Rangachari ANAND <email@example.com>
Manoj KUMAR <firstname.lastname@example.org>
Anant JHINGRAN <email@example.com>
IBM T. J. Watson Research Center
A number of trends in e-commerce have become evident in recent years. Online storefronts have proliferated, and advertisements for these storefronts in the form of clickthrough banners have become popular. Recently, the Internet has also become a distribution medium for conventional coupons. The shopper prints out such coupons with his or her printer and redeems them like any other coupon. An earlier paper by the authors (M. Kumar, A. Rangachari, A. Jhingran, and R. Mohan, "Sales Promotions on the Internet," Third USENIX workshop on Electronic Commerce, Boston, 31 Aug.-3 Sept. 1998, pp. 167-176) describes a form of sales promotion for online merchants called e-coupons. A buyer need not print e-coupons as they can be captured electronically in an electronic coupon-caddy, and later redeemed electronically. E-coupon issuers enjoy a high degree of flexibility in choosing which e-coupons are given to shoppers and when they are offered. For example, e-coupons could be offered to shoppers when they enter an online store, when they view a product description, or when they finalize their purchases. Similarly, e-coupons could be offered for a product for which a shopper has expressed interest, a product related to the product a shopper is buying, or a product the shopper never buys but the storekeeper is interested in promoting.
In this paper we describe an e-coupon delivery system that offers e-coupons to shoppers at the appropriate moment in the shopper's visit to an online storefront on the Internet. We discuss how shopper activity in an electronic storefront can be monitored to infer resistance or propensity for buying various merchandise, and how this information can be used to decide which e-coupons to offer, the terms of the e-coupon, and when the e-coupons should be offered. This contrasts with the simple implementation described in the author's earlier paper (1998), where e-coupons were placed in the shopper's coupon-caddy at the instant the merchant decided to issue them.
The basis for selecting e-coupons that should be offered to a particular shopper is a collection of promotional agents such as cross-sell and up-sell agents, surplus inventory management agent, agents to detect buy/no-buy conflict, agents that model users consumption, and so on. These correspond to the various business rationales for issuing e-coupons. Each individual agent recommends a list of e-coupons, with a score attached to each e-coupon (higher score indicating greater expected benefit from displaying the e-coupon). The agents derive their lists by analyzing various data such as users' demographic information, shopper's past purchases, items viewed in the current shopping session, and past purchases of other shoppers. In addition to this data analysis, agents also factor in the e-coupons they recommended in the past, and the shoppers' acceptance and usage of them. To obtain computational efficiency, partial calculation of the score for each e-coupon is done offline in batch mode, the part that relies on static data such as purchase history or consumption model. The calculations are completed in real time using data from the shopping session such as products viewed. The Web pages requested by the shopper during his or her shopping session have assigned space for displaying a small number of e-coupons. These Web pages also have attributes associated with them that are used by the promotional agents to calculate the scores of the e-coupons being recommended. For example, the attribute of the product page would be the product being displayed, substitute products, accessories to the product, and so on. Finally, the Web pages also specify a weight to be applied to the score of the e-coupons being recommended by each promotional agent. The final act of selecting which e-coupons to show to a shopper at various times during his or her visit to the mall is simple. The weighted scores of the e-coupons recommended by different promotional agents are compared, and the e-coupons with the highest weighted scores are displayed on the Web page. If multiple promotional agents recommend an e-coupon, then that e-coupons score is scaled up by a multiplicative factor. The coupon-caddy, used by the shoppers to capture the e-coupons offered to them by the merchant, is itself an active promotional agent. It recommends e-coupons stored in it that relate to the product being viewed or searched by the shopper.
We have a prototype coupon system operational on IBM's Net.Commerce platform and we are implementing the concepts described above on this prototype.
Broadly, most marketing activity can be described as advertising or sales promotions. While advertising is concerned with conveying information to consumers, sales promotions are short-term incentive tools designed to stimulate quicker and/or greater purchase of particular products or services by customers 1. A key feature of most types of sales promotions is that they usually offer some material gain to the customer. In recent years, sales promotions have increased in importance relative to advertising. For some classes of products, expenditure on sales promotions is often greater than that for advertising.
Conventional sales promotion techniques include coupons, rebates, samples, contests, and loyalty awards. Coupons are certificates that entitle the bearer to stated savings on the purchase of a specific product or product bundle. Coupons, which could be issued by both manufacturers and merchants, are distributed via newspaper inserts, in magazines, or by direct mail. Rebates are similar to coupons except that they usually require the customer to redeem the rebate certificate by mailing it to the manufacturer along with proof of purchase. In contrast, coupons are redeemed by merchants. Some of the popular uses for coupons include:
Although coupons are one of the most widely used forms of sales promotion, they suffer from a number of disadvantages. First, conventional distribution channels are slow and have long lead-times. Hence, distributing coupons requires much advance planning. Second, lower cost distribution channels such as newspapers do not lend themselves to precise targeting. Thus, many e-coupons may be given either to customers not interested in the promoted product or who might have bought the product without the coupon. Finally, since consumers must save and redeem coupons, their redemption rates are quite low. Often, less than 1% of the distributed coupons are redeemed.
While advertising on the Internet has become ubiquitous, sales promotions are not used widely on the Internet. In the case of samples and trial offers, the distributed nature of online commerce makes their distribution difficult except for soft goods such as music, software, and books. Loyalty awards have been offered recently by a number of online retailers. For example, Music Boulevard, an online music seller, features a "Frequent Buyers Club." By becoming a member, a shopper receives a free CD for every 9 CDs bought.
The coupon concept has not been adopted widely on the Internet. It should be noted however that while several Web sites 2 offer printable versions of conventional coupons, these coupons can not be redeemed online. Shoppers merely print out these coupons on their own printer and present them for redemption to shops and supermarkets. Not only are merchants wary of accepting such coupons, shoppers can print as many copies of the coupon as they wish.
In this paper we focus on e-coupons, described earlier,3 which are an adaptation of the coupon concept to the realm of online retailing. E-coupons address some of the problems observed with conventional coupons and provide some new functionality. In particular, we describe a number of methods by which e-coupons can be distributed on the Internet.
In the next section, we present a brief overview of e-coupons. Subsequently, we describe our e-coupon distribution system and present details of a prototype based on Net.Commerce, IBM's e-commerce platform. Finally we present a brief description of new work that we are pursuing in this area.
An e-coupon is an offer by an online merchant to a shopper to discount some or all items within one order, possibly subject to conditions, within a specified validity period.
While e-coupons could potentially be issued both by manufacturers and merchants, in this paper we will focus on e-coupons issued by merchants solely for use on their own Web sites. In the merchant-issued e-coupon scenario, all e-coupon-related data can be stored on the merchant's own server, thus simplifying security and avoiding the need for standards for merchant-manufacturer communication.
The main advantages of e-coupons over conventional coupons are:
The main events in the life cycle of an e-coupon are shown in Figure 1.
When a merchant decides to promote products using e-coupons, the first step is to design an e-coupon. This includes:
A key feature of e-coupons is that they can be targeted to merchant-identified groups of shoppers.
The e-coupon is then offered to shoppers by a variety of means described later in this paper. Note that since all coupon data are stored in the merchant's server, no actual data are sent to a shopper's computer. Instead a linkage between the shopper and the e-coupon is created in the merchant's database, which establishes the shopper's right to use the e-coupon.
When offered an e-coupon a shopper can either accept the e-coupon (that is, capture it), reject the e-coupon, or do nothing. If an e-coupon is accepted, it is added to the shopper's e-coupon wallet. The wallet is a tool that shoppers can use to review their e-coupons and to get recommendations on what they could purchase to take advantage of the e-coupons that they possess.
If a shopper does not accept an offered e-coupon, it may be offered to the shopper again at a later time. The maximum number of times that an e-coupon can be offered to a shopper can be specified. A rejected e-coupon will not be offered to a shopper again.
Shoppers present e-coupons for redemption when they pay for their order. At this time, shoppers are presented with a list of only those e-coupons in the shopper's wallet that could be applied to their order. The e-coupon redemption process causes discounts to be applied to the prices of items in the order. The total price of the order is then recomputed. An e-coupon can be used only once so it is removed from the shopper's wallet after redemption.
The e-coupon module will also offer tools that merchants could use to analyze the use patterns of e-coupons in order to gauge their effectiveness and refine their design for future use.
The main components of e-coupons are listed below:
Procedurally, e-coupons can be represented as two-level nested if-then statements. They follow the following pattern:
IF E-coupon condition THEN IF first item eligible for discount present in order THEN Apply discount to first item IF second item eligible for discount present in order THEN Apply discount to second item ...
Multiple e-coupons can be applied to an order with limitations. Merchants do not usually allow shoppers to obtain multiple discounts for an item. Hence, shoppers apply e-coupons sequentially. After each e-coupon application, discounted items are not eligible for further discounts. The e-coupon system does not compute the optimum combination of e-coupons (that is to obtain the greatest discount) for an order.
Having designed an e-coupon, the merchant will use the e-coupon distribution system to target and deliver the e-coupon to shoppers.
In this section, we first describe the support for targeting shoppers and then present details of the delivery mechanisms. Finally, we describe the operational details of the e-coupon distribution system.
Merchants will use information about their shoppers to target e-coupons. We use the term basis to refer collectively to all of the data pertaining to shoppers that will assist the merchant in targeting e-coupons. The basis consists of three types of data:
All three types of data are stored the database used by the Net.Commerce system. While shoppers' profiles are relatively small in size, the record of shopper transactions and especially the clickstream data may grow unmanageably large. Hence, the merchant may maintain only a small subset of the data online, with the rest kept in tertiary storage. Another possibility is that only a select subset of the clickstream will be recorded. Merchants can reference any combination of facts in the basis when creating targeting rules as we shall see later in section 3.3.
The e-coupon distribution system offers a number of options for delivering the e-coupon offer to shoppers. Each of these methods is suitable for the distribution of certain types of e-coupons.
The merchant can select the trigger actions anywhere in the Web store. To maximize impact, and for space reasons, only one e-coupon can be offered by this method at one time. To prevent shoppers from exploiting the system, the merchant can enable randomness in the offer so that not all shoppers are offered an e-coupon at a particular trigger. It is possible that multiple e-coupon offers may be tied to one trigger. In this case, one e-coupon is selected randomly for the actual offer. The main advantage of this method is that e-coupon offers can be displayed at the most opportune moment while the shopper is browsing. However, some shoppers may not be comfortable knowing that their behavior is being monitored.
The e-coupon distribution system is closely integrated with Net.Commerce, which is IBM's premier e-commerce platform. Net.Commerce has been used to build numerous high-traffic Web storefronts on the Internet. Wherever possible, the e-coupon distribution system uses Net.Commerce infrastructure.
The e-coupon distribution system consists of three main subsystems: the targeting subsystem, the filter subsystem, and the delivery subsystem. These subsystems as well as the related components are shown in the following diagram:
The targeting subsystem is a rules engine that generates e-coupon offer proposals. A proposal consists of an e-coupon identifier, a shopper identifier, and a delivery method. The generated proposals are input to the filter subsystem, which is also a rules engine. This subsystem suppresses erroneous or wasteful e-coupon offers. The filter subsystem then sends e-coupon offer proposals to the delivery subsystem, which uses either the Web server or the e-mail system to actually deliver the e-coupon to the shopper.
The targeting subsystem is a rules engine that executes merchant-specified rules to generate proposals. Rules are expressed as a set of conditions and actions. When all conditions are satisfied, the actions are performed. Conditions in rules can be based on
A trigger can be associated with shopper actions such as:
Merchants can create additional customized triggers if required. When a trigger is activated, it will assert a fact in the rule engine, which will contain details of the event, the context in which it occurred, and the identity of the shopper (if known) who performed the action.
Although merchants can use the targeting subsystem in many ways, we describe a number of useful scenarios below. We use an English-like notation to depict the rules for clarity. The rules are actually stored in a CLIPS-like format.
Merchants can set up filters in the form of rules to ensure that e-coupons are not offered to inappropriate recipients. Examples of filter rules would be:
Filters can apply to all e-coupons or can be made specific to a particular e-coupon. Merchants can also receive notifications with summaries of e-coupons suppressed.
The delivery subsystem is connected to both the Net.Commerce system and to the e-mail forwarder. All e-coupon offers other than e-mail offers are sent to the Net.Commerce system, which then merges the e-coupon offer with other data to create the HTML output to the shopper. For e-mail offers, the merchant specifies the text of the offer as a part of the e-coupon descriptive material. The URL is automatically added to the message.
E-coupons are an adaptation of the coupon concept to on-line sales. In this paper we have described the means by which e-coupons could be distributed on the Internet. At present, we have completed an operational prototype of the e-coupon system on the Net.Commerce platform.
We are investigating an agent-based approach to the problem of deciding the best e-coupon to offer a shopper, since it is necessary to limit e-coupon offers to maximize their impact. We propose to create agents associated with various business goals of the merchant such as promoting cross-sales, liquidation of poorly selling items, and increasing sales of an item.
When activated by a trigger, the agents will each propose a candidate e-coupon to offer the shopper along with a calculation of the value of the e-coupon to the merchant (using data in the basis). The e-coupon with the highest value would be selected.
1 Principles of Marketing, Philip Kotler and Gary Armstrong, Prentice Hall, 1998.
2 See http://www.e-coupon.com
3 "Sales promotions on the Internet," Rangachari Anand, Manoj Kumar, Anant Jhingran, and Rakesh Mohan, Proceedings of the Second Usenix Conference on E-commerce, Boston, 1998.