From Jump Odyssey to XP Lab
Not long ago, while working on balancing game parameters for our mobile project Jump Odyssey, we hit a wall. Manual testing was taking too much time, and we were never 100% certain if our settings were truly optimal.
Instead of guessing, we built a system that analyzes player behavior and recommends parameters in real-time. This experiment was so successful that it became the foundation for XP Lab.
The parameters we optimized
For the initial test, we focused on five key configuration variables:
- ad_hp_restore — How many hearts a player gets after watching a rewarded ad.
- inter_to_revive — Whether to show an interstitial ad to revive a player.
- tutorial_jumps — The number of jumps defined as the tutorial phase.
- initial_hp — The starting number of hearts for each run.
- checkpoint_hp_restore — How much health is restored at checkpoints.
The results
The test has been running for over two weeks, and the results have been eye-opening. We managed to increase our Ad Revenue per User from $0.015 to $0.035.
While there is still work to be done, doubling the revenue per user purely through parameter optimization is a major milestone.
[!TIP] This approach allows you to stop guessing about IAP package sizes or ad frequency and let data-driven logic find the winner.
What XP Lab optimizes now
Our system now handles recommendations for:
- IAP Prices & Bundles: Find the perfect combination of items and price points.
- Ad Placement & Frequency: Maximize revenue without hurting player retention.
- Offer Timing: Deliver the right offer at the magical moment to boost conversion.
- Player Progression: Balance difficulty, rewards, and enemy stats automatically.
- Onboarding: Identify the most effective tutorial flow for different player segments.
Join the beta
XP Lab is currently in beta, and we are looking for a few more teams to join the testing phase. If you’re interested in applying deep learning to your game’s economy and progression, we’d love to have you.
Conclusion
Optimization shouldn’t be a manual grind. By moving from static balancing to dynamic “Multi-Armed Bandit” models, we’re helping developers find their game’s true potential.