The Center for Data Innovation spoke with Luca Boschin, chief executive officer of LogoGrab, an artificial intelligence and cloud company based in Dublin. Boschin discussed how companies use LogoGrab’s image recognition technology for their digital marketing strategies and address challenges like counterfeiting.
This interview has been edited.
Eline Chivot: Why has image recognition become so critical in marketing? What are the challenges it can address?
Luca Boschin: In the last few years, a new trend has emerged in the way in which we communicate with each other. As we gained access to much more information than before, we started to share a lot more visual content. Visual content is now surpassing textual content, because it allows us to share more data faster, and enables the peers we interact with to skim through all these pieces of information more quickly. It helps us understand faster where we want to focus our attention.
Image recognition is very important in this environment—but at least from a short-term perspective, it isn’t doing anything new. Previously, we would be sharing text. The only difference really is this growing importance of visual content. When we work with social media monitoring platforms, for instance, we still are extracting data, information, and insights for their brands—by targeting pictures rather than text. We still deliver to brands the same material and knowledge. It is the methodology we use that is different. It involves new technologies, which might make it slightly more complicated.
As visual becomes the main focus of marketers both in terms of producing and monitoring content, in the long term, new metrics and ways of producing and measuring content will arise.
Chivot: How can AI technologies like yours help companies leverage visual content, and how does it benefit their digital marketing strategies?
Boschin: Our solutions represent just a small piece of the equation—but it is a very important one. A logo is that visual element allowing you to instantly link an image to a specific brand. Spotting logos through our solutions helps users to skim through those huge volumes of information shared on a daily basis. By targeting logos, our technology also proves useful for businesses to understand when conversations are happening specifically around their brands, to understand who their current and potential consumers are, when and how they share content, and which brands they interact with in their daily lives. Brands get to identify moments of the “product consumption.” Brands can also target and reach out more easily to consumer groups such as millennials which, more so than others, particularly communicate through and engage with visual content.
A number of companies in the space of social media monitoring, such as Brandwatch, Crimson Hexagon, and Sprinklr, work with us to extract visual information about brands within the content they have access to. Our technology is not fundamentally changing their products, it just allows them to continue to exist in a visually-driven age. The value we add is to help them stay relevant to their customers, which are brands, by being able to keep delivering the same information they used to, with the same level of precision and performance. Today’s massive amount of visual content has made that particular mission more challenging, so we help them leverage that content and rise above it, rather than running the risk to be buried under it.
What changes is only the format in which we provide information. A social media monitoring platform such as Sprinklr has access to a lot of content, mainly images, which they share on social channels. We offer an API which allows them to send all this content to our system. We then deliver tags for each image, including information about which logos appear. Sprinklr incorporates this data into the traditional products they deliver to their own customers, in the form of graphs and analytics, spotting new insights and trends about their customers’ brands. For example, our technology allows Sprinklr and its clients to notice that a specific product, such as a brand of ice cream or beer, is consumed in specific geographies, times and settings—which the brands themselves may not have been aware of before. This is key information for a brand to gain knowledge and start understanding who actually enjoys consuming a product in a particular time and space. This way, they can direct their marketing investment to a specific audience, thereby increasing the likelihood of a product or of a campaign’s success.
Chivot: Imaging recognition is set to improve and transform e-commerce significantly. Could you explain some of the different ways in which image recognition can be used by online marketplaces?
Boschin: E-commerce is exploding. Such growth also brings threats, such as counterfeit items. This is a challenge for all: the brand, which loses profit and revenue, the consumer, who might be buying an item without realizing it is fake, and the marketplace itself, which wants to be a safe space users and brands can trust. Our technology allows online marketplaces to flag counterfeit products and to quickly remove or ban them from their platforms.
For example, eBay, is using LogoGrab to identify counterfeit merchandise on its platform. When the picture of a product for sale is uploaded on eBay, it hits our logo recognition service. We then tell eBay whether we find a logo on this picture, which helps them figure out whether this is a real or fake product—in very simple ways. For instance, they can check if the product name that appears in the picture is also mentioned in the title or description of the product. Counterfeiters tend to avoid doing that, and this is typically how they are caught. A bag may look like it is a Chanel product on a picture, but if the title refers to “luxury purse,” this is a red flag.
Our detection system may be more or less accurate depending on the product category. Before using our technology, eBay had a 33 percent chance of detecting counterfeit products. Since they’ve been using our technology, that number has gone up to 66 percent. It is not 100 percent yet, but this performance still is a huge step ahead in malicious content removal. And of course, we’re keen on improving that number.
Chivot: Your product can identify logos that are difficult to detect by other technologies. How will this affect the way in which logos are designed in the future?
Boschin: The reason why our technology has been gaining traction is that we have always decided to be extremely focused both culturally, as a business, and in terms of product development. Many players in the image recognition space aim to catch every single possible aspect and benefit of image recognition. Their technology is based on deep learning, a specific field of AI, and it works great for the classification of objects, for scene detection, or if you’re trying to figure out a concept that appears in an image. But it does not perform well in detecting features that are as specific as logos, let alone if these have simplistic shape. For instance, the Nike swoosh can be easily confused with something else in the “natural environment” such as tree leaves. Deep learning cannot accurately identify a textual logo like Samsung either.
We took a separate methodological avenue. We do not use deep learning technology, but have built our own from the ground up, specifically focusing on logo recognition. Of course, we had to make a choice. Our technology doesn’t work for objects or scenes or concepts. It can only be applied to detect logos or certification marks that indicate conformity to certain standards, such as “gluten free” labels. But it can identify logos that are difficult to detect, so designers no longer need to worry about adapting their design. They don’t have to work for the technology, the technology works for them. Our solution is also scalable: any given amount of logos or certification marks only takes us three minutes to be “activated.” This means training the AI for a new logo that is not yet included in our library. Say you’re a client of LogoGrab who needs to expand in southeast Malaysia next week. To “activate” the logos of unknown beer brands from this region, you just need to tell us the name of the brand and the website’s URL. We’ll be able to detect them all within three minutes, through the API. This delivers significant commercial scale to this company.
This short time frame is key to brands as they must ensure they can quickly identify when and where their logos are being published on the web. It also provides a powerful advantage to e-commerce platforms in counterfeit detection, by the way. Our clients want to identify fake logos so as to remove them as quickly as possible, without having to risk weeks of revenue loss and other long-term damage. This three-minute detection capability allows this to be done almost instantly.
Chivot: Where is image recognition technology mostly used today, and where will it be used next? What may be some of the obstacles to its application and use, by companies and users?
Boschin: Markets that are most ready and mature to start using it include social media monitoring platforms, especially in counterfeit detection for marketplaces. Sports sponsoring monitoring companies are keen users of image recognition as well: their job is to inform brands about the actual dollar value of their logo appearing within events they sponsor. Say Heineken sponsors a sport event or a concert. Video content is generated during the event, and broadcasted on television or social media channels. Traditionally, a company would manually track when, how long for, and in which size their logo would appear within this content. This job can now be done automatically and at scale.
Self-driving cars and medical imaging are promising areas of application for image recognition. Image recognition is the starting point for a self-driving car to work in the existing asphalt infrastructure. You could build entirely new roads with elements that would make image recognition irrelevant or useless, but this would mean costly financial and political investment. The added value of image recognition is that self-driving cars and other new systems can be put to work and used at scale precisely without having to go down that other road.
Regarding challenges in image recognition today—one would be that brands still “don’t know what they don’t know.” Many brand managers have no idea what they may be missing out on. A lot of educational work needs to be done to let them know about the opportunities and the value of image recognition. As this requires significant investment, rather than having to build our own product, we partner up with companies from various industries that have the energy and power to “educate” the market.
Awareness-raising could address another challenge. AI may be hype, but just as many people talk about it without knowing what it is about, businesses include it in their products just to be able to sell more. This tends to pollute a bit the whole industry. However, this is being organically cleaned up rather quickly: many of those businesses making such claims may disappear, because of the industry becoming more educated.