Update on Sun Web Personalization Initiative
By Gary Zellerbach on Apr 15, 2009
I wrote an intro previously on SMILE, the Sun Machine Learning Engine, our new personalization system that utilizes predictive analytics. It went live in January, 2009, and we've been carefully measuring results since. So far, we are seeing higher click-through rates for SMILE served ads versus default ads, which is what we hoped would happen. But overall click through rates on these ads, whether personally targeted via SMILE or simply default, are very low, so we need more bang for the buck. We believe that the small ads served by SMILE in the right-hand navigation are easily overlooked -- many users simply ignore banner ads and/or the right hand column, concentrating on the central content of interest.
So our next step is to provide personalized recommendations in the body of the page. Here is a draft mock-up of a "carousel" type approach we hope to use:
There are numerous challenges with this approach, as the page real-estate "above the fold" is highly valued, but that's where we need the recommendations to be if we really want them to be noticed and of value. A likely approach will be to make the carousel collapsible, so users can close it up if it's in the way, and then open it to look for personally recommended content. We'll also provide a new "All Recommendations" page that is solely dedicated to showing all recommendations for visitors.
We did a usability study recently to test these concepts and were pleased with the results. Customers told us:
- They like what we're planning. We have literally millions of web pages, so if we can use our analytics to help customers quickly find the products, services, and content of most interest and value to them, we are saving them time and making them more efficient. Definitely a "win win" for all of us.
- They would likely not notice the recommendations if they're not centrally located high on the page.
- They don't have serious privacy concerns. Because we are an enterprise business site, they expect us to use analytics to improve the experience, and they'd appreciate the benefit. They would look at things differently if we were their financial institution, for example.
- If the recommendations are not accurate or useful, they will quickly learn to ignore them. So the onus is squarely on us to ensure our system is up to the task of making accurate, valuable recommendations that meet our customers' needs.
This is an ambitious undertaking involving a whole bunch of teams across Sun, from Design to Publishing to Engineering to Analytics, so it's at least a few months away. I'll keep you informed of progress and of course announce when you can check out your own personal recommendations!