JCMC 5 (2) December 1999
Message Board

Collab-U  CMC Play  E-CommerceSymposium Net Law  InfoSpaces Usenet
  NetStudy VEsVOs O-JournHigherEd   ConversationSearching for Cyberspace

Internet retail store design: How the user interface influences traffic and sales

Gerald L. Lohse
The Wharton School, University of Pennsylvania

Peter Spiller
McKinsey & Company, Inc.

Table of Contents

Abstract

Given the resources needed to launch a retail store on the Internet or change an existing online storefront design, it is important to allocate product development resources to interface features that actually improve store traffic and sales. We identified features that impact store traffic and sales using regression models of 1996 store traffic and dollar sales as dependent variables and interface design features such as number of links into the store, hours of promotional ads, number of products, and store navigation features as the independent variables. Product list navigation features that reduce the time to purchase products online account for 61% of the variance in monthly sales. Other factors explaining the variance in monthly sales include: number of hyperlinks into the store (10%), hours of promotion (4%) and customer service feedback (1%). These findings demonstrate that the user interface is an essential link between the customer and the retail store in Web-based shopping environments.

Cybershopping

The promises of on-line shopping touted by the popular press include convenient access to greater amounts of information that enhances consumer decision-making and increases market penetration for the merchants. Numerous articles equally bemoan these promises. With titles such as "On-line shopping - Virtually Impossible!" critics are quick to point out that expectations are not being met (Glamour, 1996). As one cybershopper stated, "I imagined that buying clothes on-line would be as easy as clicking on a outfit and having it appear on my doorstep. But after the third time I waited more than five minutes for a fuzzy picture to download and then sifted through the information, I realized that the technology has not caught up with my imagination." Regrettably, the number of shoppers and total sales are still marginal, in part, because of poor interfaces and store navigation (Baty & Lee, 1995; Hoffman, Novak, & Chatterjee, 1995; Jarvenpaa & Todd, 1997, Lohse & Spiller, 1998; Ridgon, 1996).

Account managers, production staff and merchant partners should not assume customers do not want an item in a retail store if it is not selling. Nor should they conclude that a poor response to a given store design is because of the merchandising mix. It is important to take a harder look at the possible relationship between poorly selling items and screen design and layout. Could customers be having a tough time wading through the screens? Can customers find what they want in the stores? Are customers aware of what products are in the stores? After all, diligence in browsing a store is not a virtue Internet retail marketers should expect from their customers.

While store traffic and sales are adversely influenced by poor interface features, it is important to document and quantify how much sales are impacted as well as to understand the underlying consumer behavior. The number of levels between the store entrance and end product, the number of browsing modes, such as searching by brand or by price, as well as the consistent design of lists and menu bars should influence consumer buying behavior in an on-line marketplace. Using a regression model, we examine the relationship between interface design features and traffic and sales data in order to quantify tradeoffs among different interface redesign alternatives. The model explains variance in store traffic and sales as a function of differences in interface design features. This can be used to assess the existing store and to improve features that are below average. This technique can also answer questions such as: "What is the value of implementing a search function into a site?" or "What is the value of having a product featured on the home page of a store?" This type of data provides some arguments for redesigning Internet retail stores. Even small improvements in traffic and conversion rates can have a huge influence on sales. This research identifies store and interface features that impact online store traffic and sales.

Research Methodology

Survey Sampling

A previous classification of Internet retail stores by Spiller and Lohse (1998) identified five distinct types of online retail stores. In the current research, we focus on one of those stores categories that we term Super Stores. Super Stores have a large selection of products. Average information for the customer is extensive, including information about the company, ordering, gift services and "What's new?" sections. Most Super Stores have a product index or a search function. Super Stores also provide the most text information for each product of any store group from the Spiller and Lohse study (1997). Number of products on product pages is small with most stores displaying only one product per page. The corresponding page length is one screen page in most cases. Product selection and ordering is supported by a shopping cart metaphor. The numbers of extra appetizer and customer-care features such as feedback or access to sales representatives are also extensive. In that aspect, Super Stores are similar to magalogs (Morris-Lee 1993). Magalogs are a hybrid of catalogs and magazines that build relationships with consumers. To keep readers coming back again and again, magalogs use extensive appetizers to build customer involvement and retention. Some examples of Super Stores noted in the Spiller and Lohse (1997) study include L.L. Bean, Land's End, Spiegel, Online Sports, J.C. Penney, Shoppers Advantage and Service Merchandise.

Given the confidential nature of the dependent variables, monthly traffic (number of unique visits) and monthly sales in dollars, sampling was contingent upon the availability of data from a cybermall. As such, this survey is neither a random nor a systematic sample of all Super Stores. It does, however, provide a representative cross-section of online retail stores for August 1996. Annual 1996 sales at this cybermall exceeded $22 million. Up to 300,000 users per hour accessed the site during prime time. Approximately 50,000 users per hour accessed the site at other times. Service stores offering financial services or information for sale were not considered. Stores that had changed significantly in the three months prior to data collection were also excluded from the survey. All thirty-six interface features (Appendix 1) were measured for the resultant set of 28 online retail stores in August 1996.

Retail Store Attributes

Electronic shopping incorporates many of the same characteristics as "real" shopping. Thus, we identified attributes that shoppers consider when patronizing a retail store. Considerable research exists on the evaluation of department stores by consumers. Berry (1969) empirically identified a number of attributes using a mail survey. May (1974) emphasized the importance of the retail stores' image. Lindquist (1974) categorized store components into functional areas such as merchandise selection, price, store policies and store layout. His attribute list is a compilation from 26 researchers in this field. Lindquist distinguished store image components in functional qualities, such as the merchandise selection, price ranges, credit policy and the store layout, and psychological attributes that were associated with the customer's  feeling comfortable in the store. In our research, we only concentrated on functional qualities, as we did not survey actual customers. This earlier research concentrated on modeling customer store choice behavior. Since most subsequent research refers to the store attributes mentioned in Lindquist's summary, we adopted the store attributes identified by Lindquist.

These attributes are categorized into four groups: merchandise, service, promotion, and convenience. Merchandise variables measure product selection, assortment, quality, guarantees, and pricing. Service variables examine general service in the store and sales clerk service for merchandise return, credit policies, etc. Promotion variables record sales, advertising, and appetizer features that attract customers (e.g., a "What's new" section). Navigation variables include store layout and organization features. Arnold et al. (1977, 1983) extended the navigation attributes to include ease of navigating through the store and a fast checkout.

How do features from bricks and mortar retail stores relate to online retail stores? Table 1 presents analogies between real stores and online retail stores. Obviously some features like store atmosphere are difficult to measure and characterize in online retail stores. Other features like store promotions are less difficult to measure in online retail stores. While we do not offer a complete theoretical foundation for how online retail store features influence consumer behavior, we define objective online variables that map to key retail store variables identified by Lindquist and our previous Internet catalog classification (Spiller & Lohse, 1997). Appendix 1 summarizes all evaluation questions, the data collection method and the coding of the data. We now describe the measurement of these variables.

Merchandise

Beyond product count and amount of textual information for each product, we measured the number of hierarchical levels between the store entrance and the final product page. The number of hierarchical store levels serves as a proxy variable for the number of product categories. The number of product categories is not easy to define across different stores since stores offering different types of products will have very different levels of product categories. In one store, men's, women's and children's clothing might represent three high-level product categories; in the next case, it might be business attire, casual and sporting goods.

Service

Interactive service is an important aspect of online stores (Jarvenpaa & Todd, 1997). Company information as well as extra product information was measured in number of lines. Binary questions examined gift services; the use of a frequently asked questions (FAQ) section; the use of email, phone, or mailing lists; the presence of a customer feedback section, the collection of personal customer data, extra product information and help on product size selection. One important way to establish credibility of the business of the Internet is to provide information about the history of the company, store policies and other company information. We counted the number of lines of company information. For established companies with a reputable brand name this is not as important as for a new firm operating on the Web. In addition, we measured the average number of lines of text describing each product.
 
"Real" Store
Online Retail Store
Salesclerk service  Product descriptions, information pages, gift services, search function, sales clerk on the phone / email 
Store promotion  Special offers, on-line games and lotteries, links to other sites of interest, appetizer information 
Store window displays  Home page 
Store atmosphere  Interface consistency, store organization, interface and graphics quality 
Aisle products  Featured products on hierarchical levels of the store 
Store layout  Screen depth, browse and search functions, indices, image maps 
Number of floors in the store  Hierarchical levels of the store 
Number of store entrances and store outlets / branches  Number of links to a particular online retail store 
Checkout cashier  On-line shopping basket and/or order form 
Look and touch of the merchandise  Limited to image quality and description, potential for sound and video applications 
Number of people entering the store  Number of unique visits to the online retail store 
Sales per period  Sales per period 

Table 1 Analogies between real stores and online retail stores

Promotion

The cybermall featured advertisements for products on a welcome screen after logon. We obtained the number of hours a store's products were on this welcome screen per month. It ranged from one to thirty hours. The number of hours per month individual stores were featured in other ads also was obtained from the cybermall. There were two types of featured ads designated A & B. Ad type "A" consisted of a store icon button with ad copy. Ad type "B" consisted of just a store icon button.

We recorded whether the store offered products at a discount in comparison to physical retail stores. The average discount was calculated from actual product prices and suggested retail prices displayed with the product description. We assumed that a leading serial position in the list would increase probability of store selection. The relative serial position in the cybermall alphabetic listing was calculated by dividing the absolute position by the total number of items in the lists. Exposing customers to more featured products during their way through the store should impact sales. We counted the number of featured products on the home page and along the departmental navigation path of the store. The latter can be seen as analogous to the aisle products in a real store.

Navigation

Shopping effort served as a guide to select the navigation variables. We measured features that influenced the amount of effort (time, number of steps, etc.) required to browse and navigate the online retail store. Navigation features include the number of hyperlinks into each store as well as links between products (which were surprisingly rare). Each link is like an additional entrance to the store. Other navigation features include the number of products on the end product pages, the number of buttons used to browse the store, and the type of checkout and order process. We recorded the average number of scroll list items and compared these numbers with each list boxes' size. If the average number of items in the list was greater than the box size in lines, customers had to scroll this list to see all items. Some stores provided the products' prices in the lists as well as on the final product screens. We considered this very helpful for comparing different items in a list and coded it as a binary variable.

The number of different shopping modes facilitates market segmentation. For example, a wine store can arrange the virtual store by vineyard (Mondavi, Phelps, etc.) or by variety (Pinot Noir, Merlot, etc.) or by price. Each different mode was counted. Shopping modes included search functions and product indices.

The basic product list consisted of a nine-item scroll list with two buttons: one to OPEN list items and one to display MORE items (Figure 1). An improved version of this list displays either an image or text description or both with the product name. Customers can select this product directly, without having to navigate further levels. This makes navigation throughout the store much easier because windows do not need to be closed and re-opened.

Figure 1: Scrolling menu showing a basic product list.

Interface Variables

Consistency is a principle of computer interfaces. We measured whether consistent menu bars were in place on all pages. Color attracts attention and helps organize information in a computer interface. We surveyed the background color, texture, or pattern of the home page. Help functions are most often related to helping users recover from errors or find a particular function in the documentation. We analyzed whether stores offered any initial help for customers to shop their store. This included help information about the store's navigation or the use of ordering features like a shopping cart function. Homogeneity, coded on three levels, records whether lists contained different types of items such as products, text screens or navigation functions at the same time. Next, we coded information about the use of images and image size.

Dependent Variables

For the data collection period, we recorded monthly number of sessions (unique visits based on login data) and monthly sales in dollars for each store surveyed (N=28). The mall requirement for customers to login facilitated our measurement of unique visits to the mall for August 1996. The number of sessions measures a stores' general attractiveness, and measures the influence of promotional features like ads on the main screen. All 36 interface features were measured prior to the verification of actual monthly sales data for August 1996. Monthly store sales averaged nearly $64,000 and traffic averaged 54,893 unique visits per month. Monthly sales data reflect actual purchases less returns, fraud, etc. We did not consider the number of purchases, or the conversion rate (measured as purchases per unit of traffic) as dependent variables, as those are highly affected by fraud. Fraudulent purchases were easy to back out of the monthly sales data whereas traffic associated with fraudulent purchases was not recorded.

Data coding

The second author coded all store variables using a standardized checklist over a three-week period in August, 1996. Given the changing nature of the sites, the reliability of the coding was tested post hoc using a random sample of 42 stores. The second author and a research assistant coded all measures for each of these 42 stores. Using Cronbach alpha, inter-rater reliability over this sample was 0.818.

Given time limitations, we did not count individual words associated with each interface feature. Instead, we counted the total number of full text lines that contained related information. Full text lines were adjusted to reflect different font sizes and frames. A post hoc analysis of a random sample of 14 stores surveyed found a high correlation (r = 0.98) between line count and word count. Thus, line count is a reasonable approximation for word count.

Regression Diagnostics

Because regression models with too many variables and too few observations lead to potential collinearity problems, we reduced the number of variables in the models using stepwise regressions (Table 1). Table 2 lists 13 independent variables eventually used in both a traffic model that used number of visits per month as the dependent variable, and a sales model that used monthly dollar sales as the dependent variable.

Collinearity among the independent variables causes the model to be very unstable when deleting or adding variables to the model. If two or more variables are completely collinear (i.e., one variable can be written as a linear combination of the others), the model is not full rank and regression coefficients can not be calculated. A measure for collinearity in multiple regression models is the variance inflation factor, VIFi, which should be smaller than 10 for all variables (Mason & Perreault, 1991). This criterion was easily met for all variables. Another measure of collinearity, the condition index, was below the critical value of 30 (Belsley, Kuh, & Welsch, 1980; p. 105). Plotting residuals versus predicted sales and visits did not reveal any patterns in the residuals. Also, the White Test for heteroskedasticity (White, 1980) let us maintain the null hypothesis that errors are homoskedastic and independent from the regressors (prob>chi-square was 0.85 for the traffic model and 0.42 for the sales model).

A number of diagnostics identify the influence of a single outlier on regression coefficients (Fox, 1991). Discrepancy is measured using the regression's partial residuals. Residuals reflect the amount of discrepancy between observed and predicted values that is still present after having fitted the regression (Kleinbaum, Kupper, & Muller, 1988, p. 185). A plot of the residuals for each independent variable and both dependent variables found no outliers. Leverage of a single observation on the parameter estimates can be measured by examining the impact of deleting each observation in turn (Belsey et al. 1980, p. 12). The standardized difference between the two estimates with and without individual observations is an indicator for the leverage of one observation on the coefficients. This indicator, called Dfbetasij (Belsey et al. 1980, p. 13), also did not identify any outliers.

The quality of our estimates varies across the variables. The standard error, which is a measure for confidence, was relatively high due to the small number of stores (observations=28) in our survey. In order to overcome these limitations, we would need to survey more stores with a greater variance in the store interface design features.

It is also important to note that the statistical model does not detect causalities. The model reveals correlations that might stem from a causal relationship, but correlations might also be completely accidental. We do not know whether advertising promotions caused more traffic and higher sales. We can only observe from our specific data that more promotion was associated with more traffic and higher sales. A causal model would require a detailed theory about all the different factors influencing these measures.

Results

The summary statistics for both models are highly significant (Table 1). The overall F-test is significant for both models at a<0.0001. R2 values measure the percentage of total variance in the data that can be explained by each independent variable. The usual R2  value can only improve by adding more variables to the model, even when their contribution is very small or accidental. The adjusted R2  value takes the number of variables in the model into account. Adding more variables with small contributions will therefore worsen the adjusted R2  values. Hence adjusted R2  is a less biased measure for the variance explained by the model and we use it in our interpretations. The variables in the traffic model explain 88% of all variance in the store traffic data; the sales model explains 76% of the variance in dollar sales data.
 
Model DF F Value Prob > F adjusted R2
Traffic  13  18.260  0.0001  0.8826 
Sales  13  14.648  0.0001  0.7629 

Table 1 Summary regression statistics for the models

Table 2 summarizes the variables in the regression analysis. Descriptive statistics show the minimum, maximum, mean, and standard deviation for each variable in the model. The column titled standardized estimate shows the beta weights calculated for each model. A one standard deviation change in one of the independent variables produces an Xi standard deviation change in the dependent variable. By measuring the relationship of all of the independent variables in standardized units, the relative impact on the dependent variable can be compared directly. To protect the confidentiality of these data, the regression estimates in dollars per month and visits per month are not shown. The columns headed Prob>|t| show the significance of individual variables in the regression. The variance explained by each variable is shown in the column labeled R2 .

A. Additional products in the store attract more traffic

Each additional product in the store yields additional store traffic. The variable explains 17% of all variance in the number of visits data and is significant (a<0.0001). Apparently, shoppers have an idea or some experience of which products they might find in each store. If they are looking for a particular product, the data suggest that consumers prefer larger stores to smaller ones. Indeed Jarvenpaa and Todd (1997) found that one-third of the shoppers expressed disappointment with the limited product offerings within any one store. Interestingly, the store size did not have a significant effect on dollar sales. It seems that more products result in more traffic to the store, but the additional traffic did not result in higher sales. Big stores are not as effective as small stores in converting traffic into sales. One reason for this outcome is that consumers may not find the products they are looking for in larger stores. Improved search functions and navigation features might overcome this low conversion to sales.
 
 
Traffic Model Sales Model
Variable (Appendix number) Min Max Mean Std Dev Std Est Prob>|t| R2 Std Est Prob>|t| R2
A  Number of products (1)  56  22,000  1,264  4,202  1.181 0.0001 0.17 -0.2682  0.1972  n.s. 
B  FAQ section available (5)  0  1  0.14  0.36  1.5548 0.0001 0.45 0.1876  0.3785  n.s. 
C  Feedback section (9)  0  1  0.11  0.31  -0.7292 0.0001 0.09 0.3674 0.0348 0.01
D  List + button + picture (27)  0  1  0.07  0.26  0.4966 0.0044 0.04 0.8369 0.0001 0.58
E  Lists with pictures (27)  0  1  0.21  0.42  0.2872 0.0028 0.04 0.1728  0.0702  n.s. 
F  Lists with buttons (27)  0  1  0.07  0.26  -0.0957  0.591  n.s.  0.5059 0.0201 0.03
G  Store "entrances" (22)  1  7  2.61  1.42  0.3535 0.0025 0.07 0.4122 0.0017 0.10
H  Shopping modes (21)  1  5  2.71  1.15  -0.3435 0.0206 0.01 -0.139  0.3573  n.s. 
I  Appetizers (11)  0  43  10.04  12.44  -0.2548  0.0531  n.s.  -0.0653  0.6327  n.s. 
J  Promotion hours (15)  1  30  9.89  9.07  0.1702 0.0339 0.01 0.2235 0.0146 0.04
K  No. featured products (18)  0  5  1.21  1.17  -0.1613  0.1091  n.s.  0.1146  0.2915  n.s. 
L  Number of levels (2)  2  7  4.36  1.37  -0.1461  0.1969  n.s.  -0.0321  0.7925  n.s. 
M  Consistent menu bars (30)  0  1  0.18  0.39  -0.1978  0.2611  n.s.  -0.3234  0.1062  n.s. 
Traffic  54,893  37,693 
Sales  $63,946  $95,419 

Table 2. Variables used in the regression. NOTE: n.s. means the variable was not significant. Given the confidential nature of the sales and traffic data, parameter estimates (in dollars/month or traffic/month) are not included.

B. Featuring a FAQ section in the store is associated with more traffic

The second variable records whether the store features a frequently asked question (FAQ) section about the company or its products. The variable is significant in the traffic model. This suggests that, in general, stores having a FAQ section generate more visits per month, compared to those stores without this section. However, it is important to emphasize that we do not talk about causal relationships. A possible explanation for this outcome is that the bigger stores received so many email messages per day that they felt that implementing a FAQ section would be helpful in reducing the cost of this interaction. With this interpretation, the FAQ variable is more of a descriptive indicator for the store's traffic number. In this sense, the FAQ feature is a result of the store's size, not a variable that caused more traffic. The variable had no significant effect on sales.

C. Providing a feedback section for the customers is associated with lower traffic and higher sales

The feedback parameter suggests that having this feature decreased traffic but increased dollar sales. Providing a way for customers to comment on catalog services and interface features is considered to be a method for improving the interface (Fox, 1995; Sandberg, 1996). We speculate that FAQ and feedback are related. Initially stores with feedback sections may have provided fast and friendly customer service via an email feedback button. As traffic increased, online stores may have added an FAQ section to cope with the times demands on staff to answer email. Assuming that established feedback sections already resulted in improved services and interfaces, this feature might explain higher sales. To test this speculation, we would need to measure the quality of the feedback to the consumer as well as the timeliness of the response. Unfortunately, these data were not available to us for this study. Then again having an email feature does not mean stores will reply to feedback sent electronically by consumers. Poor response time to email or failure to reply is a major problem (Ulfelder, 1980).

D,E,F. Improved product lists have a tremendous effect on sales

We expected that any improvement over the cybermall's basic product list window would yield better sales since shoppers could navigate the store much easier and are exposed to more featured products on their way through the store. All product list improvements had a significant impact on either monthly dollar sales or monthly store traffic. Product lists account for 61% of the variance in monthly sales and over 7% of the variation in monthly store traffic. These results suggest that improving the browsing and navigation capabilities of stores and especially product lists can generate significantly higher traffic and sales. The basic product list consisted of a scrolling menu listing products (Figure 1). Additional product navigation list information such as price, a thumbnail image, and a longer descriptive product name had the largest impact on sales. We speculate that this facilitated purchase decision making at the point consumers initially viewed the product.

G. A greater number of "store entrances" yields additional visits and sales

Additional links from other locations in the cybermall can be seen as additional "store entrances" because they offer multiple ways to access a store's home page. The regression found that each additional listing was associated with additional visits and sales. The variable explains 7% of the variance in traffic data and 10% of the variance in dollar sales data. The significance of this variable suggests that shoppers frequently used other entrances to locate a particular store. Of course, there is an upper limit to the number of links into the store. This relationship can not be extrapolated beyond the maximum number of "entrances" in our data set (seven).

H. The number of shopping modes has no impact on sales

Additional shopping modes should enhance the navigation capabilities of the interface and also segment customers who, for example, prefer to shop by brand or by price. The variable had no significant effect on dollar sales and a negative effect on the number of visits.

We can only hypothesize that this might be due to our data coding. The variable only codes the number of different shopping modes, not their quality. A sophisticated search function is considered the same as a very simple list sorted by price. Many of the smaller cybermall stores feature several simple modes, like lists by price or alphabetically, but none of them offered more advanced shopping modes like a search function. Still, a store with many simple shopping modes scores higher than a better store with fewer but more sophisticated modes. It might have been more accurate to weigh a search function higher than an alphabetical list. On the other hand, we also defined binary variables coding a search function or a A-Z list and did not find a significant effect of these variables. As mentioned before, the likelihood for type II errors, rejecting true hypothesizes, is relatively high due to the small number of stores in the survey. Online retail stores have significantly improved personalization features that improve consumer loyalty. Thus, one must be cautious about extrapolating these findings to more current online retail stores.

I. Appetizer information has no significant effect on traffic or sales

Nearly all the stores provided some information about the company, featured additional information or appetizers, or offered additional services. We hypothesize that the amount of these services would positively impact sales and visits. We coded whether the store provided any additional information over the basic product catalog, like information on the usage of its products, on health, or other issues. The variable was not significant in either model. Either consumers do not need and search for this kind of information, or they do but this does not alter the probability of purchasing anything. Whether consumers use appetizer information screens can be determined by analyzing browsers' navigation paths in server log file data.

J. Promotion on the Cybermall entrance screen generates traffic and sales

Each hour of promotion on the cybermall entrance screen resulted in additional visits and generated additional sales for the store. The variable is significant in both models at the level a<0.05. Four percent of the total variance in dollar sales and 1.4% of the total variance in store traffic can be explained by this variable.

While these ads seem to drive sales, the conversion from the ads to store traffic is very low. The low conversion to store traffic is probably a function of the end product page design. Promotional ads directed customers to an individual product. Often, the remainder of the store was not accessible from these individual screens. There was no navigation path available to navigate from any specific product screen into the store to see some other products or the store's home page. Figure 2 shows an end product page with navigation buttons to browse other areas of the store. The browse forward and browse back buttons allow customers to navigate from one end product page to another. Without such buttons, the consumer can not look at merchandise adjacent to this promotion item nor can he or she access information about the company's reputation, returns policies, etc.

Customers either purchase the promotion product and enter the store afterwards to search for some additional products, or they do not purchase the product and never enter the store. In this sense, these ads provide a reminder or motivate the customer to patronize the store. Promotional activities for particular products in real stores always aim to give shoppers an incentive to patronize the store and to buy some other products as well. The cybermall home screen promotion does not capitalize on these effects very well because there is no direct navigation path available from end product pages to browse other products in the store.

K. The number of featured products along the departmental navigation path has no significant effect

A higher number of featured products along the usual path from the home page to end product pages should have a positive effect on sales. These featured products can be seen as the aisle products in a retail store.

Figure 2: End product page with consistent navigation buttons and icons.

We did not find a significant effect of the number of featured products in the catalogs on sales. The variable also was not significant in the traffic model. We did not study the copy quality of these featured ads. Assuming that on-line shoppers are merely attracted by featured products, it might be concluded that most on-line shoppers are actively searching for particular products in the product lists, no matter how many "advertised products" they see on their way through the store. This, however, is in conflict with findings from a cybermall focus group survey (Jarvenpaa & Todd, 1997). Most users in this survey stated that they only dropped into the stores to browse whether or not they found anything interesting. Very few declared they were looking for something particular. It would be interesting to look at actual purchases in this context. Featured products might have a great impact on customers but at the same time cannibalize on other products in the store, leaving total sales almost unchanged. Unfortunately, we could only look at aggregate sales data in this survey. But this is certainly a promising area for future research.

L. The number of levels between home page and end product pages has no significant effect on visits and sales

The number of levels between the home page and end product pages should have a negative effect on sales because shoppers will have difficulty finding products. We assumed that too many levels between home page and end products would be confusing for shoppers and would reduce buying. The variable was not significant in either model. We tested different level-definitions and eventually used the average number of levels between store entrance and end products in the models. These new parameter estimates for the variable were not significant either. In order to test this hypothesis more accurately, more similar stores in terms of size (but differing in their level number) should be evaluated.

M. Consistent menu bars have no significant effect in the models

The variable, recording whether the stores featured consistent menu bars on the pages, was not significant in either model. A consistent menu bar meant that every product page in the store had a consistent set of store navigation icons. For example, these might include search the store, move to any other department, top of store, etc. Interface consistency is generally considered to be important from a human-computer interaction perspective (Nielsen, 1993; Polson, 1988). However, it is very hard to code consistency into variables. Studying additional variables describing the concept of consistency, such as the menu organization, wording and consistent use of colors and icons might yield a different result. It may also be the case that in the context of all the other factors influencing traffic and sales, consistent menu bars had a very small non-significant impact.

Limitations of the Regression Analysis

The study has several important limitations. First of all, online retail Web sites have evolved tremendously since our August 1996 data collection. One must be careful in extrapolating our findings to present online retail stores. Second, having 28 stores in the sample limits the overall confidence in the parameter estimates as well as increases the probability for type II errors in the hypothesis testing. A larger number of stores in the sample is necessary to overcome this problem. Third, The regression models do not distinguish between stores on the basis of product types or brands. Implicitly, we assume that the effects we found do not differ for stores selling flowers and stores selling computers. Fourth, several non-significant findings were surprising. The finding that consistent menu bars had no effect on traffic and sales might be heresy to the human computer interaction community; however, it is important to note that we did not say consistent menu bars were not important. Our data only imply that consistent menu bars did not influence traffic and sales as much as other online retail store features. Likewise, we found no effect for multiple shopping modes. Marketing pundits will cringe at this finding. One-to-one target marketing is one of the most touted advantages of electronic shopping environments. Does this mean that 1:1 target marketing is not a good thing? Of course not! Anecdotal evidence abounds on the importance of market segmentation and target marketing in electronic shopping environments (Hoffman et al., 1995). And finally, a non-disclosure agreement prevented earlier publication of these data and findings. Obviously, it would be ideal to have 1999 data implementing design features from current Web sites. Realistically, online stores will not disclose data that provides valuable insights for improving the design of the online retail store of their competition.

Implications for Store Design

Navigation

Many Internet retail stores do not enable shoppers to compare and browse products easily (Baty & Lee, 1995). With some products lists, shoppers open a particular product, go back to the product list, and open another product screen when they want to compare different products in sequence. Also, if consumers arrive at end product screens via promotion advertisements, there is no navigation path available to navigate from this specific product screen into the store to see some other products.

The regression suggests that improving navigation of product lists, product search, and increasing the use of hyperlinks within a store are the primary areas of opportunity. Product list navigation explained 61% of the variance in monthly sales. Additional product list information such as price, a thumbnail image, and a longer descriptive product name had the largest impact on sales. We speculate that this facilitates purchase decision-making at the point consumers initially view the product. As anecdotal support of this finding, we view the streamlined product list navigation using a one-click-to-purchase approach of Amazon.Com and 1-800-Flowers [http://www.1800flowers.com] as recognition that every additional mouse click reduces the possibility of a purchase (Lohse & Spiller, 1998).

Despite the advances in online retail stores since 1996, recent studies summarized by Nielsen in 1998 found similar navigation problems [http://www.useit.com/alertbox/981018.html]. For example, even when starting on the correct home page, users only found information they were searching for 42% of the time, 62% of shoppers gave up looking for the item they wanted to buy online, 40% of first-time purchasers failed to make a repeat purchase because of a negative experience, and 50% of sales wer lost because people can't find products on the web site. A 1999 report by Mark Hurst [http://www.goodreports.com/] also found similar problems with web site navigation. In user tests, 39% of shoppers failed in their buying attempts because the site was too difficult to navigate and 56% of search attempts failed.

In the survey by Spiller and Lohse, 95% of the stores did not have hyperlinks among related products (1998). Hyperlinks aid the discovery of new and useful information and allow the user to obtain more detail as needed. The links should be as context-specific as possible. The sale of a Walkman CD player should have a link to buy batteries. A shirt can be linked to pants and a matching sweater. Such links reduce the effort of browsing by directing customers to related items. Hyperlinks are like an additional store entrance that increases traffic into the store. The number of links into a store explained 10% of the variance on monthly sales. Online retailers should not waste opportunities to link to related areas!

It is difficult for designers to anticipate a user's navigation path. Since everyone will not come in the front door, every web page must have consistent navigation links (site map, index, etc.) to move around on the site. Complex URLs using a foreign naming convention making them nearly impossible to use as site navigation aids.

Promotion

Promotion on the cybermall entrance screen increased sales for stores. Promoting stores in the cybermall entrance increased traffic only by a small amount. Some types of promotions had no significant effect in the regression. We did not study the impact of advertisements placed in the remaining cybermall content that link to stores. No data about these ads were available. The impact of promotions should be studied in more detail. However, we did find that offering additional store entrances in the form of additional links positively impacts store traffic. Additional ads throughout the cybermall content that represent extra "store entrances" will improve traffic into stores.

Providing additional appetizers or customer services to attract surfers had no effect on traffic or sales in the regression. This research suggests that the provision of this additional information should not be a design priority.

Store Size

Larger stores attract more traffic. But as we have also seen, this traffic does not necessarily translate into higher sales. One reason for this outcome is that consumers may not find the products they are looking for in larger stores. Improved search functions or other shopping modes should overcome this low conversion to sales. However, the regression did not reveal any effect of the shopping modes we surveyed. Since only a few stores offered customers multiple modes of shopping, we assume that the sample size was not sufficient to show any effect of these shopping modes. By linking sales data to users' ZIP codes and demographics, future research could examine whether particular customer segments prefer shops that allow them to shop by, for example, different modes - by price rather than by product.

Store size is also reflected in the number of hierarchical levels between the store entrance and the product pages. In some stores, the consumer had to pass seven screens before arriving at the final product screen. The statistical analysis of the data did not reveal a negative effect of too many of these levels. Either the sample size was again too small to show an effect, or consumers do not bother to navigate several screens to arrive at the products sought.

Store Presentation

We did not find an effect of "store presentation" variables, such as image sizes, background patterns or the number of buttons on the storefront. Consumers want to find products quickly and effortlessly. It appears that no amount of "sparkle" in the presentation of products can overcome a site design with poor navigation features.

While this research analyzes and quantifies the impact of different features of the cybermall interface design on traffic and sales, it does not provide any detail about converting traffic into sales. Analyzing clickstream and browsing navigation data could provide an understanding of how to increase profitability of on-line markets.

Summary

In August 1996, monthly store traffic, dollar sales and a set of 36 interface features were measured for a sample of 28 online retail stores. Using a stepwise regression approach, we identified the set of interface design features that had the largest influence on Internet store traffic and monthly sales. For the monthly sales data, product list navigation features that reduce the time to purchase products online accounted for 61% of the variance in monthly sales. Other factors explaining the variance in monthly sales included: number of hyperlinks into the store (10%), hours of promotion (4%) and customer service feedback (1%). For the monthly store traffic data, having a FAQ or customer feedback section explained 54% of the variance. Again, this does not imply causality. We speculate that stores added a FAQ section to cope with high traffic generated by the customer feedback section (e.g., email to sales representative). Other factors explaining the variance in monthly traffic included: number of products in the store (17%), product list navigation (8%), number of hyperlinks into the store (7%), hours of promotion (1%) and number of shopping modes (1%). One final caveat regarding this data analysis is that other variables (Appendix A) that were not significant in the regression analysis may still be important in the design of online retail stores. Our results only imply that non-significant variables did not influence traffic and sales as much as other online retail store features.

While we did not specify model parameters á priori, these findings suggest that information search costs are a major determinant of both store traffic and sales. Navigation features that make it easier for customers to find products are essential for a successful online retail store. After all, if customers can not find what they want, they can not buy the product. Empirical research on information search behavior has a long tradition in marketing (Payne, Bettman, and Johnson, 1993). More recently, Hoque and Lohse (1999) provide a theoretical basis for predicting how subtle changes in the user interface design influence information search costs and demonstrate that consumer choice is contingent upon information search costs imposed by the media. These studies demonstrate the importance of information search costs in the design of a web site for selling consumer products or services.

By bringing computing to the masses, the Web heightens the awareness of user interface design to ordinary citizens. Store features in the bricks and mortar world are being implemented via the computer interface in online retail stores. A help button on the home page of the Web shopping site replaces the sales clerk's friendly advice and service. The familiar layout of the physical store becomes a maze of pull-down menus, product indices and search features. Now more than ever, the success of online stores depends upon how well how consumers interact with the computer interface to the store. Clearly, the user interface is an essential link between the customer and the retail store in Web-based shopping environments. It is our belief that the growth of Internet retail sales will depend, at least partially, on these interface design issues.

REFERENCES

Arnold, S. J., Ma, S., & Tigert, D. J. (1977). A comparative analysis of determinant attributes in retail store selection. Advances in Consumer Research, 5, 663-667.

Arnold, S. J., Tae, H. O., & Tigert, D. J. (1983). Determinant attributes in retail patronage: Seasonal, temporal, regional, and international comparisons. Journal of Marketing Research, 20, 149-157.

Baty, J. B., II,  & Lee, R. M. (1995). Intershop: Enhancing the vendor/customer dialectic in electronic shopping. Journal of Management Information Systems, 11(4), 9-31.

Belsley, D. A., Kuh, E., & Welsch, R. E., (1980). Regression diagnostics: Identifying influential data and sources of collinearity. New York, NY: John Wiley.

Berry, L. J. (1969). The components of department store image: A theoretical and empirical analysis. Journal of Retailing, 45, 3-20.

Fox, B., (1995). Retailing on the Internet: Seeking truth beyond the hype. Chain Store Age, 71, 33-72.

Fox, J. (1991). Regression diagnostics.  Sage Publications, Inc., Newbury Park.

Online Shopping - Virtually impossible, (1996, September), Glamour, p. 232.

Hoffman, D. L., Novak, T.  P., &  Chatterjee, P. (1995). Commercial Scenarios for the Web: Opportunities and Challenges. Journal of Computer-Mediated Communication, [Online] 1(3), Special Issue on Electronic Commerce, 23-45. Available: http://www.ascusc.org/jcmc/vol1/issue3/hoffman.html

Hoque, A. Y. & Lohse, G. L. (1999). Designing interfaces for electronic commerce: Predicting how information search costs influence consumer choice. Journal of Marketing Research, 36, 387-394.

Jarvenpaa, S. L., & Todd, P. A. (1997). Consumer reactions to electronic shopping on the World Wide Web. International Journal of Electronic Commerce,  1(2), 59-88.

Kleinbaum, D. G., Kupper, L. L., & Muller, K. E. (1988). Applied regression analysis and other multivariable methods, Boston, MA: PWS-Kent Publishing Company.

Lindquist, J. D.  (1974).  Meaning of image.  Journal of Retailing, 50(4), 29-38.

Lohse, G. L., & Spiller, P. (1998). Electronic shopping: The effect of customer interfaces on traffic and sales. Communications of the ACM, 41(7), 81-87.

Mason, C.H., & Perreault, W. D., Jr. (1991). Collinearity, Power and interpretation of multiple regression analysis.  Journal of Marketing Research, 28, 268-280.

May, E. G. (1974).  Practical applications of recent retail image research. Journal of Retailing, 50(4), 5-20.

Morris-Lee, J. (1993). Push-pull marketing with magalogs. Direct Marketing, 56(2), pp. 23-26, 47.

Nielsen, J. (1993). Usability Engineering. Cambridge, MA: Academic Press.

Payne, J. W., Bettman, J. R., & Johnson, E. J. (1993). The adaptive decision maker. Cambridge, England: Cambridge University Press.

Polson, P. G. (1988). The consequences of consistent and inconsistent user interfaces, In R. Guindon (Ed.). Cognitive science and its application for human-computer interaction (pp. 59-108).  Hillsdale, NJ: Lawrence Erlbaum.

Ridgon, J. (1996, June 17). Caught in the Web. Wall Street Journal, R14.

Sandberg, J. (1996, June 17). Making the sale. Wall Street Journal, R6.

Spiller, P., & Lohse, G. L. (1997). A classification of Internet retail catalogs. International Journal of Electronic Commerce, 2(2), 29-56.

Ulfelder, S. (1997, September 15).Consumers www.wait for e-mail responses. Computerworld, 31(37), pp. 1, 96-98.

White, H. (1980). A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica, 48, 817-838.
 
 
 

Appendix A: Variables used in the cybermall store survey

 
  Data Coding Comments
Merchandise
  1. Number of different products in the store 
 Actual number  The number was counted and summed for all product categories in the on-line store. 
  • Number of hierarchies between home page and end product page 
    		•  Average number  The straightest path through the general transition network of the catalog was evaluated for 6 different products. 
    1. Information about shipping, ordering, product quality, and returns 
    		 Number of lines  The total number of lines providing information about guarantees and the ordering process were counted. 
    Service
    1. Gift services 
    		 0: no 

    1: yes 

    		 Wrapping, reminders, gift certificates 
  • FAQ on product related questions 
    		•  0: no 

    1: yes 

    		 Did the store provide a frequently asked questions section on product related questions or the company? 
    1. Company information, mission statement, history, policies 
    		 number of lines  The total number of lines providing company information were counted. 
    1. Lines of text describing each product 
    		 Average number of lines for 6 end product pages  Any full line on the screen was counted as one line, single items of information (e.g. "waterproofed") were counted as 0.5 lines. 
  • Access to sales or service representative 
    		•  email (0,1)

    phone (0,1)

    feedback section (0,1)

    electronic mailing list (0,1) 

    		 Did the on-line store offer any way to contact a sales representative, by email, phone or fax? 
  • Feedback section 
    		•  0: no 

    1: yes 

    		 Did the retailer ask for feedback (email comments (mailto), blank email form, email questionnaire, phone)? 
  • Extra get-to-know-the-customer questions 
    		•  0: no

    1: yes 

    		 Any personal questions about the buyer on the order form were rated 1.
  • Extra product information 
    		•  Total number of types (0-43)  Any information on care, maintenance, and use of products, etc. was counted.
  • Help on product-size selection 
    		•  0: no 

    1: yes 

    		 Stores presenting tables or figures with product sizes were rated 1.

     
     
    Promotion
    1. Number of hours on the cybermall welcome screen 
    		 number of hours per month

    range: 1-30 hours 

    		 The welcome screen ad was changed every 15 minutes 
  • % of discount if regular price is also given 
    		•  percentage  Discount was calculated as mean of ten products. 
  • Hours in the cybermall featured ads per month? 
    		•  hours/month as ad type A

    hours/month as ad type B 

    		 Hours per month reported by cybermall marketing department 
  • On what position is it in the cybermall alphabetic A-Z store listing? 
    		•  Alphabetic position was determined 8/18/96  The total number of stores was displayed over 6.5 screens 
    1. On what position is it in the other store listings? 
    		 Position was determined 8/18/96  If the store was listed in more than one other listing, the lowest number (=best position) was recorded. 
  • Number of featured products on the home page 
    		•  number of products  All featured products with an image were counted. 
    1. Total number of featured products "end-of-aisle" 
    		 number of products  All featured products with an image were counted. 
  • "What’s new" section 
    		•  0: no 

    1: yes 

    		 A section introducing new products, catalog features or news was rated 1.

     
     
    Navigation
    1. Number and type of different shopping modes 
    		 number of modes  Were there different strategies the customer can use to find a product (store organization by product category, gender, price level, etc.)? 
  • How many hyperlinks into the store? 
    		•  number of links  Only counted links from within the cybermall. 
  • Number of products on end product page 
    		•  average number  The average of ten observations for each level. 
  • Number of scroll listings 
    		•  0: none, all listings are shorter than the listing’s box

    1: a few listings are longer

    2: lists on one level are longer

    3: lists on two levels are longer

    4: lists on three levels are longer 

    		 This variable is an indicator for the number of lists that are longer than the product list box, i.e. those lists the customer has to scroll down to see all items listed. 
  • Links between related product pages 
    		•  0: no 

    1: yes 

    		 Links between related products or links between products and additional information (magazine pages, games, etc.) was rated 1 
  • Is product price shown in the product list? 
    		•  0: no

    1: yes 

    		  
  • Product listing

    • three variables coded (0,1)     		 List product name? 

    List name and description? 

    List name, description and 
    picture? 

    		 Product navigation lists afforded different types of information to support purchase decisions. 
  • Number of home page buttons 
    		•  number of buttons  All buttons (navigation, sound, featured product) are counted. 
  • Product ordering and selection 
    		•  0: off-line

    1: on-line form to fill 
    manually 

    2: on-line clickable form 

    3: shopping cart 

    		 1 refers to an on-line form that has to be filled out manually; 2 was a form consisting of pull-down menus listing all product options. An on-line shopping cart only requires the shopper to hit an order button on end product pages. 

     
     
    Interface Variables
  • Menu bars on all pages 
    		•  0: no 

    1: yes, text

    2: yes, image overlays 

    		 Menu bars had a home page button and at least one further navigation button. The buttons had to be placed consistently on all pages of one level. Buttons using image overlays were considered more sophisticated. 
  • Background color or pattern 
    		•  0: no color, white /gray 

    1: one color 

    2: pattern or texture

    3: picture or image 

    		 In case of changing background colors on different pages, the modal level was recorded. 
  • Help on interface usage 
    		•  0: no

    1: yes 

    		 Pages on shopping cart usage, product list navigation, etc. counted. 
  • Homogeneity (list does not contain products and text-pages) of listings? 
    		•  0: not at all

    1: product listings are homogeneous

    2: all listings are homogeneous 

    		 A listing is homogeneous if it only comprises one type of item, like products, folders, or text pages. 
  • Image area on home page excluding product images 
    		•  measured as a percentage of the total visible screen area (549 square cm)  Image size was determined by measuring width and height of all home page images, excluding buttons. 
  • Image area on "end product pages" 
    		•  mean of 6 pictures; measured as a percentage of the total visible screen area  Image size was determined by measuring width and height of an average end product image. 
  • Use of images 
    		•  0: none 

    1: to show products only 

    2: image overlay and product display 

    		 Did the catalog use images for product display only, or for graphical buttons, too? 

    Acknowledgment

    We are grateful to Dr. Eric Johnson for his helpful comments and suggestions on earlier drafts of this manuscript.

    About the Authors

    Gerald L. Lohse is the Research Director of the Wharton Forum on Electronic Commerce and an Adjunct Professor of Marketing. His other current projects include evaluating the effectiveness of Internet and traditional Yellow Pages advertising and personalized on-line retail catalogs that dynamically adapt to consumer preferences. He has published over 30 articles in leading national and international journals on these topics. His comments have appeared in the Wall Street Journal, ComputerWorld, Interactive Week, Silicon Week, and National Public Radio and he was a member on the Digital Economy panel for the United States Department of Commerce. He is an advisor to the Innovation Factory, an Internet venture capital firm and Practical Reasoning, Inc., a text mining start-up company. In addition, Dr. Lohse has consulted for NASA, IBM, the Jet Propulsion Laboratories, GTE, America Online, Music Boulevard, R.R. Donnelley, and others.
    Address: Wharton Forum on Electronic Commerce, The Wharton School, University of Pennsylvania.  215.573.7345 phone 215.573-8817 fax

    Peter Spiller was a visiting scholar at the Wharton School of the University of Pennsylvania, completing his master of industrial engineering from the Technical University of Berlin, Germany. He studied the implications of Internet marketing and electronic commerce. After finishing his thesis in spring 1997, he became a consultant with McKinsey and Co. in Duesseldorf, Germany.
    Address: McKinsey & Company, Inc., Koenigsallee 60C, 40027 Duesseldorf, Germany  0049 211 1364 395