Highlights

The Client

The project’s client is a developer and publisher of mobile games based in the United States. They create family-friendly games that cater to a global audience, boasting millions of active users. 

The client came to XIM for assistance in creating a tool that would forecast estimates for LTV (lifetime value, or the total income for the total time spent by a user in a game), income from advertising campaigns, and income from in-game purchases. This would help the user acquisition department make more informed decisions when designing and promoting games or purchasing traffic.

The Project

The forecasting tool XIM is developing works by gathering historical data on various cohorts of real users (platform data, user location, income from advertising, and other data sources from both individual users and cohorts). This data is then aggregated and analyzed using AI to generate the forecasts. 

This tool empowers the user acquisition department to optimize their spending on advertising campaigns by reducing the budget for locations or channels that are not expected to perform well and reallocating it to more promising areas. The AI-generated forecasts can also help optimize game design, ensuring higher income from in-game sales. 

The Solution

Forming the Team

XIM decided that, for this task, it would be most suitable to form a very small team of experts in order to reduce coordination-related costs and efforts, and the client agreed. A careful selection process ensured that the chosen specialists had the necessary expertise to successfully carry out the various complex tasks of the project.

Challenges 

Building such a sophisticated tool is a daunting task involving various challenges from the very beginning: 

1. The forecasting system must be able to make estimates for periods of up to a year. Such long-term forecasts must take into account a huge number of variables using multiple models — a complex task requiring advanced application engineering skills. 
2. In addition, the models and libraries used to generate the forecasts must be constantly updated; otherwise their accuracy decreases and the margin of error in the forecasts can become too large for them to be useful.
3. Finally, real-world historical data on the company’s income is used by the tool; such data is sensitive, so its security must be ensured.

The Outcome

This project is still in development and has not been fully implemented yet, but the client is satisfied with the progress. More functionality is planned, including the addition of extra forecasting models to the system and continuous increases in forecasting accuracy. The product will be used by multiple company departments, including user acquisition and marketing.