a thousand furnished rooms

For Jared Spool, Purpose, not Personalization, is All

Poorly conceived efforts to personalize can result in clumsy, off-putting sites

Jared Spool is a long-time--and highly sought after--advocate of taking time and care in designing human-computer interfaces. His Haverhill, MA-based company, User Interface Engineering, has been advising and training development teams since 1988 and is now 20 people strong. Spool and his team provide primary research, publications, and training for software developers involved in interface design. With the massive build-out of the Web, User Interface Engineering has turned its substantial resources to understanding how users interact with Web sites.

Spool himself conducts some of the training classes, and has a vast and ready arsenal of examples, ideas, and anecdotes of things done well and ill. We talked recently about the many efforts sites have made to personalize the visitor's experience.

One Fact Does Not Tell Much

"The thought of personalizing based on one or two known facts..." Spool begins, and then leaps to an example. "A person is walking down the mall and slows down to look in the window of one store. Will they slow down at they next similar store? If they happen to go into one store that sells dishes, will they go into every store that sells dishes? Will they ever buy dishes?" In other words, Spool is advising, one fact does not tell us much, if anything, about that person's future behavior.

UIE has researched shopping behavior both in malls and online. They've learned that even what people expressly declare they are interested in does not have a determinative effect on what they will do. "We observed one woman go to Crate and Barrel specifically to buy napkins. She never bought napkins. Instead, she happened upon a pitcher she liked, and bought it and the set of matching glasses. Had Crate and Barrel shown her only napkins, they may have ended up with no sale."

These two examples go directly to the two predominant types of information that Web sites use to personalize:

1. Information that can be inferred about a user based on, for example, click stream behavior, and

2. Explicit information that the user might agree to share with the web site.

We don't, Spool would contend, end up with enough information to fully understand the user's needs. And even if we did, we probably wouldn't know what to do with it. Web developers are coming face to face with a fundamentally complex problem--trying to determine what specific information a user might want at a specific point in time.

"Take Amazon.com, for example," says Spool. "A friend was pregnant two years ago, and purchased a book there related to pregnancy. Two years later, every time she goes to Amazon.com, they suggest baby naming books to her." Spool then lets the thought sink in for a minute. Of course, the baby is now a toddler and indeed already has a name. It's a single example, but a telling one for Spool, one that illustrates how the inference made from one piece of data has clearly led to an off-putting result.

Watch out for the Daily Special

Amazon.com is not alone in clumsily offering recommendations and specials. Spool's examples include the gardening site that won't sell you certain plants based on your zip code, apparently out of fear the climate may kill them. Another favorite example is the pharmacy site that will recommend specials on a certain medication, despite the fact that you've completed a lengthy on-line questionnaire and included information that you're allergic to that very medication. "Apparently," Spool deadpans, "It's ok to kill yourself but not your plants."

Spool's examples are catchy and often very funny, but they also make the point about personalization. "I'm not sure we know enough how to do it," says Spool. "And when we do it wrong, it's at best an annoyance."

The problem is what Spool calls the "indiscriminate attention" a site may end up paying to certain information. "It's nice at times for a third party to pay attention to certain interests and make recommendations. For example, if you go to a restaurant regularly, and a waiter knows you like a certain salmon dish that is sometimes available as a special, it's nice for the waiter to point that out. But you probably don't want that same waiter to start commenting on your recent choice in friends."

Other sites have adopted the approach of always fronting certain information, presenting the user with daily specials and offers, regardless of whether the user is there for such information. In fact, UIE's research shows that users almost always bypass that kind of information in favor of the things that interest them. "It's like going running into a restaurant needing the restroom," says Spool. "And having the waiter insist on reading you the daily specials."

The lesson? Specific information about a user is could yield some useful functionality. For example, it might be helpful for a floral web site to track the types of flowers you have sent, and then warn you not to send the same ones twice (or to suggest one that you've indicated was well received). But one or two facts, Spool suggests, don't merit redesigning the whole Web site. This is probably most true in complex applications involving knowledge workers. "You would have to predict what they need at any given time," says Spool. "And the odds of you being psychic are slim."

Ask the Right Question

For Spool, the way to tackle personalization isn't to start with the question, "What can we personalize?" The right questions to ask are, "What does the user need to see right now? What information does the user need?"

Spool offers another example. "Take a user coming to eBay. If that user has some bids on some items pending, it would be handy for the first screen they see to be a summary of their current bids. Which ones our being out bid? How long do the auctions have? If there are certain items that they've indicated they are always on the lookout for (personally, I'm a big fan of the 'Elvis Pezley PEZ dispenser'), eBay could display new items and have an easy way for them to place opening bids on those items."

Spool's final advice is that the ultimate solution may not even have to be expensive or particularly complex. "Watching users come to your site frequently would give you some ideas on the patterns that show up in their regular visits," says Spool. "Sometimes it will require personalization technology to optimize those visits, but often it can just be done with cleverly changing the content, without any real sophisticated tools."