Refund fraud started rocketing up the fraud-fighting scene about five years ago, driven by a range of socio-economic factors and the growth of ‘Refund-as-a-Service’ fraud, and it seems it never really calmed down.
Now, generative AI’s non-diffusion technology, which can provide extremely high-level image accuracy, has put a new and powerful image alteration tool into the hands of refund fraudsters.
The Breakdown: Receipts
The typical proof that a customer has purchased an item in your physical store is their receipt. Even if your refund process is streamlined and online or via an app, it likely includes the request to take a picture of the receipt for the item for which the refund is requested.
With generative AI, asking for uploaded images of receipts as a means of defense to protect your business from returns fraud is no longer effective. Fraudsters are able to easily attach doctored images of receipts with all the correct details or create new ones from scratch just based on one example, even with a logo if necessary.
A 10-year-old with ordinary digital competence would have no trouble with this. That’s what’s new and concerning about this application. Previously, you might have been able to mock up a receipt in something like PhotoShop or Figma, but that would have taken time, work, and some technical know-how. This takes none of that — all it needs is a request to your friendly GenAI system.
How It Works: “Damaged” Items
The same opportunity to provide proof and added information is present for items that are “damaged” as well — to show that the item is flawed and that the customer is indeed deserving of a refund. But if that item is actually damage-free or not isn’t what matters to fraudsters. With the help of GenAI, fraudsters can easily add a dent, tear, or mark anywhere on the image of the goods, hassle-free, and upload the artificial “proof” to receive their refund or replacement.
You can start with something like this:
And within seconds, get something like this:
Fraudsters (and some abusers) have been playing this kind of trick for years, but now it’s fast, eerily convincing, grossly simple, and available to anyone with an internet connection.
You can show that a laptop has a damaged battery, that cosmetics arrived cracked, or that clothes are torn. You can get images generated as if taken from any angle, with the defect in the same place consistently — the possibilities and details are limitless.
It’s Fraud, Not Abuse
Merchants need to understand that this kind of activity isn’t an acceptable form of minor cheating, even if it’s carried out by customers who also sometimes make legitimate purchases — it’s theft.
This is a conscious decision to defraud a merchant. It requires going to a GenAI tool, making the request from the agent, and then using it. And people are increasingly doing just that:
That’s bad intent, and even though 58% of customers report feeling “guilty” about taking advantage of a retailers’ policies — if they’re doing it, it’s not ok.
In the “Flexible Policies, Risky Business” survey, 68% of consumers said that retailers make it easy to abuse flexible return policies. In other words, it’s up to the retailer to put the guardrails in place. On the other hand, 16% of consumers admit that they’ve completely stopped shopping with a brand that has initiated more strict return policies. If implementing stringent policies forbidding refunds isn’t the answer, what’s a merchant to do?
What You Can Do
When it comes to GenAI, fraud fighters need to invest — or ensure their fraud prevention solutions are investing — in internal tools to identify cheating behavior of this kind. The common GenAI tools have not yet issued their own solutions to make fake identification easy — although many have put blocks in place to prevent user misuse, in practice, these are easy to circumvent.
Manual identification is challenging because these images are now so good that you really can’t tell. The days of 6 fingered hands are in the past. (I know that wasn’t very long ago, but this tech is moving so fast that it’s a generation ago in terms of maturity.)
Forter’s approach is twofold. It combines continued investment in the technical detection side with a fundamental grounding in focusing on the identity and behaviors of the individual customer. As I’ve written before, you need to focus on customer value rather than blanket policies.
Why Focusing on CLTV Is Critical
Does this customer have a history of returns? Do they seem to abuse free shipping policies? How many refunds have they asked for, and what percentage of their overall spend does that represent? In other words, are they the kind of customer you want? If not, you might not want their business.
The GenAI aspect can make this new threat seem intimidating, but remember that the threat itself isn’t new. It’s the scale, precision, and simplicity that GenAI brings to the table that’s fresh. Despite the justifiable concerns this brings, the reality is that an assessment of this kind is valuable for your fraud-fighting department and your company.
If a new GenAI tool that puts more power in the hands of the cheats is the catalyst for kicking off that process, it might even be a positive thing in the long term — though we won’t be saying “thank you” any time soon.