Personalization Bashing – Almost

futurelab default header

by: Gary Hayes

A surprising article from New York times right in the epicentre of personalized media. Entitled “Like This? You’ll Hate That.” the article begins on the personalization-bashing band wagon, citing the recent Wal-Mart fiasco and ends thankfully on the side of ‘this-personalized-thing-seems-to-make-sense’. The article also brings key terms into the readers conciousness by referring to:

  • Personalized recommendations
  • Recommendation software
  • Cross-selling technology
  • Collaborative Filtering
  • Recommendation Engines
  • Clustering

It is good to see that the writer Laurie Flynn does refer to current sucesses in personalization such as LivePlasma, iTunes, Amazon and NetFliz as good examples of how to do it properly – personalization through nuance, clever algorithms rather than, as in the case of WalMart, manual intervention to promote poor selling items (in the WalMark case some bright spark decided to cross-link DVD box sets – at that level only!)

At NetFlix, the online DVD rental company, for example, roughly two-thirds of the films rented were recommended to subscribers by the site – movies the customers might never have thought to consider otherwise, the company says. As a result, between 70 and 80 percent of NetFlix rentals come from the company’s back catalog of 38,000 films rather than recent releases.
“The movies we recommend generate more satisfaction than the ones they choose from the new releases page,” said Neil Hunt, NetFlix’s chief product officer. “It increases customer loyalty to the site.”
“The most reliable prediction for how much a customer will like a movie is what they thought of other movies,” Mr. Hunt said. The company credits the system’s ability to make automated yet accurate recommendations as a major factor in its growth from 600,000 subscribers in 2002 to nearly 4 million today.

The article also hints at something I am very interested in, cross-profiling – the ability to use profile data from one life-genre and apply it to another. For example if you like Volkswagons then you will like this kind of music (bbc news today)…

Another development under way is matching customer tastes across Web businesses, using knowledge of a customer’s tastes in music to try to sell them books, for example. “To date, that’s been largely uncharted territory,” Mr. Andrews said, though not for lack of trying. Web sites have long tried to develop systems for cross-selling among companies that protect customer privacy but also allow sharing of data.
While large online stores are having success through recommendations, smaller Web sites are having a more difficult time using the technology to their advantage. Developing a system for cross-selling is expensive, and perhaps most important, requires amassing a huge amount of customer data to be effective, said Patty Freeman Evans, a Jupiter Research analyst.
As a result, according to Ms. Evans, fewer than one-quarter of online shoppers make unplanned purchases when they are online, a far smaller percentage than customers at actual stores.

The last line is a surprising statistic indeed. Less people make impulse purchases in virtual stores than they do in real life. This does indeed suggest that online is either confusing, providing too much choice or just not good for browsing. My feeling is that is still is a long way from feeling personally relevant, the things we are exposed too are just not right for us. As I have mentioned in many other papers and posts Amazon and other recommenders are getting there – pushing people gently into getting something relevant by watching what people are doing. Like anything in the AI/personalization domain, we have to build clever rules and not intervene with or corrupt those rules once set. First generation, let users dictate the experience other users will get, 2nd generation well that’s another post. The market is still open for gentle and intelligent exploitation – c’mon Apple and WalMart don’t spoil it for everyone.

Original Post: