Is it ever beneficial for competitors to share data with each other? Professor Subodha Kumar throws light on how a firm can use its competitor’s data in such a manner that it is beneficial for both.
The value of e-commerce has increased dramatically over the past decade. In 2013, the U.S. retail e-commerce sales grew by 16.9% to reach $263.3 billion, and are expected to grow at a compound annual growth rate of 9% to reach $370 billion by 2017. The U.S. retail e-commerce sales accounted for 5.8% of the total retail sales in 2013. Further, online retail sales in Europe are anticipated to grow at a compounded annual growth rate of 11% and reach 191 billion by 2017. The unprecedented growth of e-business has generated vast amounts of customer data; an extremely valuable resource for firms to grow their business. Recent big data initiatives employed by firms are typically targeted towards collecting more data about customers and analysing them to retain old customers and create new markets.
As there are multiple sources of data, so are there multiple techniques for analysing it. These techniques range from common business intelligence based on reporting tools (primarily techniques of the past) to business analytics based techniques for prediction, e.g., recommender systems. Such systems use insights from data to suggest appropriate products to customers (firms that use recommender systems are hereafter referred to as personalising firms). Some popular examples of personalising firms include Amazon.com, Target, Costco, Home Depot. In 2007, 41% of electronic retailers were found to provide personalised services, a value that is estimated to increase over time.
Impact of Recommender Systems
Recommender systems help consumers learn and navigate the vast array of products to select the ones that best fit their preferences. These systems gather knowledge on customer preferences through data collected about them (e.g., demographic and psychographic information) and their past online transactions. The predictive knowledge obtained about a customer’s preferences is referred to as the profile of the customer. Based on this profile knowledge, recommender systems predict items that best match the current preferences of the customers. Typically, items are recommended to help a customer easily find the item s/he wishes to purchase so that the fit cost associated with her/his purchase can be reduced. Fit costs are often incurred when a customer is unable to find her/his ideal product among a large number of items offered for sale.
Despite their recognised benefits, recommender systems are not always viable for small-to-medium sized firms because of the high setup cost associated with their implementation and use. Hence, many less prominent electronic retailers still sell products without providing personalisation services (hereafter referred to as non-personalising firms). For example, the online bookseller Buybooksontheweb.com does not provide personalisation services to its customers. Similarly, while iTunes provides personalised recommendations (through their toolbar “Genius”) for music items, another online firm mp3million.com sells music without doing so. Typically, prices at non-personalising firms are lower.
Firms that do not use recommender system can still contribute to the improvement of a recommender system used by another firm by sharing the customer data the -non-personalising firms collect. With the enhanced capabilities of information technologies to analyse data, firms are realising the benefits of information sharing. For example, in the direct marketing industry, more than 600 catalog marketers exchange information about purchases of individual customers with competitors. Some firms maintain databases of customers in order to sell information of the detailed profiles of shoppers’ habits to other marketers. Using shared data enables firms to obtain additional valuable insights that are usually unavailable in the data they possess.
Firms that do not use recommender system can still contribute to the improvement of a recommender system used by another firm by sharing the customer data that the non-personalising firms collect. With the enhanced capabilities of information technologies to analyse data, firms are realising the benefits of information sharing.
The basic setting of our problem is as follows. Customers have the option to purchase from two competing firms - a personalising firm and a non-personalising firm. A customer first visits the personalising firm and uses the firm’s recommendations to find her preferred product. Then s/he purchases this product from the personalising firm or its cheaper substitute from the non-personalising firm. From the perspective of the firms, they compete to sell to the same population of customers. However, the non-personalising firm has the ability to share its data with the personalising firm. Our study asks the question: When should firms participate in data sharing and what is its impact on economic variables such as firm profits and product prices?
We analyse a situation when the data is directly shared between firms. To respect customer privacy, the shared data is assumed to exclude personally identifiable information (e.g., name, phone numbers, complete address, social security number, and credit card details). On the other hand, the shared data could include background information about a customer such as demographic information, spatial information (zip code, locality, etc.), and age group. It could also include customer ratings and reviews, coupons redeemed, co-purchased product information, etc. Such information can shed light on the general characteristics of customers and their preferences.
We next explain the setup of the model. The model first considers the customer’s problem of allocating her total purchase across the two firms. Next, based on customer’s decision, the firms choose prices to maximise profit with the possibility of data sharing. We allow for the fact that a customer may not be able to find her ideal product at either firm, instead the product comes with a fit cost. As the profile of the customer improves, the recommender system provides better help in finding the ideal product, which in turn reduces the fit cost. The customer can usually increase her surplus by purchasing cheaper product from the non-personalising firm. However, the product purchased from the non-personalising firm may not be exactly the same as her preferred product. Additionally, the customer may need to spend some extra effort to search for a (possibly imperfect) substitute at the non-personalising firm. Hence, the customer often incurs an additional cost (referred to as the substitution cost) while purchasing from the non-personalising firm. Thus, in addition to the prices paid to these firms, the customer incurs a fit cost when purchasing from either firm, but incurs an additional substitution cost when purchasing from the non-personalising firm.
When the customer purchases from the non-personalising firm, the personalising firm loses the information about the preferences of the customer implicit in her purchase. Consequently, an opportunity to further improve the customer’s profile is lost. Hence, while purchasing a product from the non-personalising firm, the customer faces a trade-off between increased fit cost and the lower price of the product (net of the substitution cost). Given this trade-off, the optimal strategy for the customer could be to distribute her purchases from the two firms, rather than purchasing exclusively at one of the firms. Empirical investigation into purchase behavior provides evidence that consumers indeed distribute their purchases across firms. In doing so, they forgo the cheaper price offered at the non-personalising firm.
Anticipating the purchasing behavior of customers, the two firms engage in a simultaneous move in the pricing game. The personalising firm uses profiles of individual customers for making recommendations. The profile quality of a customer depends upon the capability of the personalising firm to leverage customer data. This data could originate from purchases made by the customer from the personalising firm as well as (possibly) shared data obtained from the non-personalising firm. Finally, anticipating the pricing decisions and customer purchase decisions, the two firms enter into a data sharing game. In this stage of the game, the two firms make sharing decisions (share, not share) in equilibrium.
Key Results and Insights
The analysis in this study provides insights into several questions that are of managerial interest. Under certain conditions, both types of firms willingly participate in data sharing since their profits increase. On the other hand, there are scenarios where the personalising firm alone benefits from shared data, while the non-personalising firm does not. In this scenario, the personalising firm could pay the non-personalising firm in order to persuade them to sharing. Finally, under certain conditions, the personalising firm does not benefit from the shared data and should refuse to participate in data sharing.
We also analyse the impact of improving the recommender system on the profits of both types of firms. With an improvement in the recommender system, customers may shift more of their purchases toward the non-personalising firm because they can maintain the same profile quality with fewer purchases at the personalising firm. Thus, the personalising firm may lose some demand as a result of an improved recommender system. A natural question arises then: Should the personalising firm improve its recommender system, and if so, how will this improvement affect the product/services prices offered by the two firms? A related question is: How should the non-personalising firm react (in terms of price) when the recommender system improves? Interestingly, we find that the improved recommender system may not only benefit the personalising firm but also the non-personalising firm, i.e., the non-personalising firm may free-ride on the improved recommender system of the personalising firm.
Another interesting point is the increase in substitution cost. We find that, as the substitution cost increases, the customers with less sensitivity to the substitution cost start purchasing more from the non-personalising firm, whereas the customers with high sensitivity for the substitution cost purchase more from the personalising firm. Thus, it leads to natural segmentation of the customer population. This segmentation is important for the firm to consider during product promotion initiatives; for example, the personalising firm should offer coupons to customers who are likely to reduce purchasing from the personalising firm. Thus, in addition to the conventional role of learning customer preferences, recommender systems should also aim and aid in tracking and predicting the switching behavior of customers.