Authentique Featured in The Business of Fashion
AI Takes on the Fashion Industry’s Counterfeit Problem
LVMH’s Patou has begun rolling out a new AI-powered authentication system with its technology partner ORDRE, adding a new dimension in fashion’s effort to fight fakes using AI’s ability to spot patterns indiscernible to humans.
by MARC BAIN
11 April 2023
KEY INSIGHTS
The new authentication system, created by Ordre, lets customers verify a protected Patou product is genuine with just a phone camera and an app.
As counterfeits continue to plague brands and retailers, more companies are turning to AI in hopes its ability to recognise patterns indiscernible to humans will help them separate even the best reproductions from the genuine article.
While evidence suggests the technology can be effective, experts say the best results come from pairing it with expert human authenticators.Luxury brands and resale businesses are betting on artificial intelligence to solve a centuries-old issue.
The plague of counterfeits shows no signs of diminishing as online marketplaces make knockoffs of everything from sneakers to high-end handbags easier to find. Typically companies have relied on trained authenticators to spot fakes, but more are turning to AI in hopes its ability to recognise patterns indiscernible to humans will separate even the best reproductions from the genuine article.
While these efforts have generally focused on catching counterfeits already in the marketplace, Patou is taking the next step and embedding AI-powered protections into its products from the start.
The LVMH-owned fashion and accessories brand will be the first to use Authentique, a new tool created by Ordre, an online wholesale platform that has steadily branched into other tech offerings. During production, the brand photographs a designated area of its bags to create what is essentially a digital fingerprint. Using the Authentique app, a customer can later snap a photo of the same location on the bag and verify whether it’s real, according to Simon Lock, Ordre’s chief executive.
Patou will roll it out first in its limited-edition Le Petit Patou bags in collaboration with DJ and stylist Sita Abellán. But Sophie Brocart, Patou’s CEO, said the next step is to expand it to all their limited releases. LVMH will be watching the results to see if it might be useful for other brands in the group, she added. Lock noted Authentique could scale to widely released products as well, as long as it has a large enough dataset to train its algorithm.
“It’s super simple — as easy as taking a picture,” Brocart said of the technology. “We don’t have yet hopefully too much counterfeiting. But it might arrive at one point, and it arrives quicker than you think, so at least we are prepared.”
Global trade in fake and pirated goods reached as high as $464 billion in 2019 — the most recent year with data available — according to the Organisation for Economic Cooperation and Development, with fashion products being the most frequently seized by global customs. A previous report by an EU agency concluded Europe’s fashion sector lost nearly 10 percent of sales annually as a result of counterfeits.
While it’s doubtful fashion will be able to squash the problem completely, the hope is AI can provide a new weapon in the arms race against counterfeiters, whose methods are constantly improving. The fast-growing secondhand market is where it could have the greatest impact, as resale players such as The RealReal and StockX contend with allegations of fakes slipping through their defences.
Evidence suggests AI could be helpful. Researchers at Case Western Reserve University were able to attribute similar paintings to the correct artists with high accuracy just by analysing a small patch of each one’s surface topography.
“The AI methods are applicable across the board,” said Andrew Van Horn, a Case Western Reserve post-doctoral fellow in data science in art. But he cautioned the best results will come from a mix of technology and trained humans who know what to look for.
Spotting Fakes With Technology
Authentique, unlike most other AI-based authentication technologies, is designed for average consumers rather than businesses. Brocart said what attracted Patou was the ease of use for customers and itself. It didn’t need to manufacture and attach special tags like NFC chips, which can be removed or counterfeited themselves. It just added a step at quality control.
“All we need to do, at some point in the production process, is to have a few seconds [with] each product to capture that unique digital fingerprint,” Lock said. “We don’t have to interfere with the production or the manufacturing techniques.”
Authentique’s algorithm analyses the picture on a microscopic level, translating the data into a numeric code and creating a digital ID for the product it stores as an NFT so it can’t be altered. Later, in the app, a user will see a faint overlay directing them to align the camera with the same location on the product. Once they snap a picture, the app compares the code generated from the image with the one on file. Because it’s just comparing numerals rather than matching images pixel by pixel, it requires minimal computing power, according to Lock.
He said the technology is currently able to authenticate clothing, shoes and small leather goods. In a year, he predicts they’ll be able to add watches and jewellery to the list. Its accuracy rate in separating real from fake is greater than 99 percent and improving, he added.
Of course, Authentique has the benefit of access to the products it’s authenticating before they reach the marketplace. The greater challenge for AI may be catching fakes amid the vast ocean of products already in circulation.
Entrupy was founded in 2012 for that purpose. The company, which was selected for LVMH’s start-up accelerator in 2019 and has clients including Selfridges and luxury reseller Sellier, has trained its AI on a massive dataset collected from both authentic and counterfeit items reaching back decades — all the way to the 1930s and 1940s in the case of items like the trunks and riding crops today’s luxury giants produced before expanding into handbags.
“We get a lot of that information from customers around the world,” said Vidyuth Srinivasan, co-founder and CEO, noting Entrupy works with brands, retailers and even governments. “We also do our own primary research on the different kinds of counterfeits — what’s new, what’s coming up.”
Clients use special hardware, like an optical device the company provides that can attach to an iPhone and works like a microscope, or a lighted box that can photograph products from different angles. Comparing the data from those images with the information it has collected lets it identify fakes with greater than 99 percent accuracy, according to Srinivasan.
Bringing Trust to Resale
While a shopper who purchases a new bag directly from Patou may be reassured they can verify its authenticity, its the secondhand buyer most in need of the technology.
Counterfeit handbags are a frequent problem, though it’s actually footwear that’s most often faked, according to OECD data. The World Customs Organization once even named Nike the most counterfeited brand in the world, meaning resale platforms such as StockX have to be vigilant. The company has run into trouble on this front, with Nike saying in a lawsuit it was able to buy knockoffs from its site and one seller apparently involved telling Complex he unwittingly bought 38 pairs of fake Air Jordan 1s.
Paul Foley, head of brand protection at StockX, said he wasn’t able to comment on ongoing litigation. But he noted that StockX has been stepping up its efforts to use technology, including AI, to fight fakes. One programme assesses high-risk products and sellers. StockX has also started developing machine vision to catch fakes when they go through its authentication centres. The company has used Entrupy’s tech on items such as handbags, but it’s building its own system, which takes time.
“You definitely want to train the AI correctly. If you train it wrong from the beginning, the value is greatly diminished,” he said. Early results, he added, have “been very favourable.”
The RealReal introduced its own visual AI system to spot fakes, called Vision, just over a year ago to complement another programme named Shield that uses AI for spotting high-risk products and sellers. Chris Brossman, the company’s vice president of machine learning, wouldn’t share specifics on how accurate Vision is but noted it’s already used on more than half of the handbags the company receives and said The RealReal will keep scaling it across other categories.
“We started [on handbags] because it was the highest risk and [we’re] trying to use our advanced technology to have the biggest benefits,” he said. “Nobody counterfeits a one-dollar bill. They counterfeit the hundreds.”
The next step will be to launch Vibranium, which combines Shield and Vision into a large multi-modal model. (All the names, you may have noticed, are references to the Marvel universe.)
Everyone BoF spoke to noted one of the benefits of AI is that it continues to improve over time as it gets more data and examples of real and fake products to understand. But Van Horn of Case Western Reserve University also said the role of human operators often gets overlooked. Both Foley and Brossman pointed out that their systems don’t stand on their own. They’re meant to complement teams of human authenticators with extensive product knowledge.
“They’re the OGs,” Foley said. “They touch and see all the product … It all starts with them and it ends with them.”
