Using Data Science to grow games / Robert Magyar (SuperScale)
0 likes169 views
Superscale is a profit scaling partner for mobile games, leveraging data science to optimize player monetization and growth across various genres. The document outlines their approach to understanding player behavior, improving monetization through personalized offers, and the journey of game scaling from pre to post-launch. Key challenges and learnings emphasize the importance of data-driven strategies and community adaptation for maximizing revenue and player retention.
Using Data Science to grow games / Robert Magyar (SuperScale)
- 1. SUPERSCALE Award-winning Profit Scaling Partner for Top Grossing Mobile Games for
- 2. Who am I? I am passionate about helping game studios all around the world to grow their games. Across different genres (match3, RPG games, racing games, simulations, etc) we have: ● Analyzed data of 500M+ players. ● Optimized millions of dollars in media spend. ● Brought millions of dollars in revenue uplift. Robert Magyar Head of Data Science
- 3. Who is SuperScale? We are game growth specialists on a mission to profitably scale games to their maximum potential.
- 4. Using Data Science to GROW games
- 5. Table of Content 1. Unlocking growth through data science ○ Potential and realization of growth in practice 2. Building impactful data science solution ○ Acquiring the right data ○ Optimizing players’ monetization experience ○ Putting it all together 3. Challenges & Learnings
- 6. CONFIDENTIAL Unlocking growth through data science
- 7. CONFIDENTIAL Lifecycle of the Game Pre-scaling Game Scaling is not yet possible Growing game Game is currently scaling Established games Game already peaked Legacy titles No development or growth support
- 8. CONFIDENTIAL Lifecycle of the Game and Data Science Pre-scaling Game Scaling is not yet possible DS can identify churn points (in this phase - low amount of data) Growing game Game is currently scaling DS can start leveraging data Established games Game already peaked DS can decrease churn and improve monetization Legacy titles No development or growth support DS can decrease churn and improve monetization Data science has more relevancy as product matures
- 9. CONFIDENTIAL Growth Through Player Lifecycle Optimization ACQUISITION ACTIVATION RETENTION MONETIZATION CHURN STORE CONVERSION User Acquisition (Paid) User Acquisition (Organic/Viral) ASO FTUE/Onboarding Design Core Loop Design Meta-Game & Economy Design Re-engagement with UA Ad Optimization IAP Optimization Cross Promotion Management Re-activation with UA
- 10. CONFIDENTIAL Growth Through Player Lifecycle Optimization ACQUISITION ACTIVATION RETENTION MONETIZATION CHURN STORE CONVERSION User Acquisition (Paid) User Acquisition (Organic/Viral) ASO FTUE/Onboarding Design Core Loop Design Meta-Game & Economy Design Re-engagement with UA Ad Optimization IAP Optimization Cross Promotion Management Re-activation with UA Re-imagining game’s Special Offer System with data science techniques - High monetization impact - No offer spamming
- 11. CONFIDENTIAL 15%-25% Simulation 5%-15% Match3 +40% RPG 25%-40% Racing UA traffic monetization improvement ARPU uplift based on the game genre 20%-50% ROAS relative increase Understanding potential ARPU uplift
- 12. CONFIDENTIAL Understanding potential ARPU uplift 15%-25% Simulation 5%-15% Match3 +40% RPG 25%-40% Racing UA traffic monetization improvement ARPU uplift based on the game genre 20%-50% ROAS relative increase Complexity of the game can be a significant indicator of potential uplift
- 13. CONFIDENTIAL Uplift realizes instantly Data science-based offer systems could be a significant contributor of extra revenue Revenue uplift realization over time Revenue
- 14. CONFIDENTIAL Long term consistent revenue uplift can be achieved Seasonal offers have a higher short term impact comparing to more stable and robust long term impact of Data science based special offer systems The long term uplift realization Revenue
- 15. Generally used offer systems lack: ○ Effectivity during progression of players ○ Ability to adapt to player’s demand which changes based on preferences and different game journey Why reinventing special offer systems using Data Science methods works?
- 16. Generally used offer systems lack: ○ Effectivity during progression of players ○ Ability to adapt to player’s demand which changes based on preferences and different game journey … and using data science techniques we can do better Why reinventing special offer systems using Data Science methods works?
- 17. CONFIDENTIAL Building impactful data science solution
- 18. CONFIDENTIAL Process for building impactful solution Developing proactive (progression based) and reactive models (preference based) Ensembles offer assets based on the personalization model and automatically creates offer visuals for each player in the game The right data = shows the current supply inefficiencies PRICING & CONTENT model AUTOMATIC OFFER DELIVERY SYSTEM ACQUIRING THE RIGHT DATA What drives ARPU uplift Understanding supply and demand for items, currencies, characters … and preference for pricing, content and discount
- 19. CONFIDENTIAL Process for building impactful solution Developing proactive (progression based) and reactive models (preference based) Ensembles offer assets based on the personalization model and automatically creates offer visuals for each player in the game The right data = shows the current supply inefficiencies PLAYER BASE SEGMENTATION PROBABILITY, EXPERTS AND MACHINE LEARNING MODELS MONITORING OF PERFORMANCE AND GATHERING FEEDBACK PRICING & CONTENT model AUTOMATIC OFFER DELIVERY SYSTEM ACQUIRING THE RIGHT DATA What drives ARPU uplift Understanding supply and demand for items, currencies, characters … and preference for pricing, content and discount
- 20. CONFIDENTIAL Process for building impactful solution What drives ARPU uplift Understanding supply and demand for items, currencies, characters … and preference for pricing, content and discount PRICING & CONTENT model AUTOMATIC OFFER DELIVERY SYSTEM ACQUIRING THE RIGHT DATA Developing proactive (progression based) and reactive models (preference based) Ensembles offer assets based on the personalization model and automatically creates offer visuals for each player in the game The right data = shows the current supply inefficiencies PLAYER BASE SEGMENTATION PROBABILITY, EXPERTS AND MACHINE LEARNING MODELS MONITORING OF PERFORMANCE AND GATHERING FEEDBACK Player data + monitoring community groups act as a player feedback
- 22. CONFIDENTIAL Understand player journey today tomorrow yesterday Day before yesterday Player mostly plays.. Player spends on these currencies but doesn't use ..
- 23. CONFIDENTIAL Understand player journey today tomorrow yesterday Player buys a new character, uses those items... Day before yesterday Player mostly plays.. Player spends on these currencies but doesn't use ..
- 24. CONFIDENTIAL Understand player journey today New game mode is introduced during progression tomorrow yesterday Player buys a new character, uses those items... Day before yesterday Player mostly plays.. Player spends on these currencies but doesn't use ..
- 25. CONFIDENTIAL Understand player journey today New game mode is introduced during progression tomorrow Player joins a guild/team yesterday Player buys a new character, uses those items... Day before yesterday Player mostly plays.. Player spends on these currencies but doesn't use ..
- 26. CONFIDENTIAL Understand player journey today New game mode is introduced during progression tomorrow Player joins a guild/team yesterday Player buys a new character, uses those items... Day before yesterday Player mostly plays.. Player spends on these currencies but doesn't use .. Models can learn a player preference from data Models have to adapt to changing player behavior due to changing game Liveops and progression Can also impact player’s preference short and long term
- 27. CONFIDENTIAL Creating player profile Player profile consists of necessary information to understand every player journey. Rank progression Items usage Upgrading preference Purchasing behavior Currency spending Usage of characters ● Percentage of revenue per type of purchase ● Percentage of purchases per type of purchases ● Conversion per type ● How many legendary cards equipped per character ● Frequency of changing characters ● Spending preference ● Usage of gems ● Spending hard currency on events ● Wide or depth upgrading ● Type of cards upgraded ● Mostly used items ● Current balance of items ● Last purchased item ● Season rank ● Player’s personal rank ● Duration to rank up ● How many rank downs
- 29. CONFIDENTIAL Showing only relevant offers is the key We use Data Science Models to pick all aspects of an offer based on data for each player in a game using ONLY existing content. Amount of resources Additional value Offer price Availability Type of chests Visuals & copy
- 30. CONFIDENTIAL What are we actually aiming for? Minimize additional value (value multiplier) Increase revenue per user Price Content distribution Value Availability Offer sets and their sequence Visual aspects
- 31. CONFIDENTIAL What are we actually aiming for? Minimize additional value (value multiplier) Increase revenue per user Price Content distribution Value Availability Offer sets and their sequence Visual aspects The value is in the personalization!
- 32. Impact on monetization KPIs Conversion ARPPU ARPU +14% +12% +20% +18% Data science models impact Conversion and ARPPU in a different way. Impact of those two metrics influences ARPU in a positive way. +4% +3%
- 33. Impact on monetization KPIs Conversion ARPPU ARPU +4% +3% +14% +12% +20% +18% Data science models impact Conversion and ARPPU in a different way. Impact of those two metrics influences ARPU in a positive way. Example: ○ Conversion - stable ○ ARPPU - increased by better upsell strategy from new pricing model Overall ARPU uplift
- 34. Impact on the Rest of Monetization There is always a small negative effect on baseline (called cannibalization) but this effect can be managed through understanding impact on different IAPs in the game and adapting supply based on that effect. uplift Hard currency Other offers Other items Hard currency Offer client Personalized offers Other Variant Control cannibalization
- 35. CONFIDENTIAL Path to ARPDAU uplift - the daily process Automatic daily data preparation Understanding player journey through basic and more complex KPIs. Automatic offer creation and offer delivery to the client Storing and delivering configs for every player in a game. The best next price point prediction Optimizing long term gains, revenue per impression Discount and offer visual strategy Definition of amount of discount and graphic design for offers Offer content preference Using proactive and reactive models to define plazer preference and changes in demand. Maximization of revenue per every offer impression Full Automation - System creates more than 20k unique offers daily
- 36. CONFIDENTIAL Complex models increase uplift & decrease transparency Probabilistic models Machine learning models Expert models Potential uplift: 30% Captured uplift: 5-10% Transparency : very high Maintenance costs: very low Potential uplift: 30% Captured uplift: 15-20% Transparency : medium Maintenance costs: low Potential uplift: 30% Captured uplift: 20-30%% Transparency : low Maintenance costs: high Transparency helps iterate the solution very fast, which means that fixing any potential bugs is easier (minimizing risks)
- 37. CONFIDENTIAL Transparency helps from the start Content definition based on behavior in a game Wide and Deep neural Network Combination of soft and hard currency spending with upgrading behavior, tuning behavior and purchasing behavior RFM BI method Probability modelling Price conversion modelling with adaptation and revenue maximization function Understanding payment potential through monetary parameters of player Understand which price points and sequence of price points maximizes revenue from each player individually
- 38. CONFIDENTIAL Putting it all together
- 39. CONFIDENTIAL Automatic Offer Creation Engine Automatic offer creation system: Ensembles offer assets based on the personalization model and automatically creates offer visuals for each player in the game. Benefit: ● Usage of templates which enables to create thousands of special offers in a matter of seconds ● Creative team chooses special offers templates and composition Graphical Assets Created in cooperation with Partner Offer Delivery Supports delivery on player level Offer Configs Storage Offer Creation Every player receives a personalized offer Personalization engine Players data
- 40. CONFIDENTIAL Monitoring reveals strengths and weaknesses Continuous improvement that does not negatively affect other aspects of the game Monitoring cannibalization development over time on different level of detail (segment/subsegment/offer etc.) Monitoring community discussions on different platforms (e.g. Discord) Monitoring performance over time comparing iterations of offers with different content distribution per segment/subsegment
- 42. Challenges ● Adaptation of community to lower discount offers than what they used to ● Sometimes, data tracking needs to be adjusted to create successful player profile ● Number of players significantly improves the ability to quickly test new ideas though A/B testing ● Huge amount of content can lead to a lot of offer variations with many different creative templates (QA) ● QA of thousands of automatically generated offer creatives ● Maintenance of some ML models can be difficult and time consuming Challenges & Learnings Learnings ● Personalized offers unlock high extra revenue potential ● Personalized offer impressions have no negative impact on retention ● Uplift potential is driven by minimizing gap between player demand and the offer supply in the game ● Cannibalization can be easily managed by understanding sinks and taps in the game ● Introduction of new offer types are usually successful if there is demand for them in your playerbase ● Impact of pricing and content model is additive ● “Fancy” visualization for seasonal offers can boost your revenue by additional 5-30%
- 43. THANK YOU If you have any questions you can reach me on Linkedin. Robert Magyar, Patrik Blanarik, Viktor Gregor Robert Magyar