Insights from Inside the Machine
Feedzai Research continues to identify fraud patterns unique to digital commerce and banking.
For a human, these insights are interesting. For a machine, they're something even better:
critical decision-points that end in a recommended action.
Because banks today are product-centric, rather than customer-centric, they make decisions in silos. As customers cross multiple channels and payment types, their transactional data ends up in databases that don’t talk to each other.
Meanwhile, valuable external enrichment data goes unused, because many organizations lack the infrastructure to integrate internal and external data sources. For these organizations, big data is not asset. It’s an exposure.
This is a sample of our findings based on 2017 data
Most of fraud occurs within the first 100 hours of account creation.
Trends show that less time elapsed between device boot time and when the transaction occurred is correlated with more fraud.
Time Elapsed in Days
Number of Consecutive Digits
Email addresses of 2-4 consecutive digits have more fraud.
Percentage of Battery Left
79% of internet users download their bank’s mobile apps, which introduces mobile as a channel for fraud.3
Research shows that fraud rates are higher when there’s more battery left in a mobile device.
Fraudsters prefer certain email domains over others.
Email Domain Name
Identifying fraud signals is hard. Connecting these insights to make a decision is even harder. No one signal is enough to say “fraud” or “not fraud.” This page shows a handful of signals, but a normal transaction profile can have thousands.
Humans alone cannot connect these signals together into decisions. That’s the power of machine learning: combining thousands of data points into recommendations for action.
1 Forbes, “What Will We Do When The World’s Data Hits 163 Zettabytes In 2025?,” Apr. 2017.
2 Aite, “Fraud is Now a Competitive Issue” Oct. 2017.
3 eMarketer, “Most People Have a Mobile Banking App, but Do They Use It?” Oct. 2017.