What we Can Expect to Learn from Image Intelligence

Customer Experience
With so many digital images out there, businesses have a prime opportunity to mine those images for consumer insights and other learnings. But how far along are companies in analyzing images and how can they best leverage those insights?

How many photos are floating around the Internet? The answer is staggering. In 2014, people uploaded an average of 1.8 billion digital images every day, according to Mary Meeker's Internet Trends Report. And according to data released by Google, in 2015, we uploaded 24 billion selfies on Google Photos alone, not counting photos shared on other sites like Facebook or Instagram. To put those numbers into perspective, there are about 7.4 billion people on earth. 

With so many digital images out there, businesses have a prime opportunity to mine those images for consumer insights and other learnings. But how far along are companies in analyzing images and how can they best leverage those insights?

I posed these questions and others to Altimeter analyst Susan Etlinger, who recently examined the value of analyzing images in her report, "Image Intelligence: Making Visual Content Predictive."

If you think of analytics in terms of being descriptive, predictive, or prescriptive, at what level is image intelligence? 
Most are still in the descriptive phase. It's so early on that there's no such thing as an image intelligence platform today. What you see are mainly descriptive tools to identify the content in images and show sharing behavior.

Then there's the piece that's turning into predictive intelligence. You have to take the data from the content of the image--image recognition--as well as data about sharing behaviors and put that into something that has an algorithm that can tell you, for example, that images of people on the beach in the summer tend to be shared more that images of people in the winter.

That gets interesting when you start to get into influencer or celebrity campaigns because there's an expectation that when you sign up a celebrity to tweet about your product that'll be shared widely, when in fact a lot of companies I've spoken with have said that's not the case. It's not necessarily the celebrity endorsement that gets the shares. It could be people who have a cooking blog with a devoted following who are focused on a particular issue. Putting together analytics about image recognition, sharing behavior, and "predictiveness" is how you get to an image intelligence platform.  

How close are companies to applying real-time predictive analytics to images?
As a market we're not close. But there are enterprises that are using this technology and the more you use it, the better it gets. The point of artificial intelligence is that it learns. But for every use case, you have to train it. There's no perfect data set out there that you can use to train your data. Everything needs to be tuned to the use that you have for it, whether that's understanding cars or coffee or shoes, etc. and that takes time. 

How can marketers quantify the value of image intelligence?
Because this is a new market, the main way to quantify it is to use it in a pilot. You could look at it as a way to understand the investment in a sponsorship because you can see how the images are being shared. Another way is to use it to find user-generated images and get permission to use those in campaigns. That can save money from using stock photos or photoshoots. If you're a pharma company and worried about off-brand uses of your drug or fraudulent drugs being advertised as yours, that ability to find off-brand products and packaging could save millions of dollars. 

Do you see image intelligence becoming a commodified feature like social listening?
I'm hoping it doesn't become a commodified feature because if we think of image intelligence as just another feature, then that undermines its potential impact. This technology can be used for everything from brand metrics to innovations in product development. Even the supply chain can use it to look for defects or new product uses by seeing these products out in the wild and understanding them at scale. In terms of turning it into an industry standard, it's still a few years away. We're probably at least three years away from seeing this as a function that sits in a platform. But companies do have technologies that can handle some of these things, so the advancements are happening. 

Which types of image intelligence are showing the most innovation? Are more companies analyzing selfies versus emojis for instance?
Emojis are a different animal; they're structured data. There are relatively clear meanings of an emoji. But when you look at a photo of a woman holding a can of soda, what if there's a burning car in the background or a looted store. There are so many different ways to interpret a photograph and the systems today are just starting to understand context at a basic level. Also, all of the major players have something related to image analysis. It's just that it hasn't cohered as a market yet and it's not clear how far it will be used in enterprise.

When we look at social, it took a few years to be subsumed. Businesses are transforming into more digital enabled services, and this is another piece of that. And while image intelligence provides insight, it doesn't do everything. Image intelligence should be complementary to a data strategy.
What are some low-hanging fruit examples of using image intelligence?
I'd start with looking for brand mentions. Ask yourself questions like, are you finding images of your product being used in surprising ways? Do these images help you understand what your customers want and need and do they point to any undesirable uses or weaknesses in your product? Start with understanding what's out there. You can't measure what you can't see. I would also recommend companies do focused pilots to answer fundamental questions about how people use their products.