How VIEWAPP protects delivery services from AI photo fraud

Just a few years ago, return fraud looked simple: a customer would send the store a photo of a broken item or spoiled product, and if the story sounded plausible, the money would be refunded. Generative AI changed the rules of the game. Now a photograph of "moldy grapes" or a "smashed package" may not be evidence, but a product of a neural network — created in a few seconds and indistinguishable from a real shot.

This is no longer an isolated case, but an industry trend. Since the beginning of 2026, industry publications have been writing about it systematically: retailers are facing a new, more sophisticated form of return fraud, in which buyers send AI-generated images of damaged or defective goods in order to get money for products that actually arrived in perfect condition — and this practice is spreading fast enough to provoke a broad reaction from retailers and their logistics partners [1]. Analysts record a 175% increase in the number of faked damage photos since the beginning of 2025, while complaints of the "item does not match the description" format grew by almost 50% in just the first months of 2026 [2].

The scheme works at scale precisely because it is cheap: a fraudster can generate several versions of the "evidence" in minutes, adjust the text of the complaint, and repeat the process across many accounts with different sellers [1]. In one documented case, the company discovered AI artifacts in a customer's photograph right during the dispute review — the item, as it turned out, had arrived intact [3].

The damage in numbers: from the US to Europe

The scale of the problem is already measured in concrete sums, and they are growing worldwide:

•            In the US, the share of fraudulent and abusive returns in retail reached 15.14% in 2024 — this brought retailers a loss of 103 billion dollars, versus 101 billion a year earlier [4].

•            In global e-commerce, the damage from fraud as a whole is estimated at 56 billion dollars in 2025, with projected growth of 133% — to 131 billion dollars by 2030 [5].

•            In France in 2024, two people were arrested who defrauded a food delivery service of almost 2 million euros over two years — through a series of false claims [6].

•            In Spain, the damage from returns of damaged goods alone exceeded 1 billion euros per year, and 88% of online stores acknowledge cases of damage [7].

Food delivery — an especially vulnerable segment

If in e-commerce as a whole fraudulent and abusive returns already account for around 15% of all returns [4], then in the food delivery and quick commerce segment the situation is even more acute. According to industry data, return and promo fraud together make up 48% of all consumer fraud on delivery platforms — that is, almost half [8]. The classic scenario: a customer who actually received the order claims that the item disappeared, spoiled, or arrived wrong — and demands a refund or reshipment [9].

The problem lies in the very nature of the category. Perishable goods and food are unprofitable and impractical to take back and verify after the fact — unlike clothing or electronics. The refund decision is made almost exclusively on the basis of a photograph sent by the customer himself. This is exactly the point where generative AI breaks the basic assumption on which the entire system of fast returns rests: that the photo sent actually reflects the condition of the product.

Similar dynamics — in Spain and other markets

The Spanish market gives an idea of the scale of the related problem. By the end of 2025, one fifth of online buyers in Spain had received at least one damaged item [7]. At the same time, the country is recording an explosive growth in online fraud in general: the number of cyberattacks related to fake purchases more than doubled in the last quarter of 2025, over 45 million attacks on fake online purchases were blocked, and fraud in purchases remains the largest category of cybercrime in the country [10].

This creates a breeding ground specifically for the type of schemes in question: the higher the volume of legitimate returns and the overall level of digital fraud in the market, the easier it is to hide fake claims among real ones.

But the situation also works in the opposite direction

This problem has another side. If a company cannot reliably verify a photograph, it risks making a mistake not only in favor of the fraudster. Because of distrust of evidence, genuine returns may be rejected — even when the item really is damaged or spoiled. For a delivery service, this means no longer a direct loss from a return, but reputational losses. A customer who really received a defective item but could not get compensation is far less likely to use the service again and to recommend it to others.

The result is the same problem from two sides: the company either risks paying for a fictitious return, or does not pay for a genuine one and loses customer trust. The solution is not simply to tighten return rules, but to obtain reliable confirmation of the condition of the product at the moment of the claim.

The technological answer: verification at the moment of capture

VIEWAPP founder Alexander Fokin believes that the problem requires a solution not at the level of after-the-fact moderation, but at the level of the very moment of capture:

"We could integrate our VIEWAPP technology into this process: the user would photograph the return directly from the delivery service's app, and this would exclude fakes — with geolocation and the rest of the authenticity verification."

The idea is to remove the very possibility of sending an arbitrary file. If the photograph is taken not from the gallery, but directly at the moment of filing the return — through the camera inside the delivery service's app — with geolocation and technical authenticity marks attached, the fraudster has no window left to substitute the frame with a neural network. Verification happens not after the money has already been charged to the retailer, but before the refund decision is made at all.

For platforms like Glovo, where order counts run into the millions and the margin in food delivery is already thin, closing even one category of fraud — fake photo evidence — could directly affect the economics of returns.

Sourses:

1. PYMNTS, «AI-Generated Damage Claims Trigger Retail Crackdown on Return Fraud» — https://www.pymnts.com/news/retail/2026/ai-generated-damage-claims-trigger-retail-crackdown-on-return-fraud/
2. Signifyd, «2026 State of Fraud Report» — https://www.signifyd.com/ecommerce-fraud-trends/
3. Alians Software, «Fraud vs Friction: AI Return-Abuse Detection» — https://aliansoftware.com/en/blog/fraud-vs-friction-ai-return-abuse-detection
4. Appriss Retail & Deloitte, «Consumer Returns in the Retail Industry», через termsandconditionstemplate.com — https://termsandconditionstemplate.com/return-fraud-statistics-2026-costs-rates-and-methods
5. Juniper Research, данные через ecosistemastartup.com — https://ecosistemastartup.com/compras-online-seguras-2026-131-000m-en-fraude-y-como-proteger-tu-ecommerce/
6. Deonde, «How to Reduce Failed Deliveries and Refund Requests in Your Food App» — https://deonde.co/blog/reduce-failed-deliveries-refund-requests-food-app/
7. El Publicista / DS Smith, «El e-commerce en España cierra 2025 con 15,2 millones de devoluciones» — https://www.elpublicista.es/mundo-online/commerce-espana-cierra-2025-15-2-millones-devoluciones-productos
8. QSR Magazine, «How Restaurants Can Tackle Delivery App Refund Abuse» — https://www.qsrmagazine.com/story/how-restaurants-can-tackle-delivery-app-refund-abuse-without-losing-customer-trust/
9. SensFRX, «Food Delivery Fraud: Common Scams & How to Protect» — https://blog.sensfrx.ai/food-delivery-fraud/
10. Digital Perito, «Estafas online +125% en España» — https://digitalperito.es/blog/estafas-online-espana-125-por-ciento-aumento-ciberfraude-2026/