Sam Cox is CEO of Milabra, an advertising targeting technology company based on visual or image-based content.
How did Milabra begin?
Milabra the company was founded in October 2008 with the mission to bring a visual recognition technology platform to market. However, the Milbra story starts in earlier in 2007 when I met my business partner Naveen Agnihotri. At the time, he was a few years out of his post-doctoral fellowship at MIT, where he’d been running the Neuroscience lab. He had left to try and find commercial applications for his ideas and capabilities in cognitive science, pattern recognition and machine vision. He’d already achieved tremendous academic distinction having worked with two Nobel Laureates during his MS in Computer Science at the University of Georgia and his PhD in Neuroscience at Columbia University. In fact, Naveen’s original PhD work on how the brain sees and stores memory, contributed to his adviser Eric Kandell winning the Nobel Prize.
When I saw what Naveen had been working on, I was very impressed. He had developed a suite of technologies that were able to identify people, places and things within images. His software was truly capable of making machines see- not just pixel matching but concept matching: find a car, not just this specific car. I knew we had to start a company together.
We recognized that photos and video were going to be the fastest growing type of content online and would be the hardest to monetize, so we focused the company on the advertising vertical where adjacency was hard to filter in real time, and where visual content was not used for targeting. Both could be overcome with our software.
What problem is Milabra solving?
Milabra’s advertising solution is a demand-side technology (DST) that fixes two problems of buying media through networks and exchanges:
- We increase the amount of high quality, premium content available for purchase across the Internet. We do this by identifying and moderating risk factors across any web content when the greatest risk and reward are exist. This is especially effective when buying audience for premium advertisers.
- We empower buyers to optimize their advertising to what the user is seeing in real-time through our visual optimization process.
The last year has been dominated by conversations about being able to purchase “brand safe” content and audience-based media across networks and exchanges. The phrase “brand safe” seems to have become synonymous with the concept of “finding porn”. We realized that for an advertiser, brand safety means much more. It’s really related to the quality of the user’s advertising experience- finding the right kind of visual content that is complementary to advertiser’s message (think beach scenes and travel), being able to target to specific layouts and color schemes on websites, blocking ads from adjacency to not only adult content but “hate imagery”, brand & logo conflicts or poor quality images (like consumer photos). We really wanted to return the same kind of editorial controls media buyers and planners had with traditional media back to them across all media on the Internet.
Unfortunately, the reality of delivering on “brand safe” content has meant that addressable inventory becomes severely constrained. Brand safe campaigns typically have only been run on a limited number of websites that are known to have fairly static, well-engineered user experiences with very controlled content (i.e., Comscore 200 sites). Additional safety and site reach has been created recently by semantic technology that allow buyers to target or anti-target based on the text information the user is reading. The constraints become worse buying brand-safe audience because the opportunities to deliver become further constrained trying to find those high-value users again. All of this is wrapped around limited transparency that’s been brought to the process by verification systems that only give buyers after-the-fact information about when there was an “oops” in delivery to less-than-desirable content.
Ironically, for display advertising, an inherently visual experience, there has been no way for buyers to make media decisions based on the total visual environment the user is sees on the page that defines their overall ad experience. Milabra’s technology can see the image content, image quality, page layout and design and pass that data back to the ad server for decisioning in real-time. As a result, demand-side players have more safe impressions available to them and have visual data to optimize their buys. Publishers will benefit too by seeing increased eCPMs across exchanges and networks.
Discuss your data and media businesses. Where do you see momentum going forward? How does it breakout percentage-wise – data vs media biz?
We currently run two businesses, both data, and media, and we’re split about even between the two. We’re getting a lot of momentum from advertisers because they recognize visual value- they want to find those visual display ad opportunities while protecting themselves from inappropriate content. However, buying visual media is still a developing practice, so we offer media as a service to our clients until they feel comfortable buying it on exchanges directly.
On the data side, we’re seeing a real pick up in momentum. Because our service is available through REST APIs, integrations are generally pretty straightforward. We’re getting good traction with verification companies looking for a real-time solution, analytics firms that want to better understand how the visual environment affects consumer behavior and DSPs who are looking for a partner who can add another data-point in their RTB stack. We love the DSPs and RTB exchanges, and think that our business will grow along side them. Our media business is proof of demand for our product.
Please provide a use case of how Milabra data is created and the resulting targeting parameters.
It’s a pretty straightforward process, Mii (Milabra Image Intelligence) is given a URL to process. If we’re given a page URL instead of an image URL, our technology takes a screenshot of the whole page so Mii can analyze the images and the visual characteristics of the whole page. This is where the technology gets really cool. The Mii platform takes the image and sends it to our Visual Dictionary for lookup and identification. The Visual Dictionary is where our Neuroscience technology comes into play. The dictionary “looks” at the image and finds things like people, places or abstract elements that impact consumer decision making like color, light and volume. The output from the Visual Dictionary is then sent back to Mii to be delivered to the client. The data is a threshold that indicates the probability of a match.
The use of this data depends on who you are. A media buyer can leverage our data to see if the impression they are about to buy is next to certain types of content (or avoid other types). Publishers generally want to store the data directly in their CMS for auto-tagging, moderation or ad targeting purposes (passing it into ad tags). Multivariate testing or dynamic creative shops can use our data to increase advertising performance (for example, page color matching).
What is a DST? And, why do you call Milabra a DST?
A DST is a Demand Side Technology. Although we produce data, we felt it was important to clearly contrast ourselves to data players/audience segment providers. We call Milabra a DST because we’re not a cookie technology, but a technology company that works to improve the value of media and the performance of media in the context of an advertising campaign- what we call the “commercial alignment” of visual data. We don’t just categorize images and put that data in someone’s database, this information is immediately actionable. However, because we are a technology solution, we provide value to the entire ecosystem not just one side of the equation.
How do marketers access Milabra data? What types of marketers does this work for? Brand, DR?
Marketers can access our data in several ways: through the exchange on places like AppNexus, where Milabra is an integrated Data Provider or by placing media buys directly through us. They will also be able to access it soon through several DSPs, although I’m not at liberty to discuss which at present.
Because the data is rules-based from without manual input, it’s ideally suited to be another tool in the arsenal of performance marketers who are looking for pockets of gold. Our data allows DR marketers to consistently “see” the Internet and figure out how to make the visual environment work for a direct response offer. Milabra data is ideally suited for brand advertisers too, whether they are looking for visual contextual relevance or safe ad placements. Our software expands scale, reach and availability of additional inventory.
How does the revenue model work for the data?
The model is volume based, rev share or flat fee, depending on the integration, customer type, and volume. We can integrate throughout the ecosystem and as a result, it’s pretty easy for us to figure out deal terms that work for our partners.
Is there a connection between what Milabra is doing and data exchanges?
Milabra is not connected to a data exchange. Since we’re a data technology, and not a cookie, we’re a bit different than the type of data traditionally traded on data exchanges. If those data exchanges grow to be more than user segment warehouses, we’ll probably start providing Milabra data there, but not in the immediate future.
Do you pay publishers for visual data? How can Milabra help publishers?
We don’t pay publishers, and publishers are not really the focal point for Milabra because of the way they are aligned with the value chain. What we’re doing is providing transparency about the publisher experience to the buyer- not reselling data about the publisher’s audience. Although our alignment is on the demand-side we do work for publishers who are participating on the exchanges because as buyers identify visual elements that are valuable to their campaigns, they’ll bid-up for those impressions as they are available increasing publisher yield.
That being said we do have some direct integrations with publishers who have a significant amount of user generated image content (i.e., photo sites). Our data helps them manage their business soup-to-nuts from business operations to ad operations. Remember, approximately 95% of visual media assets are uncategorized so it’s a very unwieldy asset to manage- especially for content moderation. Milabra has seen some pretty horrible things on top sites. Although there may be EULA’s in place, they are toothless without enforcement, and there is very, very little enforcement.
Because our solution operates in real-time we create a two-tier marketplace for impressions: an expanded set of premium inventory for risk averse advertisers and identified “risky” inventory set for advertisers who can tolerate the risk. Our research has found that in general UGC content receive about 60% of the total lifetime of impressions (commercial opportunities) in the first 24 hours as it’s shared with friends and family (or hopefully goes viral!). Within 7 days that content fades into the tail to be visited rarely if ever again. Most manual moderation schemes take 24-48 hours to classify image content and as a result, either most of the commercial opportunity is lost since there is no data or the risk for a brand advertiser is greatest because most of the impressions are delivered when the content is not moderated. Using Milabra technology allows publishers to streamline their revenue management.
What’s the status of the company’s funding? Will you be looking for more?
Currently Milabra is privately funded. It’s been a good way for us to weather the macro economic storm and the pullback in investing from VC’s, especially on the east coast. As our operations are starting to grow, we’re considering VC investment.
Follow Milabra (@milabra) and AdExchanger.com (@adexchanger) on Twitter.
