The Invisible Internet: How AI is making Image Analytics a Reality
For brands, knowing what consumers are saying about them is a vital part of the marketing and customer engagement process. However, with the number of pictures being uploaded to social media growing daily, getting a true perspective can be challenging. Lucy Ingham speaks to Lou Jordano, CMO of Crimson Hexagon, about how the company is using AI to make image analytics a reality
It has long been said that a picture is worth a thousand words, which may be why so many of us favour them over text on social media. But for brands trying to understand how consumers are interacting with their products, they often represent an invisible part of the wider internet.
Some platforms have begun to combat the problem with basic image recognition, allowing brands to identify images where their products feature. But without insights about sentiment, place or people, this only has a limited use.
Crimson Hexagon, however, is taking a different approach. By teaming AI with all the images uploaded onto the world’s most popular social media platforms, the company is able to provide deep insight for brands. Welcome to the world of AI-enhanced image analytics.
“Crimson Hexagon is a consumer insights platform and as such we help global brands better understand their consumers,” explains Lou Jordano, CMO of Crimson Hexagon.
“So we give access to those brands, to the world's largest volume of text and images across a number of sources. And we use our AI-powered capabilities to help them analyse insights that are relevant to maybe their marketing campaigns or their brand health; help them better understand their audiences or even understand what's happening in the broader market or specific to their competitors.”
Lucy Ingham caught up with Jordano to find out how the technology benefits brands how AI has made it possible.
Lucy Ingham: How important have images become to marketing?
Lou Jordano: “They're very important right now and are increasingly important. If you just look at the world overall and if you look at just social media in particular, more than three billion photos are shared every single day on social media platforms. And most of those, we estimate 85% don't have any text associated with them.
“So there's a huge volume of images out there. Those images are rich with context; they help these brands understand their product, influencers, all those kinds of things. But without specific text, those brands might be missing what's actually being said in those images if they're not able to analyse the images themselves.”
LI: What kind of insights do image analytics provide?
Lou Jordano: “There's lots of things. So most image analytics products out on the marketplace can detect a logo within an image. That is kind of like table stakes, that's not really that interesting anymore. Crimson uniquely provides the ability to go beyond just logo detection within an image; we can also spot a particular scene.
“Imagine you're looking at your brand logo in someone's picture, you probably want to know: are they on the beach or in the mountains? Are they in an office setting or in a pub? So what's the scene? It will detect facial characteristics: is it a man or a woman? What's an approximate age? And then what's the general emotion: happy, sad, frustrated, angry – all those types of things.
“We'll also detect are there other objects in the photo, so is your product always frequently in the same picture with this other product? That could be really interesting to know, right? So what's associated with it. And then is your product being involved in some kind of action? Is it appearing very frequently in sports cars or on a beach or those kinds of things?
“So it really gives you a very rich view of the photograph beyond just simple logo detection to figure out what's the setting; what's the general context. It can give you that information across any of those parameters to really help more finely understand the context of how your product is being used in the image.”
LI: Are brands surprised by any of the insights image analytics gives them?
Lou Jordano: “Absolutely, there are so many great examples here. We've got a great story around Grey Goose. When they analysed Twitter posts where their product has a branded mention, they figured out that their audience was mostly male: 73% was mostly male and they tended to be 35 years old.
“So that's an interesting insight, but that's based on text. When you filter those results to narrow it down to just looking at the images, you can get a deeper understanding of the audience and you can detect the female population, and the females that are associated with that brand and tend to be younger.
“So you could set up a campaign targeting a specific gender, for example, to say 'hey, here's our brand. Let's run a campaign just at women who are maybe 25 to 35' and that can be more specific messaging; a different call to action; a different value proposition perhaps. It gives the brand the ability to segment and more carefully define how they engage with different portions of their audience."
LI: I understand Samsung Electronics has also been using the platform?
Lou Jordano: “You bet. They use us basically for all of their product lines so they're very focused on brand. They use our platform to determine the brand health of each of their products; to monitor their performance of their marketing campaigns, so what are the topics that are frequently being associated with their brand mention? What influencers are also having an effect or are closely associated with their brand?
“They're able to also look at their product particularly around launches. They used us when their competitor was launching a new phone and they realised that the people who bought the competitor's product when that product launched were complaining about the life of the battery.
“The audience who was complaining about that battery life was of interest to Samsung, so when Samsung launched their new product for their phone they were very careful to point out how excellent their battery life was to appeal to that particular segment.”
LI: Has the technology that is driving your platfor m only recently become possible?
Lou Jordano: “Yeah, it really has come to maturity. The technology I think has been out there for a while, but it's really the AI capabilities that wrap around the image that really have come to the fore in the past couple of years.
“We capture, store and analyse every single image posted every single day to Twitter and Instagram. We store it all; it's all in our database, over 160 million images every single day. So we have this gigantic – really we think of it as the world's largest collection of unstructured text together with images. We have this really rich treasure trove of information that brands can use, and it's very helpful when they marry that text with the images because it gives them a more complete view of their consumer.”
LI: How do you anticipate this technology evolving in the future?
Lou Jordano: “I think the technology will evolve in correlation to the amount of images that are out there. Basically, the images are used as training data; you train your system to detect certain emotions, or to detect certain scenes or objects, right? So the more data that's out there, and as we know every day, there's lots of new images posted, it just becomes a more fertile training ground for the AI technology to become more sophisticated.
“Right now, for example, dumb image analytics providers struggle with detecting humour or sarcasm, those kinds of things. Just because they haven't trained their image analytics technology to be sophisticated enough, because maybe they don't have enough access to enough images.
“So that's why we have such a gigantic image library here, it's because we're able to be more sophisticated with our training data. So I think in general at the market level, image analytics will become easier to use and more sophisticated and more informative for brands as they train it with more images.”
Images courtesy of Brolly