The Go-Togethers

Finding the right combination of statistical data and human intuition in cross-selling has for the most part eluded Internet retailers. Sure, collaborative filtering and data mining applications have made it possible to suggest accessories or companion products based on how frequently shoppers purchase those items in conjunction with another item, but there is no great mystery to the science behind the technology. Sales volume ultimately determines which products are offered and how high a product is ranked.

That`s not to say developing a cross-selling strategy around a rules-based mathematical model is a bad business choice. After all it has worked well for Amazon.com Inc. But it can be argued that books are an easier product to cross-sell in an online environment, especially for first-time customers for which there is no sales history. All that is really needed is a sense of the genre and the authors that appeal to the customer. Such data can easily be gleaned from the book the customer purchases. Someone who buys a novel by David Balducci, for example, is likely to enjoy other suspense thrillers by best selling authors.

The challenge to creating a more targeted cross-selling strategy in an online environment is much greater when it comes to products such as apparel, jewelry and furniture; that is, items that are driven more by taste than popularity.

The reason is simple: Rarely online is there is a sales associate who spends time with the customer learning about the customer`s wants, needs, likes and dislikes during the sales process to suggest an accompanying product. Whether the sales representative is servicing a first-time customer or repeat customer, such information is invaluable, because it enables the sales associate to suggest accessories more likely to appeal to the customer`s tastes.

A new generation

With this in mind, vendors of cross-selling software are developing a generation of applications that utilize data about a customer`s likely preferences, needs and wants to make the cross-sell a more effective proposition.

"Cross-selling applications are being geared to the principles of merchandising and that can pick up on variants of how people shop, rather than basing decisions on just mathematical data," says David Fry, CEO of Fry Inc., an Ann Arbor, Mich.-based web design, hosting and consulting company that helps Internet retailers develop branding and merchandising strategies. "It`s too easy for mathematics to lead to coincidental success in cross-selling."

The new generation of cross-selling applications infuses a larger dose of human intuition into the sales equation by not only tabulating historical sales data on a customer, but also including such data as the pages viewed, the types of products bought during a current shopping session and the price range of the products viewed and purchased. This type of information is then weighted based on information gleaned from customers with similar purchasing and behavior.

"Web analytics is no longer just looking at what people buy over time and using that data to create a list of the popular items bought," says John Squire, vice president of product management for San Mateo, Calif.-based Coremetrics Inc. "It includes weighting customer purchasing and behavior patterns to include more of a human element in the cross-selling process so that suggestions can be made in response to specific situations."

As part of its suite of web-based analytics applications, Coremetrics has developed Coremetrics 2005, a software platform that tracks all customer click data and stores it in a customer profile.

Within Coremetrics 2005, which is offered on an ASP basis, are reports that detail response rates to cross-sell promotions. Internet retailers can use the information not only to sharpen their cross-selling strategy, but also to measure response rates to suggested items, promotions, and even their overall merchandising strategy on a daily, weekly or monthly basis. Armed with this information, retailers can adjust their cross-selling strategy and product pairings sooner.

Taking the lost opportunities

Retailers can expect to generate a single digit lift in sales from implementing a cross-selling application, according to Bob Chatham, a principal analyst for Cambridge, Mass.-based Forrester Research Inc. That`s not a bad return considering that Coremetrics charges a one-time set up fee of $2,000 and $2,000 a month for its cross-selling application. In comparison, installing a proprietary application starts at about $500,000 and can easily exceed $1 million. Other providers of cross-sell applications include SPSS Inc., a Chicago-based analytics and data mining company.

"The goal is to avoid lost opportunities to cross-sell products by getting the right product in front of the customer, whether it is an add-on or a promotional item, by seeing what is working and what is not," Squire says.

As retailers view this information they can, in theory, suggest accessories more likely to appeal to an individual customer`s taste or a customer subset. They can also adjust how accessories are presented to the customer, such as a pop-up window or text directing them to another page or presenting the item during the checkout process to create an impulse buy.

Permutations

"There can be a lot of permutations of price, presentation, and promotion when it comes to merchandising," says Chatham. "Rotating content through the web site and creating multiple test cells that run simultaneously can help retailers find improvements in sales in places that they might not have looked."

Retailers can track each test cell by attaching a cookie to the customer`s computer. Test cells can be set for first-time customers, repeat customers or customers with specific browsing or purchasing habits.

IRCE 2008 Conference Multi-Media CD-ROM

While such features look attractive on paper, what matters most is whether the application can actually increase sales. So far the indications are affirmative. Burnsville, Minn.-based Northern Tool & Equipment, for example, has increased the value of its average order by 5% in the more than 18 months it has been using Coremetrics 2005.

Northern Tool, which offers more than 10,000 SKUs, segments its product line into 50 categories and 450 sub-categories, making it difficult for the five-member support team to accurately track customer behavior patterns to suggest appealing accessories. As a result, its cross-sell suggestions were based on the popularity of the product. "Most of the decisions were based on logic," recalls Nate Miller, e-commerce marketing manager for Northern Tool. "If someone bought a chainsaw, we`d suggest safety glasses, chain oil or a replacement chain."

Utilizing the tracking and analytical capabilities within Coremetrics 2005, the company has begun to spot less likely accessories, such as a pressure washer, but which are just as appealing to customers. About 3% of customers click on suggested cross-sell items, Miller says. Northern Tool predicts this capability will help it generate $3 million in additional sales annually. "We are taking the guesswork out of the cross-selling process," he adds.

Other Internet retailers report improved performance from using the new generation of analytical cross-selling models. Footsmart.com, a Norcross, Ga.-based multi-channel retailer of foot and lower body healthcare products, for example, has boosted the sale of suggested accessories from 1.56% of total sales to about 3.6%. The company attributes the increase to its ability to track sales of each item and to know how well it sells when paired with another item. It then uses that information to create other cross-sell suggestions.

Different approaches

Analytics reports supplied by Coremetrics helped Footsmart.com spot opportunities to adjust prices, improve product pairings and even presentation of merchandise on its site, according to company officials. Anecdotally, industry experts say that other multi-channel retailers have successfully used the new generation of cross-selling applications to develop separate merchandising strategies for their web site, stores and mail order/telephone order business.

L.L. Bean, for example, altered the merchandising strategy for seasonal clothing in its outlet stores and web site after discovering demand for products such as swimsuits started more than a month earlier on its web site than in its retail outlets located throughout the Northeast. "Retailers can find variants in data that identify trends of how people shop between the web site and the store," Fry says. "Overlaying customer data with purchasing data for individual products in each product category can lead to more actionable cross-selling decisions."

Trouble is that Internet retailers use less than 1% of the data they gather, according to Fry. The primary reason is that gathering the data is easier than disseminating and interpreting the data, even when using the latest generation of cross-selling applications. Northern Tool`s Miller says decisions on which products to pair for cross-sell opportunities are based purely on raw sales tabulated by Coremetrics. "We are probably using about 75% of the capability," he concedes. "We intend to get all the way there, but we are taking it in steps."

Miller adds the step-by-step approach has been dictated in part by the company`s decision to improve the navigation capabilities of its site, which was relaunched in October.

Learn as you go

Such diversions aside, most Internet retailers are only experimenting with cross-selling applications, if that. One reason is a lack of resources. Many Internet retailers are small companies with limited resources. Their natural inclination is to deploy a cross-sell application and let it drive the decision process. Only after a few weeks or months of examining daily or weekly reports is the retailer likely to begin tweaking its cross-sell strategy. "A lot of retailers are learning as they go," says Amy Hiatt, senior data analyst for Fry. "There is a lot of testing of trends identified by the reports, but there is also a tendency to rely too heavily on the reports when making merchandising decisions."

As a result, there remains a substantial gap between what cross-selling applications can do and how they are actually being used. "There really are no best practices yet and retailers need to take the time to think about the sales environment they want to build," says Coremetric`s Squire.

If nothing else, cross-selling applications can open the door to a continual dialog with customers. Squire says opportunities exist for retailers to pitch non-merchandise, such as extended warranty plans on electronics or monthly newsletters aimed at enhancing customer loyalty. The data gathered by the application can even be used to create customized promotional mailings that can drive traffic to the retailer`s web site. "If a retailer can offer the customer information in which they have an interest, there is a better chance the customer will think of that retailer as a first choice for shopping," he says.

Higher tickets

Without a doubt, consumers tend to favor retailers who provide a more satisfying sales experience, especially when it comes to suggesting accessories. Satisfied customers also tend to have a higher average ticket and generate repeat business.

The promise cross-selling applications hold for combining statistical analysis with human intuition can go a long way toward helping Internet retailers increase overall sales while improving customer loyalty and satisfaction. But so far, Internet retailers have shown more inclination to rely on the raw data supplied by the application in devising their cross-selling strategy than utilizing the technology`s fullest capabilities. Until that changes, retailers will merely be chasing a promise, instead of cashing in on reality.