How does network tokenization lead to higher authorization rates for merchants?
Authorization rates are incredibly important to merchants because they increase both revenue and customer satisfaction. The higher the card authorization rate, the greater likelihood for repeat customer transactions, which results in higher business revenue.
One way to boost auth rates is through network tokenization. “Think of a network token as a fake 16 digit number that’s assigned to each of your card numbers that you have within your wallet,” explained Magats. Network tokens provide an alternative number for the consumer’s card, and when the issuer receives this transaction, it recognizes the number in the same way as the actual debit or credit card.
So how does network tokenization increasing auth rates? Well, the network token is a number known only by the merchant—PayPal, for example—the issuer, and the network provider, making it difficult for a criminal to access or use the card number fraudulently. It also offers a cryptogram, or a piece of data only known between the above stated three parties, which increases the safety and security of a transaction.
If a customer’s card is lost or stolen, they can get a different credential that allows them to continue to make purchases with the same account, even if the card has been canceled, because the token is not known outside of the intimate ecosystem. If the lost or stolen card is attached to billing agreements, they can continue to seamlessly pay those merchants without having to reenter the replacement card information, avoiding any possible declined transactions.
“Tokens are a huge advance in that they’re no longer tied to the plastic. They’re now a digitally enabled capability that can be deployed on a one-to-one basis, on a one-to-many basis, or any other way necessary,” noted Sloane.
How are PayPal’s machine learning models predictive?
Consumers like to be reassured that their data is protected, but it can be frustrating for the valid card holder to have their transaction declined for security reasons. This can happen for a number of reasons such as:
- The vendor suspects fraudulent activity because it’s outside of the card holder’s normal purchase pattern, or
- The vendor had a systems outage
PayPal is able to prevent this in many cases because it has at least 50 petabytes of data collected on online transactions, which it uses for pattern recognition. But how does that pattern recognition technology assist ML models in predicting whether a transaction is legitimate or fraudulent?
Well, “if you had never gone to Montana, or you had never made a purchase from your phone, or the same phone that you’re making that particular purchase, we’d have suspicion of you not necessarily making that transaction,” said Magats. But within PayPal’s ecosystem, it can see all of the behind-the-scenes transactions that may point toward a different conclusion. For example, the customer just collected money on Venmo from six different friends, which adds up to the amount of the large purchase.
“It’s not the obvious that we often are looking at,” remarked Magats. “It’s the less obvious that we look for correlative type of behavior that then triggers data for us to say, ‘that’s a very legitimate transaction.’”
How does PayPal ‘stand in’ for a purchase when merchants face technical issues?
System outages don’t happen often, but when they do it is a costly occurrence for the merchant. This is usually an issue with the external party, such as loss of internet connectivity, which prevents any transactions from going through until the server is back online.
PayPal has the ability to recognize the buyer through its data and verify their identity. After the buyer is verified, PayPal essentially extends a line of credit to the card holder, having high reason to believe that even if the consumer does not pay for the purchase immediately, they are likely to pay for it in the near future.
“So effectively, what we do during the outage is say you’ve got it, the payment has gone through. And we collect and do the transaction processing when the systems come back up and are available,” explained Magats. From a merchant’s point-of-view, they can find comfort in the fact that, regardless of whether there is a problem on their end or within the payments ecosystem, they are not going to miss a sale and the customers will not be dissatisfied.
“It will all be taken care of because we’re standing in for them,” assured Magats.
What is a ‘two-sided network’ and how is that beneficial for PayPal’s data science?
Not every transaction stems from a legitimate buyer, so it’s important to strike a balance between increasing auth rates and minimizing fraudulent activity. “One of the things that makes [PayPal] quite unique,” said Magats, “is that we have a community of consumers, or payers, and a community of merchants and businesses that are basically payees.” Because PayPal has a relationship with both sides, it is able to connect them and complete both sides of the transaction.
Traditional ecosystem players typically only have access and visibility into one side of the transaction, not both. “The ability to effectively adjudicate and make decisions based upon that richness of data, and those [two-sided] relationships, are things that we feel are really differentiating for us and allow us to create great offerings for our customers,” continued Magats.
PayPal’s two-sided network also gives its ML technology an advantage, in that it provides a larger data set for it to learn from. With over 320 million consumers’ accounts and 28 million merchants accounts, PayPal suffers no deficiency of data on consumers and risk profiles. With that insight, PayPal has a better idea of what is and is not a fraudulent transaction, even when faced with the most sophisticated fraudulent behavior.
Why is it important to have a ‘retry strategy’ and how does PayPal help?
A ‘retry strategy’ is exactly what it sounds like. It is a process by which PayPal tries alternatives ways to process an initially declined transaction. “[PayPal has] created almost customized routing logic that works for us and our customers, under the auspices of we want to make sure that every good actor gets their transactions approved,” said Magats. While PayPal strives for 100% success in their transaction rates, it understands that there are bad actors to look out for.
ML algorithms help to identify the best retry strategy based on the card used, issuer, merchant, transaction-level parameters, processor, and acquirer combinations, and even the day and time of retry. PayPal can also retry a transaction with a network token or card number based on success patterns identified by those ML models.
There is also the option of leveraging alternative funds that the customer may have available to them. And last but certainly not least, “our retry logic is effectively to say we’re going to retry later, when we know the systems are going to work based upon knowing that [the customer is] a good actor, and approve those transactions now,” said Magats.