Popular searches
Not sure where to start?Talk to our team about running sovereign AI in your environment.
↑ ↓ Navigate↵ OpenEsc Close
At PayPal's scale, trust is the product. Its data science teams use H2O to stop fraud and churn in real time, before the loss lands.
PayPal's purchase-protection guarantee reimburses buyers who never receive what they paid for, so every missed fraud is a direct loss. The team has to catch it in real time, before the payment transfers, against fraudsters who keep changing their methods.
Using H2O Driverless AI, the PayPal team found significant new modelling features that raised model accuracy by almost 6% in a single test. For a team with over 10 years of feature-engineering experience on the fraud problem, that was a striking result.
To get ahead of the fraud rings, PayPal looked beyond individual buyer and seller behavior to the signals of a connected network: shared assets, shared IP addresses, shared shipping addresses.
The team implemented a graph database and used node2vec to learn continuous feature representations for nodes in networks. Once a bad account was identified from payment transaction data, other bad accounts sharing the same network structure could be located.
PayPal combined feature representations from the graph network with their expert-engineered features, then applied H2O Driverless AI to the merged feature set. Driverless AI automatically engineered additional features and models, greatly improving performance across a graph of roughly 1.5 billion edges and half a million nodes.
The top 5 features extracted by H2O Driverless AI bested 10 years' worth of expert-engineered features. Model accuracy increased from 0.89 to 0.947, and running on an IBM Power GPU-based server let the team train the model six times faster than in a CPU environment.
“H2O Driverless AI gives amazing performance in terms of feature performance and also model performance.”Venkatesh RamanathanSenior Data Scientist, PayPal
Retaining a customer is worth as much as winning one. PayPal wanted to flag likely churn early enough to act, with real-time predictions instead of after-the-fact reporting.
“I trained and scored on the entire data set, not a sample, and it was orders of magnitude faster than other tools. Kudos to H2O.”Julian BharadwajData Scientist, PayPal
PayPal has been revolutionizing commerce globally for more than 25 years. Creating innovative experiences that make moving money, selling, and shopping simple, personalized, and secure, PayPal empowers consumers and businesses in approximately 200 markets to join and thrive in the global economy. For more information, visit https://www.paypal.com, https://about.pypl.com, and https://investor.pypl.com.