What if you could predict your app’s expected revenue a few weeks or months in advance? Or know which ads will pay off rather than falling short of the target, and whether a customer is worth the acquisition costs?
To help you figure all this out, myTracker introduces a new, unique feature — LTV predictions for in-app purchases coming first as part of a broader set of predictive analytics capabilities. Leveraging data on billions of past events and their patterns, and based on your app’s individuality, predictive algorithms make it possible to predict the value of the users you acquire. The future is already here, all you need is to run myTracker and look at what is has to offer!
How it adds value
Use our machine learning algorithms to predict customer behaviour.
● Big Data from experts
myTracker analyses thousands of actions of each user, adding up to billions of tracked events overall, and automatically generates audience value predictions together with user acquisition and product analytics data.
● Marketing budget savings
myTracker leverages the patterns identified in historical and current data sets and provides predictions the very next day after you launch your ad campaign. This enables you to measure the potential value of each campaign or customer cohort, and swiftly reallocate budgets from ads with disappointing LTVs to stronger performing ads.
● Identifying the best promotion strategy
myTracker provides LTV predictions split by country, traffic type and partner, so you can identify the most effective tools to promote your app and allocate your budgets wisely based on the potential effectiveness of advertising channels.
● Improving the quality of the audience
Using predictive LTV metrics, you can quickly identify your core audience, minimise costs and acquire more valuable customers. You can also see how much revenue each user cohort will bring you, not to waste time on those customers who are not worth the cost of their acquisition.
Any app promotion gets much easier if you can predict returns in advance. LTV predictions enable high accuracy of ad campaign budgeting making sure you recoup the costs. Handling this valuable feature the right way is the best way to always make a profit. myTracker is a mobile app tracker that provides accurate user-generated revenue LTV predictions.
How it works in myTracker
LTV as a measure of post-install app revenue per user is one of the top metrics for your app. Our LTV predictions are based on in-app payments and require the payment verification to be activated in the app. Knowing the expected LTV in advance would be crucial for understanding how much money is worth investing in app development and customer acquisition.
Now, as myTracker added a new feature, you can make LTV predictions and forecast your ROI and profit. You can measure the expected revenue, the potential quality of traffic sources, and the effectiveness of a future ad campaign, skipping months of collecting statistics and accumulating actual data. The predictions are available starting the very next day after the app is installed by users representing the relevant cohort.
Red line: LTV predicted by myTracker. Yellow line: actual LTV.
LTV predictions are made using machine learning models based on aggregated data. myTracker and the Mail.ru Group team have accumulated about the performance of thousands of applications that were used to “train” predictive algorithms. The up-to-date massive array of data coupled with the complementary use of several predictive models and continuous “training” of algorithms — all contribute to high accuracy of predictions. As new data arrives and actual app performance helps to calibrate the algorithms, the predictions get even more accurate.
myTracker-generated LTV predictions are built in four stages:
● Data collection.A built-in SDK collects your app’s statistics and compares it with the accumulated data.
● Predictive model training. The users of your app are divided into groups; the model analyses historical data and calculates coefficients.
● Predictive model application. When the model learns to factor in your app’s individuality, it begins to break the input data down to the smallest user cohorts (with the granularity going to the level of individual devices), making individual predictions by each of them. In this fashion, LTV predictions get more and more accurate with each day.
● Prediction evaluation. The resulting predictions are compared with the app’s actual performance. Coefficients are constantly calibrated getting tailored specifically to your app. By storing predictions for various time intervals you can later compare them with actual data sets (e.g. predicted 180-day LTVs vs. actual 180-day LTVs).
All this means that myTracker helps you make the best decisions by leveraging both current and forward-looking data on your app’s financial performance over a horizon of several weeks or months.
Approximation and linear models
Our LTV predictive mechanism adjusts to the current stage of your app’s life cycle, the amount of data collected, and the number of new installations per day. LTV predictions are available for different time intervals, including 30, 60, 90 and 180 days after installation. Depending on the length of the data collection period, myTracker will use either an approximation (less accurate) or linear (more accurate) model. Put simply, the logic behind the prediction is as follows:
1. If there is little actual data generated by the app and the payment history spans no more than 1 to 30 days, the approximation model is used. It can make predictions as early as the very next day after the first payment, though its margin of error may be above average.
2. For a payment history of more than 30 days (but less than N+30 days, where N is the day for which you want to make a prediction), the system also uses the approximation model. In this case, the available data is still insufficient for a full-fledged deployment of the linear model, but the approximation model shows a much better accuracy than for a span of 1 to 30 days.
3. If your app's payment history saved in myTracker is more than N+30 days, and a certain number of payments have already been made (e.g. 100+ payments over a 30-day period), the prediction is based on a linear model. It will have a higher level of accuracy, leveraging various statistical indicators, such as average LTVs by country of users’ residence, rather than payment data alone.
If you want to build a prediction for a 30-day horizon, it is desirable that the system has at least a 60-day history of your data, which enables it to use the linear model. For a 90-day horizon, 120-day data history is needed and so on. So, it is advisable to start using myTracker in advance, even if you are not going to use predictive analytics or other tracker capabilities right away. The data accumulated by the system can be later used for detailed analysis and accurate predictions.
Both approximation and linear models are perfected automatically for continuous improvement of their predictive accuracy. If you feel that one of the two models fits your app better, you can choose to have it used by default for any subsequent predictions.
If necessary, you can also use myTracker to generate tubular and graphical reports to compare the predicted and actual LTVs.
How to generate your LTV predictions
To generate LTV predictions for your app:
You can also combine predictions for different user cohorts in a single report if you need to compare the effectiveness of ad campaigns and to access traffic sources or data for specific weeks/months in which the app was installed.
More information about myTracker’s predictive analytics, including the calculation methods, margins of error, confidence intervals, model specifics and much more, can be found in our documentation.