Nowhere is the need greater for automated, behaviorally targeted content than the mobile phone. The size, navigation and input issues of the handset demand smarter, personalized media. At long
last, this week Sprint rolled out the first attempt by a U.S. carrier to solve the content discovery problem. Using artificial intelligence technology from Dublin-based ChangingWorlds, the new
"Sprint Web" learns from a mobile browser's history her media tastes and priorities and begins pushing those categories onto mobile Web portal pages. ChangingWorlds developed the
technology over the past decade and has 50 international customers. Stephen Oman, program director, Changing Worlds, explains how this technology works and some of the ways carriers and marketers
ultimately can leverage a mobile phone interfaces that really listen and learn from customers.
Behavioral Insider: What is Changing Worlds' basic approach to user tracking and content optimization?
Oman: The founders were researcher in artificial intelligence-based personalization and how information can adapt itself to suit users. The basis thesis of their research is that information should adapt to whoever is looking at it; the user shouldn't have to adapt their behavior to look at the information. Out of that research came this idea of managing the interaction with the subscriber of a mobile device. Because of the limitations of the device there should be a lot of more intelligence around the content discovery problem.
BI: So this effort always was focused on mobile, not just a port of Web-based personalization?
Oman: Very much so. Browsing on early mobile device was pretty bad, and to an extent it had unique problems compared to online in terms of screen real estate and input. We took this technology to phone companies, who were the custodians of user information. The system watches what the subscribers do on a mobile portal and then essentially restructures that portal to suit that subscriber. The idea is to reduce the amount of effort it takes for a subscriber to get to the content they want. The content you go to quite often starts to move up through the portal hierarchy and eventually appears on your home page. So your home page is completely different from someone else's. It does this automatically -- not like online portals where you decide what it is you want to have on different places on the portal.
BI: How does the AI distinguish between serendipitous browsing and real interest in a topic? What triggers recognition of a pattern?
Oman: There is an AI engine that looks for patterns and there are certain parameters that govern when it starts learning about subscribers. If you go to look at something once, it is not an indication that you actually like it. If you go twice -- maybe, but if you go three or five times you can infer that you like it. There is also some decay mechanisms built in to counteract some things. If you have gone to a service like a baseball team and you aren't interested in them anymore, than over time it starts to degrade the priority for that service.
BI: How exactly does the mobile Web page itself change?
Oman: The first is straightforward. The services on the home page can be reordered. The items you go to most often go to the number one spot. The second technique is for items buried two, three, or more levels down in the portal to start appearing at higher level pages towards the top. If you are a baseball Mets fan and drill into that category all the time then eventually you will see the link to the Met move up to the top of the sport page and then onto the homepage. The more you use the system the quicker it will learn.
BI: Is the system only taking your specific behavior into account or also inferring affinities with content based on group behaviors?
Oman: The system is a form of recommendation. Yes it can do separate recommendation system as well. There is a rules engine that lets any carrier generate business rules on top of the subscriber. For example the traditional model will be targeting based on demographics. You might put this link onto peoples' pages if they fall into this demographic or that. The same thing goes for changing the home page to suit a group of subscribers if they are all very similar. And then getting more sophisticated than that the recommendation engine can try to identify different patterns of consumption across all the subscribers. So it can look at how similar people are and the similar items people buy. Recently we started looking at it from an advertising perspective. How do we expose consumer information to allow ads to be targeted while protecting the privacy of those users? We've come up with a way of logically grouping people together based on their shared behaviors and then those traits can be exposed to an advertiser.
BI: What do we know about how this personalization changes mobile Web usage?
Oman: We started [A/B] trials with operators. Very quickly we noticed a lot of people use a mobile portal but don't actually consume content. They spend all of their time navigating through the menus and then we lose the customer. We call them failed sessions because we are not delivering value to the subscriber. The first thing we noticed is how that comes down rapidly. Content starts appearing much closer to the user so they get to it quicker. Second, because subscribers can get to content quicker they tend to come back more often. You see an increase in usage by individual subscriber over the period of a couple of months after the system is launched. We were able to correlate that against direct increases in revenue. It generates additional usage, more eyeballs and impressions, so that eventually you can monetize them.
BI: Even though there is a clear value add for users, it also must make some customers more aware that their carrier does see and retain a lot of information about them. How are some of your partners handling user privacy and notification?
Oman: There are geographic difference but that is a key questions. Certainly in Europe they have had to be much more open because of the very strict EU data protection laws. But what we found is that once you explain to subscribers what it is you are doing with this data and the benefits you get from it there is very high opt-in to this service.