ORLANDO – Canada’s largest retailer said it has already saved more than $2 million since going live with a system to monitor for fraudulent returns last year.
Speaking at the Teradata Partners 2006 user conference last month, Hudson’s Bay Co. team lead Shelly Perrior said the system, which is linked to its data warehouse, Web service server and point-of-sale systems, has already helped catch at least one major incidence of fraud.
Also called a returns control system, the tool was designed to allow authorized cash refunds on items that are returned with a receipt, or an HBC gift card.
“We piloted it in four stores, and in one case there was someone who had taken a receipt and photocopied it 100 times,” Perrior said. “They were just going across Canada, essentially, trying to get a refund on the same transaction over and over.”
Before the system was implemented, HBC estimates the individual in question had managed to get away with $26,000 from one purchase.
HBC created a separate database for the returns control application, in part to meet performance requirements, Perrior said. At peak times, between 50 to 100 transactions may be loaded into the database each second. “We’ve found that up to 100 per cent of fraud can happen within the first 30 minutes (of opening a store),” she said, “so we have to get the information into the data warehouse as fast as possible.”
When a customer buys something at HBC, a request is sent in SOAP XML to the Web service server, which is then stored for query. When a customer returns an item, the HBC sales associate can check with the database and receive validation to allow the transaction to continue, or an alert that the return may be fraudulent.
Yves Pouliot, a senior consultant at Teradata who worked with HBC on the project, said the tricky part is setting up a system that will make sense of the “tree structure” that evolves from multiple transactions. In some cases a customer might buy one item and then return it. In other cases, a customer might buy 20 items but only return one or two. This adds complexity to the overall database.
The returns control system holds three days’ worth of data – the same day plus the last two days. According to Pouliot, about 60 per cent of all returned items are processed in that period.
Perrior said the system is put under considerable pressure, but it has been managing the workload well. The average store response time for a validation on returns is usually less than one second, she said.