Options
Personalization without Interrogation: Towards more Effective Interactions between Consumers and Feature-Based Recommendation Agents
Journal
Journal of Interactive Marketing
ISSN
0892-0591
ISSN-Digital
1520-6653
Type
journal article
Date Issued
2009-05
Author(s)
Murray, Kyle
Abstract
Software agents that provide consumers with personalized product recommendations based on individual-level feature-based preference models have been shown to facilitate better consumption choices while dramatically reducing the effort required to make these choices. This article examines why, despite their usefulness, such tools have not yet been widely adopted in the marketplace. We argue that the primary reason for this is that the usability of recommendation systems has been largely neglected - both in academic research and in practice - and we outline a roadmap for future research that might lead to recommendation agents that are more readily adopted by consumers.
Imagine that you are considering buying a new car. You have some general ideas about what you like and do not like, but your preferences are fairly vague and your knowledge of the marketplace is quite limited. Fortunately, you have a friend who is an automobile expert, with an exhaustive knowledge of what is for sale and a deep understanding of the consumer decision making process in this domain. You meet your friend for coffee and, after some small talk, you tell him that you are looking to buy a new car.
As you might expect, your friend begins by asking you a few questions about how you plan to use the car. However, his approach is a little unusual - he runs through a long list of potential uses and asks you to rate, on an 11-point scale, how important each one of them is to you. Nevertheless, since he is the expert, you play along. He then asks you about the price range that you are interested in. You tell him around $30,000, but he will only accept a range of prices, and so you say $25,000 to $35,000. At this point, you are starting to feel a little annoyed, but you remain hopeful that this strange interrogation will lead you to the car of your dreams. Your friend then goes on to ask you about various brands, body types, interior features, engine types, and safety features - and he wants you to tell him how desirable you think each of these is, again using an 11-point scale.
In some of these categories, you are really not sure what your friend is talking about. In others, you don't have a strong preference one way or the other. He tells you to just skip any questions you do not understand or are uncomfortable with, but warns you that this could reduce the quality of the advice he will (eventually) be able to give you. So you go ahead and dutifully answer everything that he asks. Just when you think he has run out of questions, he asks you to tell him which of the long list of car features that you have been discussing are the most important to you and which of your preferences are most deeply held. You again provide an answer, and after that he (finally) tells you which cars he is recommending - he presents his top-five list of the models that he believes would be best for you. You exchange a few pleasantries and head home to decide which car to buy. At this point, you probably feel confused, maybe a little frustrated, and quite certain that the process you just went through is not something that you want to go through again anytime soon.
Yet, this is precisely the type of approach used by many of today's "best" feature-based product recommendation tools for consumers. (Although, while your friend at least talked to you, most of these tools would require you to type your answers). In this article, we argue that the knowledge and technology exist to allow us to build better recommendation agents (RAs) and facilitate more effective interaction between consumers and such agents than what is evident in current practice. We begin with a brief review of prior research suggesting that RAs have the potential to substantially improve consumer decision making. We then argue that the usefulness of recommendation systems will not be recognized by consumers until these tools become more natural and easy to use. We sketch out a roadmap for future research in this area and comment on the theoretical and practical implications of improving our understanding of consumer-agent interaction.
Imagine that you are considering buying a new car. You have some general ideas about what you like and do not like, but your preferences are fairly vague and your knowledge of the marketplace is quite limited. Fortunately, you have a friend who is an automobile expert, with an exhaustive knowledge of what is for sale and a deep understanding of the consumer decision making process in this domain. You meet your friend for coffee and, after some small talk, you tell him that you are looking to buy a new car.
As you might expect, your friend begins by asking you a few questions about how you plan to use the car. However, his approach is a little unusual - he runs through a long list of potential uses and asks you to rate, on an 11-point scale, how important each one of them is to you. Nevertheless, since he is the expert, you play along. He then asks you about the price range that you are interested in. You tell him around $30,000, but he will only accept a range of prices, and so you say $25,000 to $35,000. At this point, you are starting to feel a little annoyed, but you remain hopeful that this strange interrogation will lead you to the car of your dreams. Your friend then goes on to ask you about various brands, body types, interior features, engine types, and safety features - and he wants you to tell him how desirable you think each of these is, again using an 11-point scale.
In some of these categories, you are really not sure what your friend is talking about. In others, you don't have a strong preference one way or the other. He tells you to just skip any questions you do not understand or are uncomfortable with, but warns you that this could reduce the quality of the advice he will (eventually) be able to give you. So you go ahead and dutifully answer everything that he asks. Just when you think he has run out of questions, he asks you to tell him which of the long list of car features that you have been discussing are the most important to you and which of your preferences are most deeply held. You again provide an answer, and after that he (finally) tells you which cars he is recommending - he presents his top-five list of the models that he believes would be best for you. You exchange a few pleasantries and head home to decide which car to buy. At this point, you probably feel confused, maybe a little frustrated, and quite certain that the process you just went through is not something that you want to go through again anytime soon.
Yet, this is precisely the type of approach used by many of today's "best" feature-based product recommendation tools for consumers. (Although, while your friend at least talked to you, most of these tools would require you to type your answers). In this article, we argue that the knowledge and technology exist to allow us to build better recommendation agents (RAs) and facilitate more effective interaction between consumers and such agents than what is evident in current practice. We begin with a brief review of prior research suggesting that RAs have the potential to substantially improve consumer decision making. We then argue that the usefulness of recommendation systems will not be recognized by consumers until these tools become more natural and easy to use. We sketch out a roadmap for future research in this area and comment on the theoretical and practical implications of improving our understanding of consumer-agent interaction.
Language
English
HSG Classification
contribution to scientific community
Refereed
Yes
Publisher
Elsevier
Publisher place
Amsterdam [u.a.]
Volume
23
Number
2
Start page
138
End page
146
Pages
9
Subject(s)
Division(s)
Eprints ID
220732