Shopify's machine-learning tactics take aim at false declines
Widespread data breaches have sparked a rise in account-takeover fraud using stolen consumer credentials, prompting merchants to crack down on suspicious purchases. But too many legitimate customers are getting caught in the process.
Improperly blocked transactions or false declines are on the rise—in 2017 they will exceed $300 billion in the U.S.—and merchants are trying various strategies to offset the problem, according to Julie Conroy, research director at Aite Group.
To battle false declines, e-commerce platform Shopify last year began applying machine-learning technology to gain insights from its vast pool of customer data available from hosting 500,000 websites, and this year it’s seeing improvements.
“The network effect from the collective data we have from all of Shopify’s merchants combined with continuous machine learning helps us rule out fraud from accounts that may have suspicious features or discrepancies," said Solmaz Shahalizadeh, Shopify’s director of merchant service algorithms.
Shopify uses a variety of internal and third-party tools to help its e-commerce customers counter card fraud. Its merchants are responsible for blocking their own fraud, but the Ottawa, Canada-based platform provider has come a long way in the last year in its ability to help merchants avoid blocking good customers, according to Shahalizadeh.
“The goal is enabling our merchants to accept the highest level of sales with the highest confidence, and particularly for smaller merchants who don’t have a dedicated fraud-analysis department, this can be a huge factor for profitability,” she said.
Common conditions that lead to false positives are discrepancies in consumers’ names, addresses, payment and shipping information, including purchases originating away from their home location, Shahalizadeh said.
“When Shopify merchants’ customers travel, a legitimate transaction can look suspicious, say if someone buys concert tickets in a city away from their home, and has them delivered to an unfamiliar address,” she said.
Shopify’s machine learning algorithms can often immediately see from that customer’s other purchases on other Shopify-hosted websites that the person is traveling and the purchase is legitimate, saving the merchant from aggravating or losing a good customer.
Shopify’s data science team uses open-source technology, and recent advancements have made sophisticated solutions more readily available and adaptable to retail settings, Shahalizadeh said.
“Not many years ago, the neural networks and deep learning we’re using today were cumbersome and hard to apply to our business, but these tools are evolving rapidly and we’re seeing huge benefits for developing our own tools to fight fraud,” she said.
Shopify is also applying machine learning to its data to refine and personalize the user experience for merchants operating sites on its platform.
“It has taken a lot of steps to get here, but just within the last year we have the kind of tools and a mass of data available so we can easily test and improve our own products using machine learning, Shahalizadeh said.
A cash-advance tool for merchants, introduced last year, is another byproduct of machine learning, crafted with Shopify’s in-house team of about 15 data scientists, she said.
“Using our own data with machine learning has made it possible for us to qualify a cash-advance borrower and make an offer within about 40 minutes versus days,” Shahalizadeh said.