Rules vs. AI - how best to stop fraud?

“What’s better for fighting fraud: rules or AI?”

Let’s face it, this theological question doesn’t make a lot of sense. It’s like asking a woodworker what’s a better tool: a hammer or a saw? Recently it seems like everyone wants to jump on the AI hype train and so look for AI-powered solutions to fight fraud. Often, though, what I see is that uninformed buyers tend to miss two important aspects:

AI is not black magic: It will not miraculously solve fraud with a key-turn solution (yes, even if the vendor promises that). Deploying an AI solution is like growing plants - they need to be tended daily by professionals that can care for them.

AI is a specific tool, designed to achieve a specific result: Using the example above, sometimes a woodworker just needs to drive a nail in a wall and it doesn’t matter if they have the best saw (Japanese, obviously).

Simply put, these folks don’t realize they are trying to run before they can walk.

When do you want to shop for AI fraud solutions?

• You already have a basic fraud prevention setup: your data is in order, your team is operational, processes are in place and the standard tools are there (e.g. rule engine, monitoring dashboard, analytical sandbox, etc.). Without having this bare-minimum setup, you’ll likely won’t be able to utilize AI at all.

• You have a complex decision matrix: while rules are binary, AI scores enable more nuanced decisioning. Using different score bands can allow you to choose between multiple options: block, insert assets to lists, send to manual review, invoke SCA flows, flag as suspicious, etc. The more actions are at your disposal, the more it makes sense to employ AI.

• Fraud is a constant battle: if you experience a fraud spike every few months, it’s very unlikely that AI is the answer. Sporadic spikes require the ability to react and deploy changes fast, which is what rules are good for. But if fraud is a constant battle and you need to deal with multiple fraud rings, on multiple fronts and consistently - AI can do a lot to alleviate the pressure from your team.

Are there exceptions to the above?

Operating and tweaking AI solutions can also be offered as a managed service by a vendor. While this can potentially remove the burden from your team, it’s worth asking some questions on what it actually entails: how frequently would the model be refreshed? Is there a dedicated person to monitor and tweak my score strategy? What KPIs are used to optimize the performance? 

At the same time, it’s important to make sure the vendor has a high product-market fit with your business, vertical and region to make sure they really have the experience with servicing similar clients with such a delicate task.