Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

LTV predictions for Growth and UA activities

73 views

Published on

Intermediate and sophisticated approaches towards LTV measurement. Examples of custom and advanced solutions built to understand LTV of the users. Connection between LTV and user segmentation, offers based on it. Presentation based on examples and case studies.
- Presentation run during on of GameCamp webinars; http://www.gamecamp.io/events/understanding-prediction-ltv/
- All GameCamp webinars: http://www.gamecamp.io/events/

Published in: Mobile
  • Be the first to comment

  • Be the first to like this

LTV predictions for Growth and UA activities

  1. 1. Understanding, optimising and predicting LTV based on data in mobile gaming Robert Magyar Head of Data Science for GameCamp August 13th 2020
  2. 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, etc) we have: ● Analyzed data of 1B+ players. ● Optimized millions of dollars in media spend. ● Brought millions of dollars in revenue uplift. Robert Magyar Head of Data Science & General Manager at Superscale
  3. 3. What is SuperScale? We are the growth partner for world's top game developers focusing on UA, business analytics & monetization.
  4. 4. Our Partners
  5. 5. Usual problem that game studios have Day Situation 1. studio does UA 2. it works well
  6. 6. Usual problem that game studios have Day Situation 1. studio does UA 2. it works well 3. spend is increased 4. performance drops
  7. 7. Usual problem that game studios have Situation 1. studio does UA 2. it works well 3. spend is increased 4. performance drops 5. recoup too far away 6. UA cannot scale further Day Increase in spend → lower quality players and higher CPI
  8. 8. How to combat scaling issues in your game?
  9. 9. How can games really scale their revenue?
  10. 10. How can games really scale their revenue? Optimize LTV of your playerbase Optimize engagement metrics (retention etc) Continuous Scaling Bring quality players through User Acquisition
  11. 11. How can games really scale their revenue? Optimize LTV of your playerbase Optimize engagement metrics (retention etc) Continuous Scaling Bring quality players through User Acquisition … without development of new game features / game modes & without adding unique things to the game
  12. 12. How can games really scale their revenue? 1. Measure 2. Predict 3. Optimize Optimize LTV of your playerbase Optimize engagement metrics (retention etc) Continuous Scaling Bring quality players through User Acquisition … without development of new game features / game modes & without adding unique things to the game 1. Measure 2. Predict 3. Optimize
  13. 13. 2 areas to make the biggest impact using data 1. UA optimization through data - Marketing & Analytics working together ● predict results of UA actions, enable seeing UA trends and patterns ● target better quality players (lookalikes creation based on playerbase segmentation)
  14. 14. 2 areas to make the biggest impact using data 1. UA optimization through data - Marketing & Analytics working together ● predict results of UA actions, enable seeing UA trends and patterns ● target better quality players (lookalikes creation based on playerbase segmentation) 2. IAP LTV optimization through data ● Optimization of rotating / seasonal / progression offers ● If IAP makes 5% revenue -> you need 300% improvement to have significant impact on overall revenue
  15. 15. 2 areas to make the biggest impact using data 1. UA optimization through data - Marketing & Analytics working together ● predict results of UA actions, enable seeing UA trends and patterns ● target better quality players (lookalikes creation based on playerbase segmentation) 2. IAP LTV optimization through data ● Optimization of rotating / seasonal / progression offers ● If IAP makes 5% revenue -> you need 300% improvement to have significant impact on overall revenue Doing both at the same time is the key (synergizing).
  16. 16. Improving LTV of attributed players Marketing & Analytics working together
  17. 17. Marketing without proper analytics support? Issues that we frequently see: - Many different sources of data - Hard to drill down (campaign, cohorts, creatives, adsets, countries, platforms etc) - No accurate prediction system
  18. 18. Marketing & analytics working together Short term benefits - better informed day-to-day decisions about changes in creatives, campaigns, ad sets etc Mid term benefits - understand if weekly and monthly UA strategy regarding targeting and creatives work Long term benefits - de-risking scaling - spend budgeting and understanding recoup/breakeven day
  19. 19. Do you have the right tools?
  20. 20. Do you have the right tools?
  21. 21. How are your creatives doing? Improving creative targeting => bringing higher quality players => improving LTV Creatives with higher D3 and D7 ROAS brings higher quality players to your game. It is important to spot them as soon as possible. Marketing & analytics working together Important to see outlier as soon as possible and act on that
  22. 22. How are your strategies performing? Quickly identifying better strategies => bringing higher quality players => improving LTV Comparison of ROAS (Return On Ad Spend) benchmarks can give you great idea how your weekly strategies perform. Marketing & analytics working together Starting to see decline in D7 and D28 ROAS
  23. 23. What is actually UA recoup/breakeven day? Identifying the most long-term profitable campaigns and reallocating spend to them → improving LTV Yearly prediction can help with: - Spend budgeting - Spend allocation (prediction on the campaign, adset or creative level) Marketing & analytics working together CPI Break-even day - 16th day Profit Threshold
  24. 24. Connecting LTV with spend strategy Understanding possible spend increase to achieve desired recoup → improving LTV Increase in spend means increase in CPI. You can estimate what CPI you get when you increase your spend and compare it to your LTV to understand UA payback. Marketing & analytics working together
  25. 25. LTV/ROAS predictions & decision making [Careful] Your prediction model can overestimate the revenue ● This can result in overconfidence in the current performance and overspending, thus affecting UA manager decision making Model error (%, * 100) Numberofcohorts Underestimating real revenue Overestimating real revenue Marketing & analytics working together Leads to wrong decision making Leads to missing opportunities
  26. 26. Creating lookalikes based on player data Targeting most profitable players for your game → increasing LTV Our experience from different games : Creating your own lookalikes based on player segments can improve ROAS relatively by more than 20%. [changes coming to iOS, still great for Android] Marketing & analytics working together
  27. 27. Creating lookalikes based on player data Targeting most profitable players for your game → increasing LTV Our experience from different games : Creating your own lookalikes based on player segments can improve ROAS relatively by more than 20%. [changes coming to iOS, still great for Android] 1. You need to find the best possible representation of your top players Marketing & analytics working together
  28. 28. Creating lookalikes based on player data Targeting most profitable players for your game → increasing LTV Our experience from different games : Creating your own lookalikes based on player segments can improve ROAS relatively by more than 20%. [changes coming to iOS, still great for Android] 1. You need to find the best possible representation of your top players 2. You need to ask at least these question: - Is this group great at buying IAPs, do they do it frequently? - Is this group heavily engaged, does their engagement grow over time? - Does this group of players watch ads frequently? Marketing & analytics working together
  29. 29. Creating lookalikes based on player data Targeting most profitable players for your game → increasing LTV Our experience from different games : Creating your own lookalikes based on player segments can improve ROAS relatively by more than 20%. [changes coming to iOS, still great for Android] 1. You need to find the best possible representation of your top players 2. You need to ask at least these question: - Is this group great at buying IAPs, do they do it frequently? - Is this group heavily engaged, does their engagement grow over time? - Does this group of players watch ads frequently? 3. A/B test this group against your best lookalike / audience to date (your benchmark) Marketing & analytics working together
  30. 30. Creating lookalikes based on player data Targeting most profitable players for your game → increasing LTV Our experience from different games : Creating your own lookalikes based on player segments can improve ROAS relatively by more than 20%. [changes coming to iOS, still great for Android] 1. You need to find the best possible representation of your top players 2. You need to ask at least these questions: - Is this group great at buying IAPs, do they do it frequently? - Is this group heavily engaged, does their engagement grow over time? - Does this group of players watch ads frequently? 3. A/B test this group against your best lookalike / audience to date (your benchmark) 4. Evaluate and Profit Marketing & analytics working together
  31. 31. Creating lookalikes - Player segments in your game Marketing & analytics working together Example overall LTV of the game LTV LTV model predictions Legend Real LTV numbers
  32. 32. Creating lookalikes - Player segments in your game Marketing & analytics working together Convex-like development (Cluster 2) Faster early growth, closer to straight line (Cluster 1) Logarithmic-like development, earlier flattening (Cluster 3) LTV LTV LTV LTV Example overall LTV of the game LTV model predictions Legend Real LTV numbers
  33. 33. Creating lookalikes - Player segments in your game Marketing & analytics working together You need to find those player segments LTV LTV LTV Example overall LTV of the game Convex-like development (Cluster 2) Logarithmic-like development, earlier flattening (Cluster 3) LTV model predictions Legend Real LTV numbers Faster early growth, closer to straight line (Cluster 1) LTV
  34. 34. Do you have enough data for LTV predictions? Marketing & analytics working together Unusual growth of revenue in the cohort can be the clue of not having enough players to work with. => Number of players needed is based on conversion and amount of spend. LTV
  35. 35. Improving IAP LTV Maximizing revenue from your special offers
  36. 36. Are you leaving money on the table? Imagine having 10000 special offers in your favourite game, which one would you want to see in your next session? Would you be happy with any random offer? Improving IAP LTV
  37. 37. Are you leaving money on the table? Imagine having 10000 special offers in your favourite game, which one would you want to see in your next session? Would you be happy with any random offer? Improving IAP LTV - Would you want $5 or $100 price? - What amount of coins? - How much discount? - What visual aspects would offer have that would impress you? - ….
  38. 38. Are you leaving money on the table? Many games are showing offers to players that players simply don’t want. Either are offers random or picked only by simple rule-based systems. How do you battle this? Improving IAP LTV
  39. 39. Are you leaving money on the table? Many games are showing offers to players that players simply don’t want. Either are offers random or picked only by simple rule-based systems. How do you battle this? Personalize offers better. Many other industries do personalization of content which helps them improve monetization experience for each customer. Improving IAP LTV
  40. 40. Are you leaving money on the table? Showing relevant price & content at the right time = increasing LTV of your playerbase. Improving IAP LTV
  41. 41. 3. Dynamic personalization - 1000s of dynamically created offers - Targeting based on many different parameters (even complex ones like purchase aggressivity) - Use of AI / Machine learning or probability model 1. Random offers - 100-1000s predefined offers - No targeting - No filtering of offers 2. Rule-based system - 100-1000s predefined offers - Targeting based on few “ifs”, e.g.: - Previous purchase price point - Conversion on certain point - Amount of currency Special Offer Systems - Usual Types Potentially missing IAP revenue using this system: Missing 50-100% IAP revenue Missing up to 50% IAP revenue Maximizing IAP revenue Improving IAP LTV
  42. 42. Personalization results in significant IAP LTV uplift Works for any IAP focused game genre: From 20% => 100% (match 3 => RPG game) Better for complex games Different game modes + Lots of different items to sell Can be done safely Start with 5% of players then expand Case study: Observed IAP LTV growth from personalization system. Day TRUE LTV UPLIFT Legend: LTV curve - dynamic personalization LTV curve - rule-based system Improving IAP LTV +35% LTV
  43. 43. Showing only relevant offers is the key We use Machine Learning 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
  44. 44. Your special offers are great if you can …. Increase revenue per user Improving IAP LTV Minimize discount/ value multiplier
  45. 45. Your special offers are great if you can …. Minimize discount/ value multiplier Increase revenue per user Price Content distribution Value Availability Offer sets and their sequence Visual aspects Optimization: => The value is in the personalization! Improving IAP LTV
  46. 46. Personalization process 1. DATA GATHERING 2. LEARNING PREFERENCES OF PLAYER SEGMENTS 3.NEW OFFERS GENERATION 4.DELIVERY & INCREASE IN LTV ● Preprocessing Data into model-ready state ● Analyzing of player behavior (behavioral parameters) ● Learning preferences about segments of players ● Taking into account the Player’s browsing behavior along with their friends’ and their segment colleagues + interaction with customizations ● Creating offers based on available Cosmetics, chests and resources in the Inventory and the Shop slots layout ● Each player receives personalized offer he is most likely to buy ● Time Limited Offers only ● From the response (buy or not) we strengthen understanding of players’ preferences Improving IAP LTV
  47. 47. Special Offer delivery system - Infrastructure RAW DATA PLAYERS EACH PLAYER/SEGMENT GETS THEIR UNIQUE PERSONALIZED OFFERS DATA GATHERING MACHINE LEARNING (LEARNING PREFERENCES OF PLAYER SEGMENTS) OFFER ASSETS INCREASE LTV PLAYERS DATA Improving IAP LTV
  48. 48. Special Offer delivery system - Infrastructure RAW DATA PLAYERS DATA WAREHOUSE GOOGLE BIGQUERY EACH PLAYER/SEGMENT GETS THEIR UNIQUE PERSONALIZED OFFERS DATA GATHERING MACHINE LEARNING (LEARNING PREFERENCES OF PLAYER SEGMENTS) INCREASE LTV PLAYERS DATA OFFER ASSETS Improving IAP LTV
  49. 49. Special Offer delivery system - Infrastructure RAW DATA PLAYERS DATA WAREHOUSE GOOGLE BIGQUERY DATA PROCESSING DATAFLOW SCHEDULER CLOUD FUNCTIONS EACH PLAYER/SEGMENT GETS THEIR UNIQUE PERSONALIZED OFFERS DATA GATHERING MACHINE LEARNING (LEARNING PREFERENCES OF PLAYER SEGMENTS) INCREASE LTV PLAYERS DATA OFFER ASSETS Improving IAP LTV
  50. 50. Special Offer delivery system - Infrastructure RAW DATA PLAYERS DATA WAREHOUSE GOOGLE BIGQUERY Behavioral Modeling: Understanding players preference through the set of behavioral parameters Price Modeling: Understanding monetary potential of players DATA PROCESSING DATAFLOW SCHEDULER CLOUD FUNCTIONS DATA MODELING ML ENGINE EACH PLAYER/SEGMENT GETS THEIR UNIQUE PERSONALIZED OFFERS DATA GATHERING MACHINE LEARNING (LEARNING PREFERENCES OF PLAYER SEGMENTS) INCREASE LTV MODEL STORAGE GOOGLE CLOUD STORAGE PLAYERS DATA OFFER ASSETS Improving IAP LTV
  51. 51. Special Offer delivery system - Infrastructure RAW DATA PLAYERS DATA WAREHOUSE GOOGLE BIGQUERY Behavioral Modeling: Understanding players preference through the set of behavioral parameters Price Modeling: Understanding monetary potential of players DATA PROCESSING DATAFLOW SCHEDULER CLOUD FUNCTIONS DATA MODELING ML ENGINE OFFERS CREATION OPTIMIZED FOR BEST ARPU EACH PLAYER/SEGMENT GETS THEIR UNIQUE PERSONALIZED OFFERS DATA GATHERING MACHINE LEARNING (LEARNING PREFERENCES OF PLAYER SEGMENTS) INCREASE LTV MODEL STORAGE GOOGLE CLOUD STORAGE PLAYERS DATA OFFER ASSETS Improving IAP LTV
  52. 52. Special Offer delivery system - Infrastructure RAW DATA PLAYERS DATA WAREHOUSE GOOGLE BIGQUERY Behavioral Modeling: Understanding players preference through the set of behavioral parameters Price Modeling: Understanding monetary potential of players DATA PROCESSING DATAFLOW SCHEDULER CLOUD FUNCTIONS DATA MODELING ML ENGINE OFFERS CREATION OPTIMIZED FOR BEST ARPU OFFER DELIVERY UNIQUE TO PLAYER SEGMENT EACH PLAYER/SEGMENT GETS THEIR UNIQUE PERSONALIZED OFFERS DATA GATHERING MACHINE LEARNING (LEARNING PREFERENCES OF PLAYER SEGMENTS) INCREASE LTV MODEL STORAGE GOOGLE CLOUD STORAGE PLAYERS DATA OFFER ASSETS Improving IAP LTV
  53. 53. Special Offer delivery system - Infrastructure RAW DATA PLAYERS DATA WAREHOUSE GOOGLE BIGQUERY Behavioral Modeling: Understanding players preference through the set of behavioral parameters Price Modeling: Understanding monetary potential of players DATA PROCESSING DATAFLOW SCHEDULER CLOUD FUNCTIONS DATA MODELING ML ENGINE OFFERS CREATION OPTIMIZED FOR BEST ARPU OFFER DELIVERY UNIQUE TO PLAYER SEGMENT CONTINUOUS IMPROVEMENTS EACH PLAYER/SEGMENT GETS THEIR UNIQUE PERSONALIZED OFFERS DATA GATHERING MACHINE LEARNING (LEARNING PREFERENCES OF PLAYER SEGMENTS) INCREASE LTV MODEL STORAGE GOOGLE CLOUD STORAGE PLAYERS DATA OFFER ASSETS Improving IAP LTV
  54. 54. Personalization results in significant IAP LTV uplift Works for any IAP focused game genre: From 20% => 100% (match 3 => RPG game) Better for complex games Different game modes + Lots of different items to sell Can be done safely Start with 5% of players then expand Case study: Observed IAP LTV growth from personalization system ($15M+/year game) Day TRUE LTV UPLIFT Legend: LTV curve - dynamic personalization LTV curve - rule-based system Improving IAP LTV
  55. 55. Summary 1. Struggling with the UA after increasing the spend to the levels when CPI gets higher than LTV is a common issue. 2. Leveraging your game’s data to the full extent can significantly improve LTV of your game, which can unlock further spend scaling ● Special offer personalization utilizing existing game content can bring +20%-100% extra IAP revenue ● Optimizing your UA decision making using data can result in additional LTV/ROAS increases: ○ Spend reallocation ○ Spend budgeting ○ Lookalike creation ○ Creative strategy testing ...
  56. 56. Thank You robert.magyar@superscale.com

×