Sign up
Sign up
Free performance evaluation
View more

How we grew a top 100 grossing game by 11% with special offer personalization

Tags:

Simulation game

When the performance of a game is flagging, we know there’s more to the story. Our performance evaluation on this top 100 game identified specific areas of IAP monetization that could unlock significant revenue.

What our performance evaluation identified:

  • Lower conversion in certain player segments, like veteran gamers focused on late-game currencies and casual players focused on fashion.
  • Some consumables, including late-game currencies, were hard to get without dedicating crazy amounts of game time, so a lot of players had stopped pursuing them.
  • Excessive offer discounts were cannibalizing monetization and driving currency hoarding.

We decided to run an in-game offer personalization strategy to connect with and convert players without the need for deep discounts.

In 3 months we aimed to:

  • Create new pricing and content modeling
  • Connect these to A/B test structure
  • Set up monitoring and reporting
  • Run the pilot A/B test

Our goal:

Revitalize year-on-year revenue for this leading mobile game.

Harnessing probabilistic and machine learning to generate over 50,000 unique daily offers

Our first task was to build a personalization engine that could produce more than 50,000 unique offers each day. The engine would learn players’ behavior and spending patterns and configure game assets into unique offers for every player.

Every day, each player would receive up to 5 unique offers of which 3 they would see right away, and 2 would be queued and waiting. Offers changed daily based on the player’s behavior in-game. Essentially, the idea was to create a more interesting and customized in-game shopping experience by changing the offering frequently, which the above KPIs show was the right move.

The last piece of the puzzle was the A/B testing which helps understand the impact of the improvements and test new hypotheses. We started by testing just 10% of the players. After seeing the initial results, we scaled the optimizations to a greater proportion of the playerbase – 10% to 30% and later, from 30% to 90%.

We developed the initial models (that had the necessary data pipeline and transformation) within a month and a half. The success of the first A/B test meant we could keep up the relative performance numbers while scaling absolute revenue significantly.

To date, our team still uses probabilistic and machine learning models (XGBoost, Neural Networks) to understand not only the price but also the content of the offers.

The results after in-game special offer personalization 

The new personalization boosted game revenue by 11%. The benefits extended to UA, unlocking an extra 23% in spend with no fall in retention.

Other results included: 

  • +7.25% daily paying users
  • +4% conversion

Podcast transcript

This podcast has no transcript yet.