Information

How Insight works

The core of Insight is a data sampling methodology like the one used in political polls around the world.

Distilled down to its basics: we count how many people in our sample have a game and then extrapolate that onto the world to estimate total players. The details, however, are much more complex.

PlayTracker maintains two samples – its cross-platform correlated database of users, and a rolling sample of anonymous public profiles. However, neither of these data samples are truly random because users must opt into either registering for PlayTracker or simply setting their profile to public, thus the samples are not statistically representative.

To deal with this, we use machine learning. Both samples are inputs into an algorithm that learns from publicly available data such as reviews, concurrent players, and most importantly confirmed player number data points like when a publisher brags about sales numbers.

By comparing tens of thousands of these reliable confirmed datapoints to the data in our samples, the algorithm develops its understanding of which users are overrepresented or underrepresented, and then adjusts our estimates accordingly.

On the other hand, many graphs use achievement data – like for example our “New users over time” graph. These are straight from the database and are not affected by machine learning, so keep that in mind when using their figures.

If you have any further questions about PlayTracker’s methodology, feel free to contact us.