For Finish Line, higher tech yields lower fraud


While fraud is a risk assumed by merchants of all shapes and sizes, the nature of ecommerce adds a dimension of intrigue. The inability to physically see the customer, the anonymity of the online world and the ever-growing sophistication of criminals consistently present special challenges.

One issue is the need to reconcile two competing priorities: the necessity of remaining vigilant against online scams and the problem of over-declining — rejecting legitimate orders due to an excess of caution. In other words, striking the right balance that will enhance security and still allow volume to grow.

The problem will only become more urgent. According to the U.S. Commerce Department, ecommerce represented 9.6 percent of all retail sales in the United States as of the second quarter this year. Online retail spending for all of 2017 totaled $448 billion, a 16 percent increase over 2016.

Taking note of the trend and with 20 percent of its total revenue coming from digital sales, in 2016 Finish Line decided to reevaluate and upgrade its ecommerce protocols and began working with fraud detection technology from Riskified.

“Previously, Finish Line was using a rule-based fraud screening tool in combination with an in-house manual review team,” says Dajana Gajic-Fisic, head of ecommerce risk management at the athletic footwear and apparel chain.

Riskified operates in real time and can offer approval or denial of a transaction based on its proprietary technology. The company offers a 100 percent chargeback guarantee on transactions it erroneously approves.


“Up to today, a lot of merchants used systems similar to Finish Line’s, which are based on ‘rules’ models,” says Darren Migneault, an account manager at Riskified.

“If a certain set of criteria were met, a sale would be declined. Other methods were score-based, but in either case liability for chargebacks remained with the retailer.”

Those methods tend to “cast a wide net,” he says, exaggerating the fraud potential in the interest of thoroughness, but in doing so decline legitimate orders.

Riskified’s system uses algorithms to work with raw customer data to produce a highly accurate approval/denial decision. It considers several behavioral models, including the customer’s browsing habits and social media presence. The system uses those and other data points to build features, or statistical models, which form the basis of approving or denying a given transaction.

When a new client begins working with Riskified, the company assesses patterns of a merchant’s risk profile and contrasts them with that of similar companies.

“When a merchant comes in, we’re able to look at the type of data associated with good customers and fraud chargebacks and compare that with similar merchants, or across an entire industry,” Migneault says.

“This allows us to scale, since a new merchant generally only sends us their riskiest orders. As time goes on and more orders are received, more data is generated, which allows the machine learning process to self-improve.” It also eliminates the need to constantly tweak rules and manually modify the system.

That means the system becomes progressively more knowledgeable and accurate over time, allowing it to respond quickly to the ever-changing fraud landscape.

“If we’re working with a merchant with synchronous integration, we can complete our review and return a decision within one to three seconds from when the customer clicks ‘complete purchase,’” Migneault says.


Riskified’s system uses different types of integration depending on which ecommerce platform a merchant uses. Migneault says Riskified also aggregates data to build models of legitimate customers versus fraudsters that are specific to particular merchants and even to distinct product lines, such as sneakers — Finish Line’s mainstay product.

“Since sneakers have a strong secondary market, the sneaker model is very adept at locating this particular type of fraud,” Migneault says, “and steps up its game when new products are introduced.”

Machine learning can respond to threats of that nature instinctively and in less time than conventional methods. Riskified’s research indicates that trends affecting one merchant tend to occur across an entire industry or even beyond one industry. Since Riskified works with clients over a wide range of verticals, variations of data over different industries can be analyzed. The insights gained working with one client are then used to enrich the entire system, making it more effective.

One test of that effectiveness is a merchant’s willingness to send Riskified orders it had previously declined through previous fraud detection methods or internal channels; Migneault says Riskified can typically approve between 30 and 70 percent of those orders.

“We have seen a decrease in chargebacks as well as an increase in our approval rate,” Finish Line’s Gajic-Fisic says. “With the chargeback rate falling in a healthy range, we have an expanded opportunity to partner with Riskified further.”

“The more data we have in our models, and the more diverse the body of orders we process becomes (by region, customer type, payment type, etc.), the more accurate our performance will be,” Migneault says. “The growth potential is really exponential.”

Gajic-Fisic says the partnership is an excellent example of meeting a problem with a tailor-made solution.

“With Riskified, we have found a partner that can help us provide a great customer experience at minimum risk,” she says. “A machine-learning fraud solution helps us make the right decision fast while minimizing fraud and increasing revenue.”

The experience happens in both the virtual and real worlds. Based in Indianapolis and Boulder, Colo., Finish Line operates 900 stores in 47 states and Puerto Rico. About half are free standing or mall locations; the others are Finish Line-branded departments in select Macy’s stores.

Detroit-based Paul Vachon writes for various trade publications, in addition to feature stories for consumer magazines and books on Michigan history and travel.


Comments are closed.