Mobile wallet providers strive to let consumers pay quickly and easily, but users are often forced to manually authenticate their purchases. How can payment providers balance security with usability?
Paygilant analyzes each user’s spending habits and stores a “map” of their behavior on their smartphone. The authentication process is only triggered when the map detects an unusual transaction.
Enablesthe vast majority of transactions to be completed without manual authentication
3 timesmore fraud attempts can be prevented, compared to traditional detection methods
Up to 60%cut in operational costs for issuers’ fraud detection departments
Business challenge story
The mobile wallet paradox
Usage of mobile wallet apps is booming. Two years ago, wallet transactions amounted to USD 300 billion—over the next three years they are expected to reach USD 1.5 trillion. This growth has attracted considerable attention from banks and big-name businesses, yet even with all the mobile wallet apps available, one factor continues to undermine adoption and end-user satisfaction.
Ziv Cohen, CEO at Paygilant, explains: “The premise behind mobile wallets is to streamline the payment process for shoppers—but that’s not what was actually happening. As the risk and cost of fraud are so high, payment providers have to put the burden of anti-fraud checking on end-users by forcing them to provide their credentials for almost every transaction. But manual authentication takes time, and for the user it can cause considerable friction—which defeats the purpose of using a mobile wallet in the first place.
“Simplistic approaches to authentication also make mobile wallets more confusing for users and more risky for providers. For example, in many countries, large transactions require authentication every time, while small transactions get approved automatically with no checks at all. Fraudsters who know the rules can therefore game the system, making a number of small payments to steal significant sums of money with very little risk of being caught.”
Paygilant realized that the traditional back-end fraud detection systems used by banks and other payment providers were the root of the problem due to their lack of broad visibility. In addition, these systems rely on operation teams to manually process alerts, and call centers to contact customers when suspicious activity is detected. As a result, operating costs for most providers were high.
Paygilant also believes that due to their limitations, these traditional systems and processes only detect around 30 percent of fraud attempts on average. The behavioral models they use are imprecise and rarely updated, and often do not cover all fraud use cases.
Cohen continues: “As more and more people start using mobile wallet apps, there will be more transactions and more fraud attempts, requiring even more resources to manage at the back-end. As it stands, providers have little choice but to resort to manual end-user authentication as a way to counter fraud. But while having to authenticate for every transaction does mitigate some of the fraud risk, it is detrimental to the customer experience. On the other hand, the alternative option—to have no authentication at all for low-value transactions—makes life easier for consumers, but it also plays into the hands of fraudsters.
“We knew that there had to be a better way to balance security, usability and costs to make mobile wallet apps an attractive option for both end-users and payment providers.”
Revolutionizing mobile wallet fraud detectionPaygilant devised a revolutionary solution to the mobile wallet challenge: deploying fraud detection systems directly on the user’s mobile device.
The Paygilant solution uses big data technology to generate behavioral “maps” for each individual customer. The maps are then downloaded and stored on the customers’ smartphones and used to assess whether each new transaction fits with the customer’s usual purchasing behavior. The system only triggers an authentication request when it detects unusual or suspicious activity. The process is instant and takes place on the device and at the point-of-sale—without any need to send or receive data from the provider’s back-office system, and without any manual intervention.
“We had a vision that would completely transform the mobile wallet space,” says Cohen. “The first step to making that vision a reality was to find a partner to help us design and build our solution architecture. We spent some time evaluating the options, but it wasn’t long before we chose IBM.”
He continues: “IBM’s philosophy and technology perspective really resonated with us and they have a great deal of experience in the mobile payment arena. What’s more, we were very keen to get involved with IBM’s Alpha Zone program.”
The IBM Alpha Zone Accelerator is an intensive 20-week deep immersion program that helps startup businesses develop solutions for the enterprise market, with on-site support, mentoring, and technical training.
“We got immense value from the program,” remarks Cohen. “Our engineers worked closely with the Alpha Zone team from the outset to refine our initial plans and design our architecture in such a way that it would be highly scalable and supported by various IBM products and services.”
At the heart of the Paygilant solution is a patented multidimensional database that captures customer information and historical transaction data. This data is then processed using a unique algorithm that builds detailed multidimensional maps of users’ purchasing behavior.
To fit onto a smartphone, these complex multidimensional maps are then condensed into multiple smaller maps that focus on the specific areas where the customer has the most transactions. Using these personal maps, Paygilant can predict with a high degree of certainty whether any given transaction is genuine, and can therefore automatically approve the non-suspicious transactions without requesting manual input from either the payment provider or the end-user.
Each new transaction feeds back into the algorithm, ensuring that the model is constantly kept up to date, and a new map is generated for each user at least once every 72 hours.
“We believe that the accuracy of our system is leagues ahead of any other solution on the market,” explains Cohen. “Traditional fraud detection models might only be updated every few months, whereas ours is automatically updated every few days.
“The fact that our solution is located on the mobile devices themselves also yields unprecedented advantages from a security and privacy perspective. Ordinarily, a mobile wallet app would not be able to utilize location data or call history for fraud detection, because capturing that kind of personal data and transmitting it to a back-end system would raise serious privacy issues in many countries.
“But since all of the decision-making for Paygilant takes place on the smartphone itself, personal data never has to leave the device. So we can use it to improve the accuracy of fraud detection without compromising on privacy.
“What’s more, the local deployment enables the solution to work entirely offline, so it doesn’t matter if the transaction takes place somewhere where there’s no mobile data or wifi connection. This is a big plus for merchants, because they don’t need to worry about patchy network coverage in their stores—there’s no risk of losing a sale because the customer’s device can’t connect to the internet for authentication.”
Paygilant’s big data technology is built with IBM® BigInsights® and Apache Spark, and utilizes the IBM Bluemix® application development platform to enable easy integration with mobile wallet apps. The solution uses Spark’s sophisticated machine learning and predictive capabilities to generate its multidimensional maps, which are then stored in an Apache Hadoop cluster.
“We designed Paygilant to work both as a cloud-based and an on-premises solution,” adds Cohen. “The default option runs in the cloud on SoftLayer, providing an easy way for our clients to spin up our detection service and embed it in their apps. But since many organizations—banks in particular—aren’t comfortable sending data outside of their own networks, we also offer an on-premises version that they can run on their own Hadoop cluster in-house—giving them total peace of mind.”
To manage all the requests coming from the mobile wallet to the big data environment, Paygilant uses a mediation layer based on IBM WebSphere® Liberty, which routes the requests to the appropriate location in the multidimensional database to fetch a new map.
Finally, Paygilant is using IBM Watson™ Natural Language Classifier in its new-user registration processes. Cohen explains: “One of the main problems that we’ve seen mobile wallet companies struggle with is registration: how can they tell whether the person registering is really who they say they are.”
He adds: “With Watson, we can build a customer profile based on information from available data sources—such as public Facebook pages and other social media information—and match it against the registration details. When the two profiles don’t align, we get notified instantly, so we can track the user’s transactions to ensure that no fraud is occurring. Watson Natural Language Classifier significantly accelerates the process of verifying users’ identities and increases our accuracy.”
Secure, cost-effective, user-friendlyWith Paygilant, shoppers only need to authenticate a small fraction of their mobile wallet purchases, and fraud detection is increased dramatically. As a result, banks and mobile wallet providers can make considerable savings.
“From the end-user perspective, Paygilant really delivers the shopping experience that mobile wallets originally promised,” says Cohen. “It’s fast, easy and secure, and there’s very little friction between the user’s decision to make a purchase and the completion of the transaction itself. On average, customers only need to authenticate 10 percent of their payments with Paygilant, compared to 100 percent with many other systems.
“When the risk of fraud for a certain purchase is particularly high, Paygilant triggers multiple innocuous authentication methods, all without subjecting the user to phone interrogation from a bank’s back-office call center. And whereas fraud flies under the radar of traditional back-office systems, our locally deployed solution triples the fraud detection rate.”
From a payment provider’s point of view, Paygilant also has the advantage of being completely transparent to the end-user. Cohen comments: “Users interact with their respective mobile wallet apps, not with us. Paygilant works entirely in the background so the shopping experience is fully streamlined.”
Paygilant also offers increased control, significantly reduces the risk of fraud losses, and can yield considerable back-office savings. “We expect that banks will be able to cut operational expenses for fraud detection by up to 60 percent,” says Cohen. “Paygilant takes care of authentication and detection, and eliminates the majority of the call center workload.”
Cohen concludes: “We firmly believe that Paygilant is the answer that the mobile wallet space has been waiting for. It will enable mobile wallet apps to strike the perfect balance between security, usability, and costs—and it wouldn’t be possible without IBM’s technology and expertise.”
Paygilant was established in 2014 with the goal of transforming the mobile wallet space by deploying fraud detection systems on smartphones themselves. Paygilant’s innovative mobile-based fraud detection solution, built on cognitive and cloud data services from IBM, has been recognized by the European Union Commission as a disruptive technology.
- Apache Spark on IBM Open Platform with Apache Hadoop (BigInsights)
- Cloud Services - IBM Cloud infrastructure F2F (Cloud BU)
- WebSphere Application Server
Take the next step
IBM Cloud Data Services offers a complete portfolio of data and analytics services providing unique and seamless product integrations to build apps faster and gain new insights easier with flexible deployment and pricing options. For more information about how IBM Cloud Data Services can help businesses solve tough big data problems rapidly and cost-effectively, please visit ibm.biz/clouddataservices.