is trained to predict revenue retention (revenue decline rate) split by ['network','country','campaign','adgroup(adset)','placement'] based on cohort characteristics known within the first 10 days. The model is then used to predict the revenue decline rate for each cohort. The resulting data is then combined with the current revenue retention level for the cohort, which gives us the final forecast used to calculate the LTV.
How often do we update the forecast?
Revenue forecast is updated on a daily basis. This allows adjusting the forecast as the new data appears, making it more accurate.
What affects forecast accuracy? How can it be improved?
The more data is available about the cohort, the better:
-
A bigger cohort (number of installs) shows more stable retention and lower forecast dispersion.
-
A larger number of days after the install improves the accuracy of the revenue retention forecast.
For example, the forecasts after days 3 and 7 may differ dramatically, since the actual user retention may change.
When does forecast accuracy decrease?
When there are big changes in the app, such as:
-
New updates that increase or decrease user engagement and affect retention.
-
Changes in monetization strategy resulting in increase or decrease in app ARPU
-
Launching campaigns in new ad networks with a different amount of traffic
What happens next? Forecasting algorithm will use historical app data to retrain itself, taking into account recent app changes as well as new data. This may take from several days to several weeks.
How do we get a more specific forecast?
As an example, let's look at the sample where revenue's split by ['network_name','campaign_name', 'creative_name'].
In this case, the forecast cohort revenue is distributed proportionally to the actual revenue that has already been received.
Revenue for the cohort acquired through AAA campaign
| Campaign | Revenue To Date | Revenue Forecast D365 |
|---|---|---|
| AAA | 9 | 27 |
Actual and forecast revenue distribution within the cohort
| Campaign | Creative | Revenue To Date | Revenue Forecast D365 |
|---|---|---|---|
| AAA | BBB | 1 | 3 |
| AAA | CCC | 5 | 15 |
| AAA | DDD | 3 | 9 |