Google Lighthouse - Financial Services

Talk given on 2017-03-29 about Ravelin's technology stack and GCP.

Transcript

1. Ravelin How Ravelin is leveraging data to combat fraud rather than managing infrastructure. Leonard Austin | 29.03.2017 ravelin.com
2. Meet the exec team Martin Sweeney CEO Leonard Austin CTO Nick Lally COO Mairtin O’Riada CIO Gerry Carr CMO Mairtin is as obsessive about catching fraudsters as he is about drinking tea. Before Ravelin, he had a storied career in international intelligence which gave him the perfect insight into the criminal mind. Previously Head of Fraud at Hailo, he’s now channeling all that experience into the perfect fraud product. Leonard has been designing, developing and deploying online technologies for over a decade. He is passionate about the power of agile teams, iterative development, continuous integration, automated testing and empowered engineers. He can often be found in the far corners of the earth. Martin was Founding Engineer at Hailo the taxi app, building the systems and team there over four years into a global business. He is a Physicist by training, but left behind an academic calling for the glamour of fraud detection and software. Martin is a father, husband and sporadic fisherman. Nick loves creating teams and processes in fast-growing organisations. Before Ravelin, he held Finance Director roles at Hailo and Canonical, the company behind Ubuntu, the OS that dominates cloud computing. An accountant at heart, he hates seeing any company lose money to pesky fraudsters. Gerry was the original marketing hire at Ubuntu and over five years there launched its cloud products, seeing it become the most used cloud OS in the world. Now juggles breaking phones with making people in the fraud prevention business aware of what Ravelin can do. ravelin.com
3. ravelin.com Fraud, the numbers. €4.79 out of €100 in online sales is at risk of fraud ⇧215% increase in the rate of online payment fraud since last year 27 attacks per 1000 transactions in Q4 2015 1.3% of revenue is lost by merchants on average *Source: Global Fraud Attack Index (2016)
4. Many M’s where is the N? Detection is hard
5. 3D Secure
6. 3D Secure
7. 3D Secure
8. The Fraudster ravelin.com
9. The Fraudster ravelin.com 50 iMacs + 19 Kebabs = £100k+ Taxi Driver Fraudster Dublin to Galway twice a day Chargeback Flights Tweets “looking forward to my flight”. Forgotten Bag 40 Stolen Credit Cards and Cloning Gear Fraudsters Top Five Budwiser Jack Daniels Guinness Gelato Malibu Cardz & Deetz & Fullz allthefreestuff@ youllnevercatchme@ ilovesmokingpot1992@ moneyzzz@
10. Ravelin Smarter Fraud Detection ravelin.com
11. Requirements Better than 3D Secure Fraud Detection Fast We sit within the checkout flow <300ms Process & Store Big Data All historic data from all clients. All real-time data from all clients. Global Markets Clients operate all over the world. Fraud is different from country to country. Highly Available Clients rely on Ravelin to decide whether to accept an order Secure We store/analysis bank cards, location data and PII (e.g. address, email, phone, transaction history etc) data ravelin.com
12. Coverage
13. ravelin.com Ravelin’s Graph DB is1000x the speed of traditional graph database solutions. Can be integrated into real-time scoring. Ravelin maps every user, card number, phone number, device, email, IP Address etc Ravelin Network
14. ravelin.com Ravelin’s rules engine allows configure custom rules based on client’s experience of fraud. Set the boundaries acceptable behaviour. We track purchasing behaviour prior to the transaction so that each decision is made with as much information as possible. Rules & Fingerprints
15. No other approach provides the accuracy that our customers demand at a speed and scale that fits in with how their businesses operate. Our data science teams deploy models across industries and fine tune them for clients. These models learn based on the fraud that our clients experience, taking out much of the analysis work that goes with traditional rules based systems. This leads to higher accuracy - less fraud, and fewer genuine customers being turned away. ravelin.com Machine Learning
16. Machine Learning
17. ravelin.com Bob.Smith71@qmail.net
18. Provider name free provider % Digits % Vowels average key distance capitalised % upper/lowercase length total number of characters tld total number of digits all uppercase disposable contains special characters (+,#,& etc) contains “.” contains given name contains family name Levenshtein distance (email & name) Email Features
20. Ravelin System - Deeper Drive Client Event→API →Decode →Enrich →Store →Read All History & Graph Data →Feature Extract →ML/Rules/BlackWhite Lists →Respond →Fanout ravelin.com After We Respond →Store all features →Notification - callback →Fanout →Re-Score →Train Models
21. Category Technology Type Storage Cassandra Infrastructure Queue RabbitMQ, Kafka, NSQ etc Infrastructure Big Data Hadoop Infrastructure SQL Postgres Infrastructure Instrumentation Graphite, Influx, Grafana Infrastructure Distributed Synchronization Zookeeper Infrastructure Cache Memcache, Redis Infrastructure Search ElasticSearch Infrastructure Machine Learning Custom Python Services Business Software 50 x Go Services Business
22. ravelin.com A Sample of Ravelin’s Platform
23. The DevOps Team Mike ravelin.com
24. ravelin.com A Sample of Ravelin’s Platform
25. ravelin.com A Sample of Ravelin’s Platform
26. Category Technology Type Storage Bigtable Infrastructure Queue Pub/Sub Infrastructure Big Data Bigquery + Dataflow Infrastructure SQL Cloud SQL Infrastructure Instrumentation Datadog Infrastructure Distributed Synchronization Zookeeper Infrastructure Cache Redis Infrastructure Search ElasticSearch Infrastructure Machine Learning Custom Python Services Business Software 50 x Go Services Business
27. ravelin.com Moral of the Story. Outsourcing infrastructure to GCP keeps Mike happy and provides me with more time to run the business.
28. Thank you @leonardaustin @ravelinhq ravelin.com