A team from Google’s fraud-detection group has started its own software as a service venture for spotting transaction fraud quickly based on rule sets and that also learns as it goes to improve its hit rate.
Simility examines online transactions to identify indicators of foul play and assigns them risk scores from 0 to 1. Customers can use the information to shut down transactions it deems suspect.
The Simility Fraud Prevention Platform service is available starting next week after a six-month private beta.
The service is aimed at online marketplaces and retailers and on financial institutions and can cut down the number of false positives these businesses now generate using manual analysis of transactions, says Rahul Pangam, the company’s co-founder and CEO.
The platform analyzes the computers from which transactions are coming in order to detect those that are being used by fraudsters to fire off multiple bogus transactions. That function, called Device Recon, can detect factors such as browser and plugins, Flash version, browser language, location and operating system to create a fingerprint of the machine. When the same fingerprint comes up over and over, that increases its risk score, he says.
Customers’ analysts check out the risk scores and can dig into the reasons behind them in order to decide whether to cancel transactions coming from suspect computers, he says. As they gain confidence in the platform they can more quickly pull the trigger on transactions whose scores exceed a certain threshold.
The platform includes a plain-language rule-making utility for customers to set parameters they want used to catch bogus transactions. It also includes a machine learning engine that seeks out correlations among attributes of known fraudulent transactions. These learned indicators are worked into the algorithms that formulate the risk scores, Pangam says.
Sorting through a vast amount of transaction attributes to discover some of these indicators is beyond the capability of human analysts due to the sheer volume of data, he says.
An important part of the platform is that its analysis engine can be tuned to each customer, making it sensitive to attributes unique to that business’s transactions, Pangam says. “That’s much more effective than a single detection model,” he says.
The platform includes a workbench where customers can view scores and also carry out analysis on their own. For example, they can look for patterns based on physical location of where suspect transactions originate to see whether there’s indications of a larger fraud network.
Simility’s three co-founders all worked together at Google as did five other early employees. The company has a mix of operations experts, data scientists, data analysts and engineers to put together the pieces of the platform.
Founded two years ago, the company has $7.2 million in funding from Accel Partners, Trinity Ventures and angel investors.
The company charges .5 cent to 6 cents per transaction depending on the volume of transactions and the sophistication of the analysis performed.