It’s getting increasingly hard to tell real from fake in so many aspects of life today, but at least one sector might be poised for a reversal of that trend, thanks to technology.
The global market for counterfeit products grew by 50% in 2022, and is expected to top $3 trillion this year. A major reason is the explosion of e-commerce, where it’s relatively easy for a purveyor of counterfeit goods to escape detection, simply by shutting down one website and opening another. Product categories plagued by counterfeiting include a wide range of consumer goods, electronics, pharmaceuticals and auto parts.
One of the biggest targets is fashion apparel and accessories. The problem extends from high fashion all the way to street brands, according to Mark Lee, co-founder of MarqVision, a vendor of software incorporating artificial intelligence for brand protection.
MarqVision aims to purge more than 1,500 online marketplaces of counterfeit goods in 115 countries, claiming better than 95% accuracy in detecting fakery. But is that good enough to put the brakes on a market that just continues to grow?
To have a real impact on the situation, Lee says, legitimate brands need to broaden their focus beyond the e-commerce giants like Amazon.com and Walmart to examine smaller and more isolated sites, which are tougher to police. Some, for instance, might be accessible only to those with an Instagram account, selling through livestreaming. Chat rooms are another popular source for counterfeit goods sales.
The problem extends beyond physical goods to include digital assets, copyright infringement and theft of intellectual property. That sector has become especially toxic since the emergence of non-fungible tokens (NFT), lines of code that supposedly validate ownership of images, videos or similar digital content.
AI might be the key to finally reining in the stampede of counterfeit goods. Lee says the technology can scrape information from all of those vulnerable marketplaces, analyzing each and every listing within a matter of seconds. The AI links up with computer vision to compare product images and logos on a given site with the authorized versions. All told, Lee says, there are 14 different ways in which illegal content can be visually detected.
The task is too massive to be carried out by humans, who would need about an hour to write up a report on a single questionable item, Lee says.
An unusually low advertised price for a luxury item is one obvious red flag. Another is the use of language hinting at the counterfeit nature of the product — a piece of phony Tiffany jewelry, for example, might be described as “Tiffany-style,” to keep the seller from violating certain trademark laws.
The system compares product images against an internal database of repeat offenders, who account for about 80% of the fakes found in the marketplaces under scrutiny, Lee says. In that way, it can flush out sellers who keep opening new stores to stay one step ahead of law enforcement. “AI can pattern-match the images,” he says.
Only recently has AI become sophisticated enough to tackle this immense task, Lee says. MarqVision has been offering its own system for about two years, tapping into relatively recent advances in deep learning — drawing on vast volumes of data and accompanying algorithms — and neural network computing models.
The technology is constantly getting better at counterfeit goods detection, Lee says. In its early days, it achieved an accuracy rate of 75% to 80%. But even with current claims of 95% accuracy, there’s room for improvement as the system “learns.”
Some applications of AI, especially in its most advanced state, find it difficult if not impossible to understand why the system reaches a particular conclusion. Lee says MarqVision gets around the problem of the “black box” by building separate models for each product, categorizing all instances of logo and trademark abuse related to that item. In that way, it’s able to determine the reasoning behind an alert. “You can’t just report and say, “I think this is counterfeit, but I can’t tell you why,’” he says.
As the processing costs of AI come down, fraud detection software will become feasible for a wider range of products. But Lee claims the system is already about one-fifth the cost of relying on a human employee to carry out the determination manually.
Fraudsters, too, become more sophisticated in their methods over time, so there’s always the challenge of playing catch-up with the next wave of scamming. “That’s the fun part of the game,” Lee says. “We’re constantly evolving our tactics.”
Timely, incisive articles delivered directly to your inbox.