How to avoid sales of counterfeit items in ecommerce Part 1

Part One

The counterfeiting epidemic

The e-commerce industry has become the target of exponential counterfeit sales on various online marketplaces in recent years. While counterfeits have always existed on e-commerce sites, the problem now appears to have grown to epidemic proportions. With Black Friday fast approaching, it’s important to question if all e-commerce platforms are enforcing some method of Counterfeit Detection to protect legitimate brands and customers while shopping online. Taking Amazon as an example, the site has adopted a complex system for finding and eliminating counterfeits in which a sophisticated algorithm scans through all the site’s pages, flags possible counterfeits, and queues them for fraud protection teams to investigate manually, which is where the system appears to collapse. This begs the question: is manual reviewing of counterfeit items really a sustainable Brand Protection solution for online marketplaces?

We have already explored the importance of Counterfeit Detection in a previous blog-post, which you can read here. Although there are many existing anti-counterfeiting solutions available, we are now going to take a different approach. What if there was a way to detect counterfeits with visual recognition? Using logo detection technology, VISUA can find and eliminate counterfeit, fraudulent, and suspect items on various e-commerce sites. But don’t take our word for it. Let’s explore a use case in which the anti-counterfeiting methods we used provided one of the biggest industry leaders in e-commerce a successful and sustainable solution to their counterfeiting problem.

VISUA’s key account team: how they approached the use case

In late 2015 VISUA started talking to a large e-commerce client about Visual-AI tests for Counterfeit Detection. While we originally didn’t know what the client wanted to use our technology for, it became clearer and clearer as time went on that they needed us to find counterfeit and suspect items found on their online marketplace.

The client was quite mysterious initially with what they were proposing. They wanted to test the capabilities of Visual-AI, so the account team tried to find out exactly what the client needed in order to move them along as quickly as possible. At this point, the client seemed to hit a wall. They hadn’t tested our technology yet, but the sales team continued to check in with them every few weeks. It was clear that it was not the right time for them as they were still awaiting approval.

Be Sure To Also Read ->Counterfeit Detection Use Case Overview

The technology

Then they had a breakthrough. The client finally agreed to a test which they ran on a representative sample of their images, geared towards luxury brands. These would include the type of brands you would see visibly on high-end fashion accessories. The test performed very well which meant that we started to get clues as to how exactly the client wanted to use VISUA’s solution. To keep things moving, the client and the account team agreed on an initial Proof Of Concept (POC) period to last around three months.

This POC was also a success. The core capability of the technology proved itself from the off and it demonstrated to the client the full flexibility of the system. It’s important to note that within the e-commerce industry, there isn’t just one type of logo per brand. So taking the Nike logo as an example, any small alteration to it still passes to the human eye as a legitimate Nike logo, even though it technically may not be anymore. Our system was able to flag this automatically and VISUA communicated these “fake” Nike logos to the client. We then altered the methodology to detect this type of “modified” logo.

Challenges and concerns

Any new use case comes with challenges along the way. The “newness” of the solution and the fact that we were an external vendor were initial concerns. If the client was going to commit to spending money on this solution, there’s a risk there as they would have had to invest some of their reputation in it, but the account team addressed that pain point from the start.

The client also faced their own business challenges. It can be an uphill battle to keep track of newly suspected counterfeit items, some of which are very difficult to spot and therefore hidden from existing anti-counterfeiting measures. A lot of the time, systems may miss a lot of counterfeit content listings on a site, so the client needed something that fitted their day-to-day workflow. VISUA’s engineering team designed a solution to fulfil that brief so that it was easy for them to use. There were also certain escalations which came from the top of the company down. So for example, someone important and higher level would spot a problem and the client’s team would then be told to solve that problem as quickly as possible.

In addition, there were concerns from the client’s side about VISUA’s technology, such as whether or not Visual-AI would work at all and if it was even the right solution to apply. The usual questions you hear from an educated client such as this one are: if it works at a basic level, will it be precise enough, will it catch all the brands, and how easy is it to activate and track new logos and products? So the client decided to test VISUA’s technology against other providers.

Other anti-counterfeiting solutions

This client also had a couple of different Visual-AI providers to benchmark at the beginning. If you work for a massive global corporation it is standard practice to test a multitude of systems against the same KPIs to see who comes out on top. The client wanted to justify their business case and the fact that we could provide evidence of our strong performance and quick response meant that they could assess the technology quickly and comprehensively.

It turned out that VISUA’s Visual Counterfeit Detection technology had the highest precision, recall, and hit rate on the market. And there are other benefits to the technology too, such as the speed of the activation of new brands and our hard-won domain knowledge on counterfeits. VISUA has done this a few times already, so we know what to look for. When something new is suspicious, we are ahead of other providers who might only be good at Visual-AI in a “general sense”. And if there are fakes that a platform hasn’t spotted which may have been there for months, the added benefit is that they can clean them up quickly. This means that you would actually see the “confirmed fake” metrics decrease over time.

The technical challenge

With any new use case, it’s important to be prepared. The account team working on this use case approached the engineering team in an open-ended way so that everything they knew about the client and their problem was laid out. Both teams worked together to scope out what the client might need, and then quickly set up direct contact between VISUA specialists and their business user to gather crucial information and scope the solution. Then our engineers put everything in place, developed the methodology, and customized it. In addition, any time constraint was neutralized by the fact that VISUA can plan and customize everything quickly, which is always in the client’s favor. If you want to know all about the engineering team’s role to play in this use case, stay tuned for Part Two.

The results: is visual the best solution?

While many Brand Protection and anti-counterfeiting solutions exist, VISUA has tested an alternative Visual Counterfeit Detection method that has proven to be more efficient for one of the biggest industry leaders in eCommerce. The overall takeaway from this use case is that the logo detection results were extremely strong and this solution works very well for detecting counterfeits. It also solves the issue of having to manually sift through product listings in order to find and eliminate all the counterfeits present within an online marketplace. Using this automated solution, it is now possible to identify, review, and even eliminate counterfeit products and content entirely within VISUA’s Counterfeit Detection dashboard.

Check out Part Two of this article.