Op-Ed: Data-Mining Lemons

By Kiran Aditham 

We welcome the return of our monthly contributor Simon Mathews, currently chief strategy officer at West Coast shop, Extractable, who’ s also worked on the strategy side at the likes of Isobar and Molecular during his career. As per usual, we’re not really sure how to preface Mathews’ entry, so let’s just let him do the talking. Take it away, sir.

For the last few days, I’ve been buried deep in a stack of spreadsheets analyzing an annual website user satisfaction survey for one of our clients. The client is a large business-to-business technology firm, primarily targeting engineers, and hence the survey is rich in specific questions around product interests, content consumption and ability to achieve key tasks. This rich data is augmented by connecting reported behavior (the survey) to actual behavior via web analytics to build an all-round picture of their audiences.

When deep in data I have a tendency to sometimes become somewhat (very?) distracted. This week, a particular weakness of mine, geographic trivia, got me sidetracked.  Did you know that, for example, since the revolution / war in Libya there is now no sovereign country in the world with a national flag that is a single solid color with no other markings on it?  So, obviously I love challenges such as the “View from your window” completion on the Daily Dish, where you have to guess a location based just on a photo taken from a hotel window, or Geo Guessr, where the challenge is based off a random Google street view.


While digging around trying to pinpoint a lone farmhouse in the Australian outback (thanks Geo Guessr) a couple of thoughts struck me.  First, that such a task is now eminently possible, thanks to the wealth of information and data available to everyone, instantly. And, how some businesses still don’t realize how this is changing their customers, but that is a topic for another day.

The second realization was how much of the information I was using to locate the Australian farmhouse was user-generated, social or crowd sourced. The photo from a family vacation posted on Flickr, the Yelp review of the local mechanic, the Tumblr of Australian farm equipment.  In my one Outback search I may have consumed some 20 pieces of such content before nailing down the location to a few hundred feet (Winning!).

Back on the client survey we had asked a number of social media-related questions this year. We asked about their personal social media usage as well as about their usage of social media in their professional life. Around half the respondents engaged in social media and user-generated content for ‘professional’ purposes and the vast majority (90%) of those that did, only used it for research. They, just like me with my geo-quiz obsession, were assimilating large amounts of content to make design and purchasing decisions about our client’s products, but were not sharing much (or any) content back out into the world.

This is not an unusual dilemma for a B2B company – knowing how important these sources of information are in purchase decisions, but how hard it is to get customers to share such content.  For this client, the insight was the highly disproportional size of the ‘consumers’ vs. ‘givers’ pools. And when looking at more specific data points—for example, sub-types of their audience—Google+ was used more than LinkedIn or Facebook, and way ahead of Twitter. We also saw a lot of variation by audience type and product and content interest. This would lead us to form a new strategy where we may micro-target very specific pieces of content to highly relevant segments and then provide a set of more narrowly focused sharing tools. And by increasing sharing in each niche, we could potentially lift the overall of content sharing and re-use.

True, that bit of data-mining led to some useful insights and a new approach. Some of the other analysis led more towards the ‘lemon’ end of success spectrum. We did find patterns, but, what do they mean?

For example, users who visited the site more often (weekly/daily) were two times more likely to be using a laptop than a desktop computer to do so, than their less-frequent brethren. Clearly such a pattern piques our interest, but how can it help us?  Do we re-configure the page layout on detection of a frequent user, assuming they are using a laptop? Probably not.

Also we found that those who use social media more frequently rank their satisfaction with the site higher than less frequent users of social media. This site is light on social features today, so it’s not likely that the social features of the site are making users more satisfied. Could it be they are spending so much time on Facebook that they are just happy people? Probably not. Again, an interesting data point, but a bit of a dud in terms of actionable insight right now.

What we did find was an interesting pattern across a number of similar variables. Any factor that showed some level of higher engagement with the site (such as frequency, looking for more types of information, etc.) or any factor that seemed to show a more heavy web user (social media usage, sources of information used, etc.) had a positive correlation with overall satisfaction.

So if one used the site more, or generally used the web more, he/she ranked his/her satisfaction with the site higher. The natural follow up question the survey could not answer is whether we have skillfully built a site that engages frequent users or whether frequent visits makes a user more able to engage?

This is clearly where we can start a long conversation on the nature of digital ‘engagement’ but we will also hold that for another time.

For us, this data is leading to an actionable insight: That by adding more personalized tools, content and accelerators for our more frequent visitors (and other sophisticated users) we can drive deeper engagement and higher satisfaction, and hence higher revenue/value. And for less frequent users, we need to drive content that helps them to start the process of engaging deeper.

Data mining is valuable even when you keep hitting lemons, as deeper understanding of user behavior always pays off eventually.