"Artificial Intelligence" as a concept still eludes some financial executives, but for Signifyd the most important thing to understand is it can be a strong weapon in the fight against fraud.
"Anyone with automated decision-making wants to brand it as artificial intelligence," said Signifyd co-founder and chief operating officer Mike Liberty, who worked as a risk manager at PayPal before starting the San Jose, Calif.-based Signifyd. "But you have to split AI between traditional rules-based screening systems, which are starting to show their age, and those that are based on machine learning, or acquiring knowledge on its own."
Signifyd serves 5,000 clients, mostly small e-commerce businesses and some larger online retailers such as Jet.com, Lacoste and Peet's Coffee & Tea. It monitored $5.6 billion in transaction volume in 2015. The company's AI-driven transaction screening is a particularly good fit for businesses lacking the manpower or expertise to manually study e-commerce transactions and make accurate authorization decisions, Liberty said.
These smaller retailers may face pressure to adopt AI, which has started to pick up steam in the payments industry.
"Artificial intelligence is by no means a new term, and it is so broad that it encompasses many different types of capabilities," said Julie Conroy, research director and fraud expert with Boston-based Aite Group. "But I'm actually quite bullish about the use of machine learning technologies for fraud prevention."
Businesses that are either piloting or in production with machine learning fraud systems have generally been positive about their experiences, Conroy said. Signifyd clients fall into that category.
Signifyd offers an online fraud screen dashboard for its customers that provides feedback on transactions monitored through a system that creates "a giant digital footprint online," said Jay Sung, chief operating officer of EcoReco, a Signifyd client and company that sells electronic scooters.
When a customer orders from the EcoReco website, the company gets a Signifyd score for that transaction that takes into account everything from basic customer information, internet protocol address and proxy, to product ordering records from a specific e-mail address and what other types of social media accounts and activity are associated with that address.
After the company gets the transaction score, it can decide whether to let the transaction through, seek more information from a customer or deny the transaction.
"When our company started, we had many fraudulent orders," Sung said. "Every time we lost something, we learned something. But we really didn't have time to investigate all of these orders."
After researching the AI field, EcoReco found Signifyd fit what it was seeking. Ultimately, Signifyd recovered 5% to 10% of lost sales, in addition to completely eliminating the fraud for EcoReco, Sung added.
Signifyd is able to boast of completely stopping fraud for a client because the company accepts the financial liability for any chargebacks on transactions it approves that turn out to be fraudulent.
Taking the financial burden away from clients is a two-way street, Liberty said. "The feedback point is critical for us in data security," he added. "So many fraud prevention solutions that exist don't really have a feedback route on whether the provider made a good or bad prediction on a transaction."
By offering to pay for any bad decision, Signifyd's customers will automatically provide feedback and alert the company about a chargeback. "We have to then reincorporate that information," Liberty said. "That is a critical step to close that loop, if you want to apply machine learning to this domain. Plus, it's an interesting positive effect of our business model."
