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.

07 - IDNOG04 - Leontinus Alpha Edison (Tokopedia) - Data Driven Innovation

469 views

Published on

07 - IDNOG04 - Leontinus Alpha Edison (Tokopedia) - Data Driven Innovation

Published in: Internet
  • Login to see the comments

07 - IDNOG04 - Leontinus Alpha Edison (Tokopedia) - Data Driven Innovation

  1. 1. Data-Driven Innovation
  2. 2. Once Upon a Time In Jakarta, Jan 2009
  3. 3. 1 Product Guy 1 Half Engineer as co-founder Don’t have business background AT ALL
  4. 4. China Japan Korea Indonesia baidu.com yahoo.co.jp naver.com google.com qq.com google.co.jp google.co.kr google.co.id taobao.com amazon.co.jp google.com facebook.com sina.com.cn youtube.com amazon.com youtube.com weibo.com google.com daum.net yahoo.com tmall.com fc2.com youtube.com detik.com hao123.com rakuten.co.jp facebook.com kaskus.co.id sohu.com nicovideo.jp tistory.com liputan6.com 360.cn twitter.com ppomppu.co.kr kompas.com tianya.cn facebook.com gmarket.co.kr wordpress.com
  5. 5. ? Online Forums Social Media Blog Platform
  6. 6. Data-Driven Mindset
  7. 7. TVC Optimization
  8. 8. Before
  9. 9. What did we do? We found that people tends to play with their phone while watching TV. IF we can captured number of people who visits our site, after watching our TVC. Means, our TVC works on that niche of people in that particular program. How? 1. Filter traffic only from Direct and Organic. 2. Filter traffic only from Mobile, WAP, Android and iOS. (Desktop show spikes too, but not all the time. IF we put Desktop in the calculation, it will reduce the contrast, making it harder to read.) 3. Use Google Trend (to analyse search terms).
  10. 10. Cancellation Rate
  11. 11. Overview
  12. 12. What did we do?
  13. 13. Result
  14. 14. Fraud Install Detector
  15. 15. Ads Agency
  16. 16. Machine Learning @Tokopedia
  17. 17. Fighting transactional fraud at Tokopedia Fraudulent transactions are handled MANUALLY using a backend system Only using Rule Engine and later on we counting on MANUAL INSPECTION However, number of cases overwhelms fraud team, it leaves a lot of unhandled cases.
  18. 18. Solutions? Using machine learning to discover fraudulent pattern rather than manually checking similar cases over and over again. We use machine learning model alongside the existing rule engine.
  19. 19. Transaction are not the only place for fraud What about MESSAGING?
  20. 20. USD 2 million USD 230k USD 600k
  21. 21. Don’t think “someone else will join and take care of this” — Mike Krieger of Instagram

×