A Two-Stage Model of Generating Product Advice: Proposing and Testing the Complementarity Principle

Abstract

Most extant research into product recommendations focuses on how advice from recommendation agents (RAs), consumers, or experts facilitates an initial (or single-stage) screening of available products and provides relevant product recommendations. The literature has largely overlooked the possibility and effects of the second stage of product advice using a recommendation improvement (RI) functionality, during which users can refine and improve the accuracy of the first-stage product recommendations. Thus, our understanding of how users make product choices is incomplete. To rectify this, we propose a two-stage model of generating product advice, and we use it to test what we propose as the complementarity principle. This principle posits that the first-stage recommendations (personalized or nonpersonalized) influence the impact of different types of second-stage RI functionality, which augment the first stage by facilitating either alternative-based or attribute-based processing. Results show that the complementary synergies between the two stages result in higher perceived decision quality, but at the expense of higher perceived decision effort. We contribute to the literature by helping researchers better understand users’ adoption of the second-stage RI functionality in conjunction with first-stage recommendations. In addition, e-commerce designers are advised to provide different and complementary types of recommendation sources and RI functionalities to facilitate online consumers’ decision making.

Key words and phrases:

• online recommenders
• perceived decision effort
• perceived decision quality
• personalized recommendations
• product recommendations
• recommendation improvement
• recommendation sources

Acknowledgment

The authors thank Vladimir Zwass, the Editor-in-Chief, and the two anonymous reviewers for their constructive comments and suggestions that significantly improved the paper’s quality and contributions. The authors acknowledge the financial support provided by the Natural Sciences and Engineering Research Council of Canada.

Notes

1. The two-stage model of generating product advice we propose is not the same as the traditional two-stage decision process documented in the literature [23, 54]. Our proposed model generates new product advice in the second stage followed by product evaluation at a subsequent stage, while the traditional two-stage considers product evaluation as the second stage.

2. Compensatory RAs lead to better decision quality than RAs with noncompensatory strategies [19, 77].

3. www.bestbuy.com/site/tvs/guided-buying-tv/pcmcat748302046702.c?id=pcmcat748302046702.

4. A content-filtering RA is different from a collaborative-filtering (COLF) RA that is based on the opinions of like-minded people to generate recommendations for users. Product recommendations from a COLF method may be less personalized than those from the content-filtering method as users may have many different interests and an item’s name may have different meanings [41]. Thus, we consider COLF recommendations as semipersonalized. As the personalization level of COLF recommendation for each individual user varies, COLF is excluded from this study. However, we expect that the complementary principle would also hold for COLF. That is, COLF should be combined with attributed-based RI (if COLF provides nonpersonalized recommendations) or alternative-based RI (if COLF provides personalized recommendations) to improve decision quality.

5. www.census.gov/history/pdf/2013computeruse.pdf.

6. To illustrate, in customized kitchen design shops, customers are often led through a showroom, and service representatives encourage customers to indicate what they like and do not like about the exhibited kitchen. These preferences can then be used to help select an appropriate alternative.

7. Among the 88 products, there were no dominant products because dominant alternatives recommended (or not) may confound the effects of the initial recommendations and the RI functionality. This is also consistent with prior literature showing that trade-off exists among product attribute values, for example, advanced features come with a higher price, and large display comes with heavier weight in general [81].

8. Although using a baseline website as a benchmark has its merit, it is likely that a participant may have looked for the same laptop in the second task. To explore this possibility, we examined whether participants selected the same product in the second task as they did in the first task. The data showed that only 14.6 percent of the subjects selected the same product in both tasks. We further examined whether these participants rushed to finish the second task by comparing the time they took in completing the tasks. The data showed that there was no significant time difference between two tasks (p = 0.55). In fact, those selecting the same products in both tasks spent slightly more time (0.68 minutes) than those selecting a different product in the two tasks.

9. We included both consumer and expert recommendations because it is unclear from the literature which one is better [9, 61, 72].

10. The equal weight strategy (EQW) strategy examines all alternatives and all attribute values for each alternative, and it has been advocated as a highly accurate simplification of the decision-making process [14].

11. The three items for perceived decision quality are: (1) I am confident that the product I just chose is really the best choice for me; (2) I am confident about the decision I made in choosing the laptop over the website; and (3) I am confident in making the choice of the preferred laptop computer [23, 27, 50, 56, 65, 71, 77].

12. Items for perceived decision effort are: (1) I took this task seriously; (2) I put in a lot of effort; (3) I was totally immersed in addressing this problem; and (4) I wanted to do as good a job as possible no matter how much effort it took [36, 42].

13. Sample item for task involvement is, The product selection task that I have experienced in the website was important to me [46].

14. The sample items are: The website could respond to my input on the Web interface and The website recommends laptops that suited my preferences [76].

15. Subjects were asked to answer the following semantic differential question with a dichotomous scale: To determine which laptop to select, (a) I examined all the laptops for a given attribute and then moved on to the next attribute; (b) I considered each alternative’s balance of values on all laptop attributes [54].

Additional information

Notes on contributors

David Jingjun Xu

David Jingjun Xu ([email protected]; corresponding author) is an associate professor in the Department of Information Systems at City University of Hong Kong. He received his Ph.D. in MIS from the University of British Columbia. His research interests include human–computer interaction, social media, e-commerce, and computer-mediated deception. He has published in MIS Quarterly, Information Systems Research, Journal of the Association for Information Systems, Journal of Strategic Information Systems, and Journal of Business Ethics, among others.

Izak Benbasat

Izak Benbasat ([email protected]) holds a Ph.D. from the University of Minnesota and a Doctorat Honoris Causa, Université de Montréal. He is a fellow of the Royal Society of Canada and professor emeritus in information technology management at the Sauder School of Business, University of British Columbia. He serves on the editorial boards of Journal of Management Information Systems and Information Systems Journal. He was editor in chief of Information Systems Research, editor of the Information Systems and Decision Support Systems Department of Management Science, and a senior editor of MIS Quarterly. He received the LEO Award for Lifetime Exceptional Achievements in Information Systems from the Association for Information Systems, and was conferred the title of Distinguished Fellow by the Institute for Operations Research and Management Sciences (INFORMS) Information Systems Society.

Ronald T. Cenfetelli

Ronald T. Cenfetelli ([email protected]) is a professor of management information systems at the University of British Columbia’s Sauder School of Business. He received his Ph.D. from the University of British Columbia. He conducts research into e-business, online customer service; the strategic uses of information technology; the behavioral and emotional aspects of technology usage; and human–computer interfaces. His work has been published in MIS Quarterly, Information Systems Research, and Journal of the AIS.