3. Forecast Results
The 3rd step of Audience Reporting is applying projections onto your user performance to anticipate future value.
AdLibertas is excited to add Forecasting capabilities to Audience Reporting.
What are the Forecasting tools?
The LTV forecasting tools provide a visualization of a curve of best fit over your audience’s LTV data. It provides a quick and easy way of seeing possible future values of the LTV metric, which may be useful in understanding future values of the audience LTV.
We apply several mathematical models to help you choose the curve projection that best represents your data and users. The model is calculated using your chosen LTV data and allows you to extend these models to predict future values. We currently offer 3 curves to choose from:
- Linear: Also called line-of-best fit, this is the most aggressive growth model and shows a straight-line growth rate. This most often occurs with a relatively flat retention curve and users are predictably returning to the app. The equation takes the form Y=aX + b.
- Power: A non-linear power regression calculation. Useful when there is a skew in early user behavior (e.g. a user drop-off) or the lifetime changes in a non-linear rate. The equation takes the form Y=aX^b.
- Logarithmic: Also called exponential; this is the “most conservative” function and takes the form via a non-linear regression with logarithmic transform. This is most useful when there is a large drop-off in retention (or profitability) in early user behavior. Takes the form Y=a*ln(x) + b.
Depending on the nature of the data, one model may be a better fit for the data than the others. We encourage you to explore and select the one that matches your needs.
Exclude: Data Filtering
When modeling LTV data, it is not uncommon to see large fluctuations in the retention rates of users over the first several days to several weeks, that then settles down and tapers off into a long tail of retained users. Since projections over longer time periods rely on the long-tail (retained) of users more than the initial fluctuations, excluding noisy days upfront may lead to more reliable forecasts.
In order to help include only the most relevant data points, the UI offers 3 different data filters:
- Include all data points: Will include all data points starting at Day 0.
- Exclude first 7 days: Excludes Day 0 through Day 6 from the forecast curve of best fit. Will only include all data points from Day 7 onward in calculations.
- Exclude first 14 days: Excludes Day 0 through Day 13 from the forecast curve of best fit. Will only include all data points from Day 14 onward in calculations.
The UI can also visualize the range of individual values spanning 2 standard deviations (containing 95% of all individual values) above and below the forecast line. The values are represented as a shaded area on the graph, with a lower limit of 0, measuring the 2 standard deviation across the actual data, extending to the upper limit using projections for the forecast. The purpose of the standard deviation lines is to show the 95% confidence interval of any individual forecast value, based on the underlying data points.
We do not provide options to show a +/- 1 standard deviation range, as it is not sufficiently rigorous in showing a range of expected values (68%). On the other hand, 3 standard deviations is overly rigorous in most cases and rarely used for this type of A/B testing.
The standard deviation view helps visualize a range of values that may be observed in the LTV forecast. For some audiences, especially those with fewer data points, there may be much more variance, or a wider range in the data, and the +/- 2 standard deviation will visually reflect that. Seeing a wide range of possible values can also be helpful when comparing LTV against cost data and making ROI calculations.
To see a quick video example watch the video below: