RS Intelliedge Blog

In-memory profiling at scale

Most rule-based fraud and risk management solutions use certain data elements from digital payment transactions and statistical data such as, mean, median, standard deviation, frequency, count, etc. of the transactions from the same card or account to construct the rules.

Hence, one set of data elements pertain to “this” transaction, and the other set is a summarized view of distant past transactions.

However, with real-time payments, where settlement to beneficiary happens in real-time, a third type of data element becomes essential – the transactions done in the immediate past few minutes / hours. Thus, inter-transaction pattern needs to be enabled for inspection and screening by the rules. Adding to this, as the action has to be taken in real-time, we do not have the luxury of latency to dig into some datastore maintained in secondary storage to select the past transactions and match the inter-transactional pattern.

The challenge is accentuated as there are a few million concurrent users and hence if the system needs to maintain the transactions in-memory it would need an enormous amount of RAM.

To tackle this, we need a smart way to maintain an optimal number of transactions in the RAM and then apply a way to remove the transactions as the time window slides ahead.

A sliding window to support this capability of referring to transactions done in the immediate past is needed over and above the transaction-based filters and statistical data-based filters. RS IntelliEdge™ implements this feature using open-source technologies and provides the power to use all these capabilities in the hands of risk analysts to build rules using a user friendly rule-editor as well as a sandbox capability to dry run the rules before finalizing the same.

Read more to understand –


Leave a Reply

Your email address will not be published. Required fields are marked *

Comment moderation is enabled. Your comment may take some time to appear.