Transitioning from legacy fraud solutions like Kount may be easier than you think.
Fraud solution offerings have come a long way. Decision accuracy took priority in the mid-2000s and early 2010s, giving any provider that offered flexible custom rule builders an edge over the competition. Companies like Kount did a good job meeting market expectations at the time—but things started to shift when Sift entered the fraud prevention market with advanced, authentic machine learning (ML).
Machine learning was already being used to fight fraud in financial services, albeit using less sophisticated models and data analysis. But it hadn’t been deployed at scale, or applied across different types of e-commerce businesses. In 2011, Sift entered the market and changed the game, bringing sophisticated ML capabilities to the trust and safety sector that could be effectively used across multiple industries and business types, even as they grew in size and expanded into new regions.
A losing legacy: Rules-based fraud solutions
Over a decade later, many rules-first fraud solution providers are still scrambling to catch up by deploying machine learning models that overlap with rules, or that offer less clear decision explainability—in other words, it’s not easy to understand what specifically contributed to the recommendation provided by the solution.
In addition to decision engine challenges, legacy solutions have also fallen behind when it comes to user experience within case management tools. The vast amount of data generated in each customer interaction can be overwhelming and challenging to present meaningfully to analysts striving to uncover the story behind a transaction—a necessity for them to accurately assess it.
Legacy solutions like Kount tend to display data in a way that requires fraud analysts to go on a digital scavenger hunt, clicking around to look at different areas of the tool, and then comparing what they see in order to come to a conclusion. In contrast, Sift guides analysts to take specific actions by summarizing and displaying relevant signals that contributed to a Sift score. This reduces the time spent on each investigation, resulting in significant productivity gains.
Kount claims that they combine supervised and unsupervised machine learning insights into a single score that’s then delivered to risk analysts. However, Kount’s transaction details page (where fraud analysts perform reviews) displays the following:
- Safety Rating: A score between 1 and 99 that represents the “safety” of the transaction. The lower the score, the riskier the transaction.
- Safety Grade: This is derived from their omniscore and translates the score into a grade that uses the well known academic grading of letters A through F. A being a passing grade and a low risk score, and F being a failing grade and a high risk score. What makes this challenging is the use of both a score and a letter grade. This takes up real estate on the UI, and leaves a lot open to interpretation (and misinterpretation). Another risk here is that a grade of F in academia ranges from 0 to 59, while the rest of the letters reside in a range of 10. Is an F that was scored 59 really the same as an F that was scored a 3?
- Persona Score: A score that ranges from 0 to 99, where 99 is the most risky and 0 is the least risky. Adding even more confusion, this scale uses the inverse of the safety rating. Imagine being a new fraud analyst and Kount user, and having to remember which score is riskier the higher it gets, and which is riskier the lower it gets. Just having to pause to remember is adding precious time to each review. In a worst case scenario, the analyst confuses the two and applies the wrong decision as they intended.
- Velocity: Kount describes velocity as “The total number of previous transactions linking in the persona (over the last 14 days).” The problem here is they define a persona as “being derived from over 200 data elements”. And of those elements, many can change without changing the persona. This makes it very unclear what this velocity value even represents.
- VMAX: This is described as “The maximum number of previous transactions within any rolling six hour period over the last 14 days.”
Although more decision outputs may seem like an advantage, when they are cryptic and conflict with each other, they quickly become a disadvantage.
The Sift advantage
In contrast, Sift displays deeper insights that don’t require analysts to consistently reference product specs to decipher. Sift offers a comprehensive suite of fraud prevention tools, including step-up authentication, chargeback prevention, and account protection. Businesses that migrate to Sift can tap into all of these tools from a single, intuitive Console and apply them across the user journey. By simplifying the tech stack, and aligning and integrating fraud prevention efforts from account creation to dispute, digital risk teams get complete transparency into data and decisioning.
We differentiate ourselves in both backend decisioning technology and front-end user experience in a number of ways:
- Advanced rule builder that has the ability to leverage thousands of signals.
- Clear decision explainability that articulates why a transaction or event was scored as risky or not risky and guides the analyst where to focus their investigation.
- Advanced case management that accelerates productivity with capabilities like visual link analysis that supports bulk decisioning directly from the visualization screen.
- Purpose-built solutions for non-payment fraud types such as account takeover, fake signups, and spam/scams.
Moving from a collection of point solutions to a single platform also drives efficiency and accuracy—data from various sources can be seamlessly integrated and analyzed to identify complex fraud patterns that may have otherwise gone unnoticed by merchants using disparate systems.