Instagram caused a Twitterstorm this week when it told users that it retains the right to use user images in advertising. Since images dominate content shared in social networks, brands are taking a keen interest. But even if Instagram, Facebook, Pinterest and others wanted to target images, it raises a technical issue: How to target advertising at the content of images if you don't know what's there?
A coterie of startups are working to sort this out, including Gazemetrix, which recognizes brand logos in images, Image Space Media, Stipple and others.
Startup Luminate took on the problem for L'Oreal brand Garnier this past fall for the launch of a new product, indeed a new class of product called BB cream (beauty balm), which has been popular in Asia for years but is just now becoming prevalent in the U.S.
Garnier was interested in targeting images of women -- not just any women, mind you -- but fresh-faced, beautiful women with clear skin, which might put other women into a moment of beauty aspiration.
"We targeted images of women, not just celebrity women but women with beautiful skin, women in fashionable environments," said Michelle Ryan, VP of marketing for Garnier Skin Care.
The thought here is that when women see an image of a woman with great skin, they are in the mindset to learn more about products. "Skin care in particular is a category where women seek out content before they go to a store and buy," she said.
But image targeting is harder than it looks and in the case of the Garnier campaign, technology could only take it so far. Luminate used a mix of technology and human hands that gives a sense of how tough it is to do at scale.
"Humans will be part of the process for years to come but with each passing month the percentage of work that must be done by humans will be reduced as image-recognition software and metadata analysis improve," said Chief Revenue Officer Chas Edwards.
Luminate indexes images from 7,000 publishers, including Yahoo and CBS. Initially, the technology throws out images where it detects porn, violence or hate speech. Then, it looks for women where the face takes up a big percentage of the page.
That's where the robots and algorithms stop and people come in. The next step -- finding women with beautiful skin -- is left up to Luminate's team of freelance screeners, which narrowed down 20 million images of women found by the software, 230,000 were considered "good look-a-likes" for the campaign.
"The software is not very good at telling you if they have beautiful skin; we use humans to validate that ," Mr. Edwards said.Garnier then attached ads that appear when a reader mouses over the image. Garnier and other beauty advertisers are big buyers of print and TV, but more and more, L'Oreal is investing in content that it can provide to women during the research process, which is what this campaign is about.
This is a little different than Luminate's first business model, which was identifying products in images for brands and retailers like Macy's , Gap and Norstrom that are similar to once they're selling. Now they're looking for images that fit a profile and might have little or nothing to do with the brand, but allows the brand to attach its message to a desirable image.
"For McDonald's we located images that express the 'look of winning;' for M&Ms. it was 'signs of affection,'" he said.