Bayesian MMM for Mobile Game User Acquisition
Supercell
The Challenge
Supercell wanted to know which of their marketing channels were actually working. When you're spending across dozens of channels to acquire players for mobile games, the fundamental question is deceptively simple: for every dollar spent here versus there, how many players do we actually get? Without a principled way to untangle overlapping campaign effects, budget decisions were guided more by intuition than evidence.
Our Approach
We built a marketing mix model that accounts for the natural hierarchy in Supercell's data — different markets behave differently, and a good model respects that structure rather than flattening it away. The model estimates the return on each channel while sharing statistical strength across markets, so even channels with limited spend history in a given region get reasonable estimates. The end result is a tool for asking "what if we moved budget from channel A to channel B?" and getting an honest answer — complete with uncertainty about how confident we should be in that answer.
Results
The model gave Supercell a clear, quantified view of channel-level returns across their markets.
Rather than relying on last-touch attribution or gut feel, their user acquisition team could now see which channels were genuinely earning their budget and which were riding on the coattails of organic growth.
PyMC Labs Team
- Christian Luhmann
- Niall Oulton
- Sef M
Let's Chat, We Respond Fast
Tell us about your problem. We typically respond within 24 hours.
Schedule a Consultation