How to Predict a Mobile App's LTV on Your Own

When placing ads to attract traffic, you need tools to estimate ad performance so you can reallocate your budget or cancel an ad.

These can be models to forecast a user’s lifetime value (LTV) over a given horizon (usually 30, 90 or 180 days or 1 or 2years).

What is this guide about?

App revenue forecasting is a tough task involving complicated data pre-processing and analysis and choosing the proper models and metrics.

In this guide, we will discuss and suggest a number of ways to predict revenue, looking at models for several use cases:

Our approach

A forecast by device. Using pretty similar techniques, you can make LTV predictions by device or user. This guide covers the former.

A 180-day prediction horizon. This time frame is most commonly used to assess ROI in this industry. However, the logic would be the same for any other period.

An 8-day data accumulation period. In the mobile app industry, data is normally accumulated for 3–15 days. For this guide, we have chosen the golden mean of eight.

Payments and traffic type as inputs. Payments are critically important for LTV prediction, while the traffic type is an additional parameter we have selected as an example.

What you will learn

With the help of this guide, you will learn to use raw data, pre-process it and prepare it for analysis, as well as choose models and properly assess them using your metrics.

For each model, we'll walk you through the following steps:

  1. Viewing the data
  2. Transforming the data and gearing up for the training
  3. Training the model
  4. Evaluating the model

Read more in our comprehensive guide to making data-based device revenue predictions

    Tags: LTV predictive analytics