Spark Summit EU talk by Johnathan Mercer
0 likes•736 views
Apptopia is a mobile app intelligence company that uses public and proprietary data along with machine learning models to provide insights about the app market. They have transformed their analytics capabilities by moving from slow single-machine models to using Apache Spark on large clusters, allowing them to train and score models much faster. Going forward, Apptopia aims to continue scaling their platform while focusing on key factors like app similarity, bundledness, and competitiveness to better understand the complex, dynamic global app market.
Spark Summit EU talk by Johnathan Mercer
- 1. Solving the app market with only grit, hustle, and a Spark cluster Johnathan Mercer VP of Analytics [email protected]
- 2. @Apptopia
- 3. 30 minutes is a lot to ask 3
- 4. 4 I want to fill your pockets
- 6. Apptopia is a mobile app intelligence company 6
- 7. 7 Company Overview >$4m raised from major investors like: ~40 employees ~6 years in business Trusted by > 35,000 publishers worldwide.
- 8. 8 Standing “on the shoulders of giants” Eli Sapir Co-Founder & CEO Jonathan Kay Co-Founder & COO Serge Balyuk VP of Engineering
- 9. apptopia.com 9 We power mobile intel data for the best
- 11. apptopia.com 11 There is so much more to mobile than games
- 12. 12 User demographics and behavior 3%97% 40% 60% GoToWebinarWealthfront
- 17. Apptopia is on a mission to solve the app market 17
- 19. 19 We combine public and proprietary data Public Data Ranks Ratings Reviews Price Publisher Descriptions Screenshots Versions Historical Performance of >200K Apps
- 20. 20 Public data is predictive of performance
- 21. 21 Rank is the most predictive feature Instagram
- 22. 22 Rank is also the most complex (i.e., fun) feature App . . . . . . Downloads Revenue
- 23. 23 There are many categories and sub-categories https://developer.apple.com/app-store/categories/
- 24. 24 The constellation of ranks is complex and constantly evolving
- 25. 25 We build models with data from connected apps Connected Apps All Apps
- 26. 26 For any app, day, country
- 28. 28 First generation training and scoring Rserve
- 29. Slow scoring (~2 months) 29 This was not a viable path forward
- 31. 2 months to 2 days 31
- 32. Systematically compare hundreds of models using 20x more data 32
- 33. Spark and other open source tools have transformed us 33
- 34. 34
- 35. 35 Where do we go from here?
- 36. The global app market is a complex dynamical system 36
- 38. 38 Focus as we scale
- 39. Here’s 4 things we learned 39
- 40. First solve for the human side of the equation 40
- 42. Make big data small data 42
- 43. Edges are more important than nodes 43
- 44. Thank You
- 45. Solving the app market with only grit, hustle, and a Spark cluster Johnathan Mercer VP of Analytics [email protected]