4 Main Takeaways From Sift Science’s Webinar With OpenTable

Recently, Sift Science CEO Jason Tan sat down with Robbie Fritts, head of Risk and Fraud at OpenTable, to plan and present an MRC-hosted webinar, Grow More, Risk Less: A Case Study in How Machine Learning is Revolutionizing Fraud Prevention.

Hearing Robbie speak is always a treat, due to his deep experience with fraud that started at UpWork and continues at OpenTable, plus his thorough knowledge of the most cutting-edge approaches to handling risk. You can find the entire webinar on MRC’s website, but in the meantime here are 4 main takeaways to chew on:

The world is moving toward real-time delivery

Industry leaders like Amazon are setting the pace for what consumers expect from a purchasing experience. Not only is it easy to purchase items with a single click, but the retail giant is pushing the limits of what’s possible in terms of delivery – next-day delivery is becoming commonplace, with same-day close behind.

Robbie discussed how this new status quo of instant gratification has fueled OpenTable’s e-gifts product. “There’s a general consumer demand for real-time commerce and real-time fulfillment. They want the value of what they bought delivered instantaneously, so that presents new challenges,” said Robbie.

In terms of fighting fraud, rules just don’t scale

Robbie gave an overview of what it takes to create and maintain a rules-based system. The 4 main challenges are:

• Rules are static and don’t adapt in real time
• They over-weigh certain purchase behavior
• They’re easy for fraudsters to reverse-engineer
• They unscalable, needing costly internal investment

“Creating a new rule set is a time-intensive process. It can take days, months, weeks to make sure that you’re not creating false positives,” said Robbie.

A data-agnostic platform is key

Every business is unique, and that means that what might be a fraud signal for one company (like size 11 shoes for a footwear store) might not be applicable to others (a vacation rental marketplace doesn’t care about shoe size). That’s why one of Robbie’s main criteria when choosing a machine-learning fraud prevention solution was that it needed to take in custom data, unique to OpenTable.

In fact, Jason explained that providing as much data as possible  is ideal for a machine-learning system.  “We have the technology now to do a lot of math on very large data sets in a quick fashion. Why not toss the kitchen sink at it?”, asked Jason.

Machine learning enables automation and efficiency

The magic of machine learning comes in when you start, as Robbie put it, “defining user experiences.” That means setting up thresholds for what’s considered a “good” user and a “not good” user, then creating specific experiences for each category – which can even be automated.

Once you’re confident enough in your machine learning model’s accuracy to auto-reject and auto-accept certain orders based on the algorithmic findings, you can focus your time reviewing the few orders that fall in the gray area of “suspicious” or “unusual”.

And being able to automate aspects of your fraud process means you don’t have to hire more people as your business expands. “You never want to be growing your fraud team at the same rate as your revenue,” Robbie explained.

***

Want to learn more about machine learning and fraud? Check out our free ebook, Machine Learning: The Future of Fraud Fighting.