nPario strip-mines online usage data for more effective ads

Last week, Bassel Ojjeh, CEO of the data analytics startup nPario, gave me an interesting look into how behavior tracking is being used in advertsing, as well as a somewhat startling look at his previous employer, Yahoo.

Prior to starting nPario, Ojjeh ran Yahoo's Strategic Data Solutions group, which he says managed one of the largest data stores in the world. When he was there in 2009, he said, it collected over 8 petabytes of usage data from 550 million users, over about 16,000 servers. This data was used to improve Yahoo's user experience, advertising targeting, and so on. Ojjeh says that his group had a $70 million budget and 600 staf, and that the value of acting on the data, to Yahoo, was about $700 million.

Other big internet companies have similar initiatives, he said. Google, Amazon, and Microsoft all mine their data to improve their services and sell ever more specific advertising plans. And, of course, Facebook is emerging as the real power player in this field.

What nPario does is strip-mine usage data. The company uses the technology Ojjeh developed while at Yahoo; in a neat feat of business dealing, he got an exclusive license to further develop and sell his analysis technology as part of his new company when he spun out of Yahoo.

Ojjeh says his "insights engine" starts when his customers connect all their usage data to his products. At the moment that means server logs, but it is being expanded to include location data. Client EA also sends over what Ojjeh calls, "gaming telemetry," to enable behavioral targeting in games. (Ojjeh says, "Everybody with a big brand is looking at advertising as a way to augment their revenues.")

NPario specializes in finding "signals" in behavior that marketers can take advantage of. For example, if you're a marketer in the wedding industry, nPario can tell you who is about to get married. As he says, "there's no 'getting married' field" in server logs, but by watching behavior he can make a pretty good, and testable, estimation of which users are in that phase of their life.

The technology also tracks post-campaign behavior to discover the useful half-life of these signals. If you're looking to buy flowers, he says, a marketer has only about 40 minutes to serve you a message once the "I want to buy flowers" signal is detected; after that you've either made your purchase or moved on. If you're looking to buy a car, however, the purchase intent window is more like 12 weeks.

Practically, nPario gives marketers a console that lets them create e-mail campaigns based on the signals they are looking for. From a customer list it finds likely candidates, and then makes it easy for marketers to test a campaign on small subsets of the returned users until they are happy with the response. This technology could also, of course, be used for generating messages on a site or mobile device, but the email campaign manager is likely the easiest and most immediate way for an nPario customer to start using the technology.

My impression of nPario is that the technology is very sensitive to teasing out small signals in large data sets; it's probably less useful on small sites. Ojjeh notes that the businesses his technology could really help are those that collect the greatest volumes of online activity data: the ISPs. However, for a variety of good reasons, like the fact that ISP customers would flip out if they were being tracked this actively, this level of analysis can't yet be applied to general Web behavior.

Or can it? Google and major portals like Yahoo still collect inclusive behavior data on their own networks of sites and through toolbars. NPario doesn't track users across client boundaries, however, as some advertising tracking cookies do.

And the next frontier in behavior analysis and tracking space is just now opening up: The mobile Web. Again, it's the connectivity providers -- the mobile carriers -- that could have gained the most from this technology, and again, they have not stepped into this role. While nPario does have two mobile operators as clients, so far it's the companies just a little closer to the consumer that are collecting this information. Why do you think, Ojjeh asks, that Google is pushing so hard on Android?

See also last week's Reporters' Roundtable podcast: Online advertisers see all, know all,