Hajs się musi zgadzać, czyli o sprzedaży na Amazon AW$

Presentation from AWS User Group (2015.09.24) in Warsaw by Łukasz Cepowski from Xstream.

Transcript

1. Hajs się musi zgadzać, czyli o sprzedaży na Amazon AW$ Lukasz Cepowski DevOps Consultant Solution Architect @ Xstream
2. Boost sales with Amazon AWS lukasz@cepowski.com Hi!
3. Boost sales with Amazon AWS lukasz@cepowski.com
4. Boost sales with Amazon AWS lukasz@cepowski.com A business case
5. Boost sales with Amazon AWS lukasz@cepowski.com Post launch reality
6. Boost sales with Amazon AWS lukasz@cepowski.com
7. Boost sales with Amazon AWS lukasz@cepowski.com Let’s play a game :-)
8. Boost sales with Amazon AWS lukasz@cepowski.com
9. Boost sales with Amazon AWS lukasz@cepowski.com
10. Boost sales with Amazon AWS lukasz@cepowski.com What are KPIs? Sample SALES KPIs: • revenue growth • customer base growth • product performance • product engagement • purchases • popularity • more... Sample TECH KPIs: • page visits • unique visitors • direct / referal traffic • engagement time • transactions • searches • more...
11. Boost sales with Amazon AWS lukasz@cepowski.com Plain old HTTP access log 1.2.3.4 - - [23/Sep/2015:23:36:59 +0200] "GET /some/page.html&foo=hoo HTTP/1.1" 200 66 http://porn.com "Mozilla/5.0 (Mobile; Windows Phone 8.1; Android 4.0; ARM; Trident/7.0 Touch; rv:11.0; IEMobile/11.0; NOKIA; Lumia 1320) like iPhone OS 7_0_3 Mac OS X AppleWebKit/537 (KHTML, like Gecko) Mobile Safari/537" • 1.2.3.4 – where you are • GET /some/page.html&foo=hoo – what you are looking at • http:/ /porn.com – where you are comming from • „Mozilla/5.0...” – what device you are using
12. Boost sales with Amazon AWS lukasz@cepowski.com Playing Ping Pong with JavaScript (function () { var x = []; x.push(['v', 1]); x.push(['t', (new Date()).getTime()]); x.push(['u', Math.random().toString(36).substring(7)]); x.push(['sr', screen.width + 'x' + screen.height]); x.push(['cd', screen.colorDepth]); x.push(['o', screen.mozOrientation]); x.push(['vp', window.innerWidth + 'x' + window.innerHeight]); x.push(['cs', document.characterSet]); x.push(['l', navigator.language]); x.push(['h', window.location.href]); var q = x.map(function (e) { return e[0] + '=' + encodeURIComponent(e[1]); }).join('&'); e = document.createElement('script'); e.async = 1; e.src = 'pong.js?' + q; f = document.getElementsByTagName('script')[0]; f.parentNode.insertBefore(e, f); })(); 1.2.3.4 - - [23/Sep/2015:23:52:41 +0200] "GET /ping.js?1443045141479 HTTP/1.0" 200 486 http://cepowski.com/ "-„ 1.2.3.4 - - [23/Sep/2015:23:52:42 +0200] "GET /pong.js?v=1&t=1443045141790&u=akjb1grdx6r&sr=1360x768&cd=32&vp=1x1 &cs=UTF-8&l=pt-BR&h=http%3A%2F%2Fcepowski.com%2F HTTP/1.0" 200 48 "http://cepowski.com/"
13. Boost sales with Amazon AWS lukasz@cepowski.com It is all data
14. Boost sales with Amazon AWS lukasz@cepowski.com It is all data BIG DATA
15. Boost sales with Amazon AWS lukasz@cepowski.com Collecting data for Business Intelligence Amazon EC2 Amazon Elasticache Amazon S3 Amazon Redshift Graphing Software
16. Boost sales with Amazon AWS lukasz@cepowski.com So what can we do with data?
17. Boost sales with Amazon AWS lukasz@cepowski.com Looking for a new Iphone?
18. Boost sales with Amazon AWS lukasz@cepowski.com Let me Cloudsearch that for you 4. Return popular items first 1. Count references to popular item from other (access log, http referer) 3. Query Amazon Cloudsearch q.options={fields:[’referrers^3',’orders^2‚’likes^1.5’]} Show customers what is „hot” now to increase purchase likehood by customizing result relevance with KPIs (popularity, purchases, referrals, etc).
19. Boost sales with Amazon AWS lukasz@cepowski.com Need a second hand car nearby?
20. Boost sales with Amazon AWS lukasz@cepowski.com Geospatial search with Amazon Cloudsearch 4. Return only relevant and nearby items • Items within customer area • Sort results by geographical distance 1. Get customer IP address from access log 2. Get aproximate location (GeoIP / MaxMind) 3. Query Amazon Cloudsearch q=restaurant&expr.distance=haversin(35.621966,-120.686706,location.latitude,location.longitude)&sort=distance asc Boost search performance by returning items locally available to customer.
21. Boost sales with Amazon AWS lukasz@cepowski.com Dealing with customers with different taste
22. Boost sales with Amazon AWS lukasz@cepowski.com What is taste? Name Weizenbock American IPA Witbier Coffe Stout Janusz 2 1 4 5 Zbyszek 4 5 1 0 Czesław 1 2 5 5 Mirek 5 5 1 1
23. Boost sales with Amazon AWS lukasz@cepowski.com k Nearest Neighbours (kNN)
24. Boost sales with Amazon AWS lukasz@cepowski.com Recommendations • Item based (taste, kNN) • User based (taste, kNN) • Simmilarity (text, TF/IDF) Amazon CloudSearch Amazon Elastic MapReduce
25. Boost sales with Amazon AWS lukasz@cepowski.com Results • Recommended items for users
 „other users like you bought this and that” • Recommended items for items
 „simmilar items based on other users taste” • Users with simmilar taste
 „who likes simmilar products” • Simmilar items (based on description)
 „likely simmilar products”
26. Boost sales with Amazon AWS lukasz@cepowski.com Challenges • Scalability (i.e. 1M users against 100k items) • Sparse matrices • Weighting • Computation time • Amount of data
27. Boost sales with Amazon AWS lukasz@cepowski.com How to use AWS Cloud for BigData computation?
28. Boost sales with Amazon AWS lukasz@cepowski.com
29. Boost sales with Amazon AWS lukasz@cepowski.com Use Spot Instances to run calculations Runing 5x c4.large for ~two days Spot Pricing History
30. Boost sales with Amazon AWS lukasz@cepowski.com Tips & tricks • Infrastructure as Code • Orchestration • Compressed data on S3 • Download results only (!)
31. Boost sales with Amazon AWS lukasz@cepowski.com Future • More data • More tracking • Internet of things • Data driven marketing
32. Questions? Lukasz Cepowski DevOps Consultant Solution Architect @ Xstream