Recently, Sift Science hosted our Architecting Trust and Safety Summit in New York City. It was the first event we’ve held that brought together industry thought leaders to share best practices and give insight into where the online risk, abuse, and trust and safety industry is headed – specifically from a technical perspective.
Here are some high-level takeaways from each session:
Topic: Managing Risk and Building Trust
By Chris McClean – Vice President, Research Director, Forrester Research
There are 3 main drivers of how a business can build consumer trust:
Integrity – Can a user trust that you are not going to do anything shady with their personal information?
Competence – Does your abuse team have the ability to find errors and fix them quickly?
Transparency – When you take an action on an account or transaction (for example, stopping the order) do you give an explanation to the user or do you treat it as a black box experience that can cause customer frustration and churn?
Topic: Building a Product Optimized for Trust
By Kevin Lee – Trust and Safety Architect, Sift Science
Online businesses are becoming more diverse
Traditional commerce, community, and content-based businesses are all combining and moving into each other’s spaces.
For example, Facebook started in the “community” space but has grown its product line to include shopping and peer-to-peer payments. Google started in the “content” space with a search engine, but has moved into the community (Google+) and commerce (Google Shopping) spaces.
In order to effectively fight abuse on your platform, you must create joint utility function goals between your abuse team and the product teams to produce win/win scenarios.
Transparency – Create a system of trust by enabling users to:
Report bad things
Get educated on abuse
Go through soft responses (for example, 2-factor authentication) instead of blocking a transaction.
Topic: Reducing Chargeback and Security Risks in Digital Currency Exchanges
By Soups Ranjan – Director of Data Science, Coinbase
At Coinbase, they use both human review and machine learning to combat fraud. Fraud analysts label tons of accounts to feed the supervised and unsupervised machine learning models to learn.
Core to the investigation process is verifying the name and address of the customer across various sources (user email, government issued ID, SSN, bank account, etc). Thus, in order for a fraudster to be successful they need to compromise name and address across multiple data sources.
They use their own in-house models and a models from Sift Science together to get maximum coverage.
They limit impact of fraud by giving users a daily buying limit similar to a credit line. The higher risk the user the less the credit line and vice versa.
Topic: Measuring the Impact of Risk Systems
By Jevin Bhorania – Senior Data Scientist & Product Editor, Square
Evaluate model performance (recall, manual caseload created, and insults) on an incremental basis
A risk system can be measured in 3 main parts: models, decisions, and the overall system
Creating a holdout group where you actively choose to let through what you think are bad transactions is a great way to get a ground truth around false positives and loss rates. Just make sure to tag these accounts/transactions and monitor very closely.
Based on the level of audience engagement and questions, the summit was a huge success. Stayed tuned for the next summit and let us know if you’re interested in presenting. Hope to see you at our next event!
Kevin Lee is VP of Digital Trust & Safety at Sift where he helps customers implement strategies that cross-functionally align risk and revenue programs. Prior to Sift, he has spent the last 14+ years leading various risk, chargeback, spam/scams, and trust and safety organizations at Facebook, Square, and Google.
Stop fraud, break down data silos, and lower friction with Sift.
Achieve up to 285% ROI
Increase user acceptance rates up to 99%
Drop time spent on manual review up to 80%
Your information will be used to contact you about our service and subscribe you to our direct marketing communications. You can, of course, unsubscribe at any time. Please see our Website Privacy Notice.