How machine learning enables real-time commerce

There is no question about it: the real-time, on-demand economy is disrupting ecommerce. These days, you can order rides, buy groceries, rent a car, make a dinner reservation, and more with a single tap on your smartphone – and each service arrives in as little as minutes. Against this backdrop of speed, more consumers are expecting – and even demanding – a fast and frictionless user experience. The challenge for businesses is to meet these high expectations and stay competitive – all without increasing risk.

Challenges of fraud prevention for real-time businesses

All types of ecommerce companies struggle with payment fraud, but time-sensitive businesses that fulfill orders in real time face unique challenges. Take last-minute hotel bookings as an example. When someone reserves a room, they may be looking to check in immediately. Delay approving that transaction, and there is a high chance you will lose that customer. When you are a business competing on speed and convenience, you do not often have the luxury of manually reviewing new orders.

Unfortunately, fraudsters are aware of this short window for investigation and exploit it for their gain. If an on-demand business does not have adequate fraud defenses, a criminal can easily maximize their returns off of a stolen credit card before they are blocked from making further purchases.

As a solution to managing risk in real time, more and more businesses that operate with limited manual review timeframes are turning to machine learning-based solutions to automate fraud decisions. With machine learning, computers use specially created algorithms and mathematical formulas to learn from historical data to predict likely future scenarios. Think of machine learning as the equivalent of a human learning from experience. As a machine learning system ingests more data, its predictions get increasingly accurate.

Machine learning vs. rules

On-demand businesses are discovering that traditional fraud prevention systems like rules engines simply cannot keep pace or deliver effective results. Rules engines are reactive, based only on patterns observed in the past, from activity that has already happened. Machine learning, on the other hand, can learn from historical examples of fraud and adapt to constantly-evolving fraud patterns, predicting future fraudulent activity based on shared attributes or actions.

Fraudsters can quickly work around rules, leaving businesses vulnerable to new types of fraud attacks. Also, rules-based systems are not capable of detecting the subtle nuances of fraud, treating it in “black and white” terms that can easily lead to false positives. Machine learning can look at thousands of different signals and identify red flags that humans cannot easily spot, which lowers the risk of mistakenly flagging good customers as bad.

Here is a table that compares some of the pros and cons of each type of fraud-prevention approach:

Qualities of an effective machine-learning solution

There are a number of machine learning-based fraud prevention solutions available today. So, what makes a machine learning system effective at fighting fraud? Here are a few qualities to look for:

• Speed: works in real time. Fraudsters constantly find new and innovative ways to commit fraudulent transactions, so an effective tool must be able to respond to changing fraud patterns as they occur.

• Scale: draws upon a vast and varied network. The key to doing machine learning well is to leverage large volumes of high-quality data. The more data the tool has access to, the more accurate it will become.

• Sophistication: can accurately analyze and process enormous data sets. Since suspicious behavior is often buried within streams of data, an effective tool must be able to extract individual elements, understand their significance, and deliver precise results.

Want to discover more about how real-time businesses are using machine learning to prevent fraud? Read our hotel booking case study or our food delivery case study. You can also read more about what to look for in a machine learning fraud detection solution.

About Jason Tan

Jason Tan is the co-founder and CEO of Sift Science, a San Francisco-based company that helps online businesses fight fraud using large-scale, real-time machine learning. Fueled by a passion for building great products and amazing teams, Jason has also held leadership and engineering roles at BuzzLabs, Optify, and Zillow.

About Sift Science

Sift Science is a fraud detection solution for websites and mobile applications. The platform utilizes large-scale machine learning to detect fraudulent users and help online, ecommerce businesses to avoid falling victim to fraud. As the internets trust layer, Sift Sciences mission is simple yet powerful: make these online experiences faster, smoother, and safer – using the smartest technology around.