Web personalization technology potentially makes it possible to match recommendations to customer preferences and thereby open up business opportunities for online merchants. However, the practice is rather more complicated than the theory as there are important decisions to make. One of the most important is when to make recommendations.

Technology allows for adaptive personalized recommendations that that are based on a customer's browsing and clicking activity on a website. This offers a real-time match to a customer's preference as opposed to a static recommendation based on the customer's stored profile, but it is also more time consuming to achieve.

Shuk Ying Ho, David Bodoff and Kar Yan Tam investigate the problem using both a lab experiment and a field study and they find that online merchants face a dilemma when it comes to personalized recommendations.

"On the one hand," they say, "if a personalization agent is designed effectively, adaptive personalized recommendations that are presented later are more likely to match a consumer's interests. This is because the recommendations are based on a longer period of observing the consumer."

"On the other hand, all else being equal, the effectiveness of offering personalized items is likely to diminish over time. The consumer may feel he has already found something satisfactory and decides it is not worth the effort to search any further."

The authors investigate the problem by asking 276 students to visit an experimental personalized online bookstore to search and select a book, then asking 199 users of an online music store to search and select a music track. The timing and type of personalized recommendations (adaptive or static) is varied among the participants.

The results show an interplay between timing and quality of recommendations. For example, in the music study, 24 per cent of participants downloaded early static recommendations but this plummeted to 4 per cent for late static recommendations, indicating a downward match to their preferences as they encountered other options.

In early adaptive recommendations the download rate was 19 per cent and this fell only slightly to 17 per cent for late adaptive recommendations. Clearly time is a factor as customers prefer to make decisions earlier, but adaptive recommendations are more likely to tailor with their preferences.

The authors also consider consumer expertise, which can affect sampling behavior, and find the combined effect of time is more favorable for experts.

"Our study shows that the benefits of dynamic content aren't easily or completely attainable. Quality and timeliness present an inherent trade-off and it is not possible to attain the best of both. Even assuming it's still worth the investment, a manager is left with determining how to weigh between the two competing values of better matches and timely recommendations. In some cases there is a higher risk from delay, for example if a user session is typically shorter, and in other cases the opposite may hold."

"Our findings also offer some practical guidance by identifying that the upward effect of adaptiveness is enhanced as user expertise increases. If an online shop knows its users are on average experts, the argument is increased for later presentation. Better still, if they can segment users into novices and experts then they can enhance the delay for experts and perhaps weaken it for novices."