The advent of advanced machine learning and AI powered fraud prevention solutions has allowed merchants to tackle rising fraud losses and the costs of preventing it more efficiently. Apart from cutting down on time and resources spent on manual reviews, AI and machine learning based solutions have allowed for fraud and risk decisioning to be done at scale and with greater precision.
While these benefits are significant, it is important to note that not all AI powered fraud prevention solutions are equal. Many providers, characterized by an older tech stack, depend on Blackbox AI models which are primarily designed to work in a ‘set and forget’ manner, working as opaque engines providing little to no insight into why a decision was made.
Why was the risk score for a transaction high? Why was a transaction flagged? What rule changes should be made to adjust performance? Blackbox AI makes answering such questions extremely difficult, leaving fraud and risk teams at the mercy of a machine. The lack of insight into these aspects can then translate into increasing false positives for the merchant, thus costing them over 75 times more than the fraud itself.
Placing power in the hands of relevant users: Enter Whitebox AI
To counter the drawbacks of incumbent and opaque systems, Whitebox AI focuses on two key elements, namely interpretability and transparency. This provides risk and fraud analysts more accurate insights in a simple human readable language along with visuals. More specifically, it sheds light on the customer and payment level attributes contributing to the final risk score.
Gained by enriching raw data points like name, payment information, IP and email address, these attributes reveal key details such as, an email being bogus, risky IP locations, a VPN/TOR being used etc. The fraud analysts can then leverage these insights to accurately tweak and customize rules. In the absence of such valuable information, teams would find it extremely hard to cut down on fraud and false positives.
Let’s take a practical example to see the advantages of Whitebox AI:
Barbara is traveling for a workcation from Germany to France. On her train ride, she realizes that her contact lenses are missing. She then visits her preferred online seller, logs into her account, and orders a new pair to her hotel in France. She also adds a lens cleaning solution, which she orders regularly from the website. However, the transaction is rejected by the merchant . Having regularly ordered the same items in the past, she is left confused.
Barbara’s transaction in the case above is declined despite her being a regular customer and the order being the same as her previous orders with the online seller. Given that there is no spike in the overall transaction value, and she is a known customer, her transaction should have been considered to be safe, However it’s declined by the merchant’s fraud prevention solution due to the IP address of the order not matching Barbara’s previous purchase addresses and an additional mismatch between the shipping and billing address. Such false positives lead to a loss of revenue for the merchant and an overall bad customer experience.
In such a situation, a fraud prevention solution with Blackbox AI, would give little to no insight into why this decision was taken and directly reject the transaction due to a high-risk score. This would impact multiple other transactions made in similar conditions, thus leading to high false positives.
In contrast, Whitebox AI powered solutions give merchants detailed insights in real time, which can be leveraged to identify reasons for the incorrect decline and fine tune rules for greater accuracy. A fraud analyst can then leverage such insights and accordingly add an exception to the decline rule, making sure recurring customers that use a previously approved email or credit card, are accepted.
Blackbox vs Whitebox AI
Blackbox AI:- Model ingests data -> Machine provides risk score -> Risk & Fraud Analysts make decisions without relevant insights into how a decision was made.
Whitebox AI:- Model ingests data -> Machine provides risk score along with verbal and visual explanations on how this risk score was reached -> Risk & Fraud Analysts make decisions based on explanation and score.
Push for overall AI transparency:
Recently, there has been an increased push to mitigate the undesirable outcomes and risks arising from AI decisions with no explainability. An EU draft legislation proposes that AI systems will need to meet specific “transparency obligations” that allow humans to review decisions made by an AI as well as to establish how that decision was reached and what data points were used in the process.
Trust being one of the core virtues of a customer experience, Whitebox AI helps ensure that risk and fraud decisioning is free of any bias against a person’s ethnicity, nationality, or sexual identity, amongst other factors.
In line with this, Fraugster built its AI engine with Whitebox philosophy at its core. Leveraging advanced data enrichment and network intelligence, Fraugster analyzes more than 2500 attributes and behavioral identifiers including:
Ultimately, AI is a tool used by humans for humans. While not giving a 100% transparency, given the constantly evolving nature of AI, Whitebox AI enables humans to understand and interpret the AI’s decision by delivering insights into key factors contributing to the overall decision.
This article was contributed by Fraugster