Case Study: Reduced False Declines and Increased Monthly Revenue Through Ekata Risk Scores

While the approach to capturing fraud and false declines varies from company to company and from industry to industry, one thing is certain: implementing an end-to-end identity verification solution into your risk assessment model is essential. In this case study, Ekata performed a historical analysis on previous transactions accepted and declined by a client prior to them implementing Ekata. Before we get into the breakdown of outcomes, let’s first present an overview of what Ekata’s data does.

 

Identity Risk Score and Identity Network Score: Improving Risk Analysis for Digital Interactions

By leveraging the power of the Ekata Identity Network, a proprietary global dataset of billions of customer transactions, Ekata enables our clients to reduce false declines and increase the precision of fraud detection via highly predictive signals that are designed to help differentiate good customers from bad customers.

Ekata has released several features to help businesses maximize predictability in finding fraud. Two such features include:

  • Identity Risk Score: Combines authoritative data (metadata, match statuses, linkages) from the Ekata Identity Graph with usage patterns of identity elements seen in the Identity Network to derive a comprehensive risk score. Identity Risk Score predicts whether the customer is who they say they are, as well as if their identity data is used legitimately.
  • Identity Network Score: Assesses the risk of a digital interaction using patterns and features selected based on predictive power in the Identity Network. Complex machine learning algorithms derive a score from activity patterns. Network Score boosts fraud detection efforts for real-time decision making and verifies genuine customers to help avoid false declines.

The two features independently predict a risk level based on the input from the transactions. The Identity Network Score spans a spectrum from 0 to 1 where 1 denotes high risk. Similarly, the Identity Risk Score spans a spectrum from 0 to 500 where 500 denotes high risk. 

Combining Risk Scores

With the two Ekata Network features, each customer’s predicted risk levels could be plotted onto a graph. For example, a customer with an Identity Network Score of 0.975 and an Identity Risk Score of 475 is deemed to be high risk. 

Plotting the two scores for each customer within a transaction period can help give us a visual profile of different categories; the green graph represents transactions that are “true negative” – these transactions were predicted not to be fraud and they were, indeed, not fraud. Meanwhile, the red graph represents transactions that are “false negative” – these transactions were predicted not to be fraud but they turned out to be fraud.

Customers that are verified to be good have a greater concentration around an Identity Network Scores of less than 0.6 coupled with Identity Risk Scores of less than 200.  

Customers that are verified to be fraudsters have a greater concentration around an Identity Network Scores of more than 0.8 coupled with Identity Risk Scores of more than 300.  

The combination of Identity Risk Score and Identity Network Score paints a very different picture for good transactions than for bad transactions. Using this information and recognizing (and accepting) that some rejected transactions are good customers, one can make an educated guess regarding which of those rejected transactions were good customers. With this set of results, one could deduce that those rejected transactions in the upper right quadrant are likely to be fraudsters; a good indicator that fraudulent transactions are captured properly. Those that fall within the upper left quadrant cannot be determined, signaling to our client that more due diligence is required to assess the level of risk. Finally, those that fall within the bottom left quadrant are likely to be false positives. 

The illustration shown to the left depicts an Ekata customer test scenario using the methodology of evaluating risk scores to identify false declines. In analyzing the rejected transactions and focusing on the “likely good” transactions, our customer was able to deduce that the bottom left sector of rejected customers accounted for approximately 20% of all rejected transactions, which equates to a monthly revenue of up to $500,000. 

These studies give confidence that the two risk scores play a key role in identifying the uncertainty of rejected transactions. By eliminating the falsely rejected transactions, good customers can encounter less friction and a more positive experience.

 

In Practice: Outcomes of the Historical Analysis

Ekata performed the following exercise on a new client to better understand their distribution of transactions, reduce the number of falsely rejected customers, and drastically increase their revenue, all without impacting chargebacks. 

For this particular Ekata client, the chargeback rate was relatively low, but they were rejecting a large number of transactions. Ekata developed a hypothesis: If a low-risk population can be identified among rejected transactions, then the client could reduce the number of falsely rejected customers, increasing revenue without a significant impact to chargebacks. 

Our historical analysis, this time using Ekata’s API, looked at verified good transactions (green graph), verified fraud transactions (red graph), and rejected transactions (blue graph). We also captured those who are falsely rejected by tracking the distribution of declines that were reversed via call center monitoring. From this data, we can see that the majority of false declines had low Identity Risk Scores and Network Scores. While it is not imperative to track false declines separately to assess the impact, our client was able to make a more informed decision on their risk assessment model.

Following this analysis, our client went on to pick a subset of rejected transactions from their production environment at random and put them in front of manual review agents to see which transactions should have been accepted. Even though the manual review agents were not privy to Ekata’s risk scores, their decisions of falsely declined transactions correlated with the predictive signals.

Finally, we conducted an A/B test. As our client was now ready to take on more risk, they decided to allow those transactions that have an Identity Risk Score < 300 and Identity Network Score < 0.8 to be accepted. To manage that risk, they started accepting transactions incrementally. As a start, only 25% of the transactions that met the Ekata criteria were allowed through. The original plan was to increase the percentage of transactions accepted gradually, but after seeing the results from the 25% test, our client decided to ramp up to 100%. 

Through the observations from implementing the Ekata solution into their risk assessment model, our client saw a monthly increase in revenue as a result of the additional acceptances that were falsely declined previously. 

 

Summary

Gaining market share is not an easy task, but it starts with providing an exceptional customer experience. To be the leader of the pack, it is necessary to consider the customer experience every step of the way. 

Implementing an intelligent layer of fraud detection proactively not only allows the merchant to capture fraudulent transactions earlier, enabling good transactions to be processed, it also reduces false declines. 

Ekata helps our clients – and can help you – reduce false declines and, in doing so, increase revenue. Give us a call today.

Start a Free Trial

See how Ekata can reduce fraud risk for your business, contact us for a Demo.