Features

In the previous KDnuggets News issue we published the beginning of the interview with Dr. Usama Fayyad, Yahoo! Chief Data Officer, covering

• Chief Data Officer title (first CDO in the industry),
• data mining challenges at Yahoo,
• his successes and failures at digiMine, and
• consulting lessons from DMX group.

Q5. Gregory Piatetsky-Shapiro: Can you tell us about some Yahoo successes with data mining?

Usama Fayyad: Yahoo! is the first company to hire a Chief Data Officer - demonstrating that data is a true, strategic asset to the company. Our goal is to create value to consumers and marketers by delivering the consumer-centric data platform & insight services that maximize user engagement and enable innovative marketing solutions. In a very short period of time, we have managed to have a significant impact on several Yahoo! products and services. The unfortunate thing here is that many of the contributions cannot be shared publicly (yet) because they involve competitive advantages or are specific to some of our advertising clients or products.

So rather than share the biggest successes, I'll share some of our more public ones:

Product integration: one of the examples that you see today on Yahoo! Mail is a visible result of data mining. One of the patterns we noticed in analyzing unexpected patterns in usage data was a strong correlation between people reading email and reading news, in the same session! When we shared this with the Yahoo! Mail product team, their first instinct was to test this effect: this was done by building a "news module" that highlights news headlines and showing this to a test group of consumers as part of the main Mail front page.

For a product like Mail, the critical business pain is to take new "light users", and turn them into "heavy users". If you do that, you reduce churn dramatically. Indeed, we showed from this test that churn in the weakest group was reduced by 40%. This led to the immediate development and launch of the News Module embedded in Yahoo! Mail's front page. Today, hundreds of millions of consumers see and use this product. I like this story because it highlights the quick reaction time of the product team who is obsessed with customer engagement. It also highlights the tremendous value lurking in the many, many patterns in the usage data.

Instant Messenger: we analyzed the drivers of the Instant Messenger usage, and out of the many factors, the data suggested that the most powerful factor to go after was to get users to grow their "buddy list" by at least 5 additional people. This led to a dedicated marketing campaign designed to achieve exactly this quantitative goal in order to increases the number of friends on your IM friends list and dramatically increase product usage.

Search box on Yahoo front page: A simple example had to do with discovering that on the Yahoo Front Page, centering the search box on the page (as opposed to having it be left-justified) would increase consumer usage. This led to better user engagement and there was no cost to Yahoo! to make the change. We discovered this by discovering the hidden pattern that showed that Netscape users tended to use search more than IE users, and by discovering that the only visible difference is the subtle position of the box! It was centered on Netscape browsers but left justified on IE browsers. A very unnoticeable difference, yet an important one. Who would figure that out???

Advanced targeting techniqiues: I would also list some of our advanced targeting techniques, such as discovering who is in the market for an automobile with high reliability, based on only anonymous browse data, as another big area of success for predictive data mining technology. This ability allows us to create very high-value audiences by understanding what they have in mind and biasing the advertising so it is more relevant to that audience: a win-win for consumers and advertisers; and of course a win for Yahoo! because reaching such purchase intender audiences commands a premium.