Online stores use AI to catch fake orders and false refund claims

Online stores in India are using artificial intelligence to spot fake orders and false refund claims as the festive shopping rush builds. The hard part is telling a crook from an excited shopper. A pattern that looks suspicious in March can look perfectly normal during a big sale.

What changed

Datum Intelligence expects festive order volumes in India to grow from 760 million in 2025 to 966 million in 2026, a 27.1% increase. Its September report says festive shoppers will rise 14.5%, and each shopper is expected to place 11% more orders. The average order value is projected to slip from ₹1,578 to ₹1,554. So stores must handle more orders while keeping the cost of fraud control in check.

Where fraud shows up

The risks go well beyond stolen card details. The article lists these:

• Suspicious cash-on-delivery bookings, where the buyer pays when the package arrives
• Fake customer identities and abuse of offers
• Account takeovers, where someone breaks into another person's account
• Fraudulent return requests and delivery tampering

TransUnion, a credit and fraud data firm, reports that 7.1% of attempted digital transactions involving consumers in India in 2025 were suspected fraud. The global average was 3.8%. In logistics, the rate was 16.3%, the highest of the sectors it analyzed, and suspected attempts there rose 31% year on year.

Why old checks fall short

Older systems follow fixed rules, such as flagging orders above a set value or repeat purchases from one account. During a festive rush, real customers also place several or high-value orders. A new user placing multiple high-value cash-on-delivery orders could be a red flag, or just normal festive shopping. A failed delivery does not automatically mean fraud either.

How AI helps

Machine learning, a way for software to learn patterns from past data, can score each order for risk. It weighs order history, account details, order value, device information, delivery experience and shipping addresses. Risky orders can then go to a person for manual review. An order that looks fine alone may be flagged when seen next to related orders.

The same approach can work after delivery. By linking order histories, product identifiers, warehouse records and manual review, companies can flag suspicious returns and refund patterns, and repeated use of multiple accounts and devices to exploit offers.

What sellers should know

TransUnion found suspected fraud was highest at login, at 3.9% of transactions. Account creation followed at 3.1%, and financial transactions at 1.2%. That suggests checking payments alone is not enough.

Krishna Gupta, CEO of logistics platform ShipMaxx360, said AI can help detect early risk, but "this relies on integration between different parts of your business." When customer, order, payment and delivery data sit on separate systems, he said, fraud may be found only later.