Charting the Course of Customers

Retail stores have used video-surveillance (tape-based) technology for many years as both a theft-deterrence tool and as an employee-monitoring tool. We strongly believe that usage of digitized video surveillance will soar to unprecedented levels because of a coming proliferation of low-cost charge coupled display (CCD) cells (the vision part of a digital camera) equipped with wireless IP technology. Although we recognize the value of this Inescapable Data technology as a crime-fighting tool, we also believe it will be exploited by retailers for another as yet little-known use: customer tracking.

For decades, merchandisers have desired a tool to better analyze customer buying habits. RFID helps in this effort, but in the end all that is truly known is what items are actually purchased (and perhaps at what time of the day and perhaps with what other purchases). What a merchandiser still does not know includes the following:

• Was the person male or female? Young or old? African American, Caucasian, or Hispanic?

• Was the person alone or with children?

• How long did the person pause in front of a particular display?

• What paths through the store did the person take, and how many places did he or she stop and for what time periods?

"We are so convinced of the value of such data that we looked into manufacturing some sort of small handheld device that we'd give to people as they entered our store," explains Robert Watson, owner of Watson-Janssens Marketing Agency, and former owner of the Black Lion retail chain. "Now, obviously, no consumer is going to want to carry something that is essentially watching them, so we abandoned that approach. Nevertheless, merchandisers are in the dark ages when it comes to having any sort of basic data about the people making purchases in their stores. This type of data is essential for growth in the hotly competitive retail world."

Using multiple, low-cost digital video-surveillance cameras that can be deployed easily (i.e., without wires), a retailer can now achieve 100 percent floor coverage and make visual fingerprints of individual shoppers or "shopping groups" as they enter a store. For example, Cognex, a major supplier of intelligent image-processing systems, offers a system that can accurately identify the pertinent details of individuals as they pass through a doorway. Even from above and at fairly modest resolution, the system can identify predominant and secondary clothing colors, approximate body shape and height, and the number of people in a shopping groupenough to fingerprint shoppers and then track them as they pass from camera view to camera view.

At the checkout, a camera aimed at face level can also distinguish major race and sex. Cameras will be in stores in any event, primarily for theft or security reasons. Adding a few additional units for shopper tracking (made possible by low cost, wireless enablement, and computer processing advances) gives store owners a completely new source of customer data. Retail stores are hungry for as much information about their shoppers as they can possibly acquire. Jan Davis, CEO of ShopperTrak, says, "More and more leading retailers are adopting conversion rate^[4] as one of their most powerful measures of success. They use the trend and level of conversion to assess how successful they are at driving customers to stores through advertising, meeting their customers' needs for service through adequate staffing, and offering their customers appropriate merchandise in attractive surroundings. A one-point change in conversion drives double-digit improvements in revenue."

^[4] The number of potential shoppers divided by the number of transactions a store sees over a given period.

Traffic data is an important measure for the sales potential of a store (conversion opportunity). For the most part, this data has been unavailable, and so retailers used far less useful data, such as sales figures, for many operational decisions. For example, staffing decisions would be mostly based on historical sales figures for given times of the day (or days of the week), data that does not correctly account for unserviced customers. The impact of out-of-stock situations, which occur frequently for a hotly sought-after sale or new-release item, cannot be understood by just a lack of sales; traffic analysis around the location could at least offer some insight into the missed opportunity. Fundamen tally, having traffic data changes how stores are managed. Instead of supplying a particular retail outlet with X units of an item based on past sales, the quantity could be based on more knowledge about the number of shoppers for a certain category of product. Similarly, stores, particular in-mall stores, rely heavily on window displays to attract shoppers into the store; however, they currently have no metric regarding window observers compared to how many then enter the store. A given "promotion" of some item can now measure the amount of attention the item received in terms of shoppers observing. With simply low sales figures, it would be difficult to know whether the low sales resulted from a lack of promotion, poor in-store placement, or price. Digital video enables retailers to passively acquire a set of behavioral shopping details that would have been impractical to gather otherwise.

ShopperTrak's Orbit

A Chicago company, ShopperTrak, makes a product that is very close to fulfilling the full video vision. The Orbit is a small four-inch by four-inch device that includes a camera and some computer processing technology. The device is located in ceilings and other locations throughout the store or mall. The system's primary application is counting the number of customers as they enter and leave a store or mall. It unobtrusively tracks customer movements and converts that data into counts. The video images are not stored, and thus individual privacy is protected. It is an improvement upon many other "customer-counting" devices in that it can better distinguish shoppers even in highly dense traffic such as in grand openings and in difficult entrances or areas of a store or mall. It counts people as they enter, as they exit, or as they pass by the store. It can also count people as they move throughout the store and measure intra-store traffic.

This data is then used by store, district, and corporate managers to understand traffic patterns, manage staffing levels, and measure the impact of advertising and promotion as well as the attractiveness of the store layout and merchandising. Traffic data can help better gauge how to staff the store. (Using just historical sales data alone does not account for how many people are left unserviced and unhappy.) If the owner has several stores, he or she can better understand how different layouts and merchandising affect sales of particular products. This system is currently used in many retail stores and malls already, with nearly 40,000 cameras installed throughout the world, including in more than 100 Louis Vuitton^[5] locations in North America. Knowing the conversion rate, these owners have a better understanding of the performance of stores. In retail, owners are constantly trying to raise their conversion rate.

According to Jan Davis, CEO of ShopperTrak, sales of their device have been briskgrowing at double-digit rates for more than 8 years with more than 40,000 installed by the end of 2004. Tracking shoppers as they shop is a reality.

^[5] http://www.shoppertrak.com/news_article53.html.

There is an old urban legend about how one large retail company, through extensive data mining (detailed mathematical relationship analysis of their sales databases), discovered that there was a correlation between the sale of beer and diapers in the early evening hours. It has been conjectured that this was due to the fact that husbands were stopping by the corner store on their way home from work (instead of the local bar) and buying diapers for the baby and beer for themselves. Whether this particular urban legend is true, such analysis is at the heart of data mining, the results of which can give retailers an edge over their competitors. Remember that for the power and size of mass retailers such as Wal-Mart, even a small change in the merchandising of one item on one day can lead to millions of dollars in new revenue the next day. Retailers are now driven by numbers and live in an age of grand analysis.

Suppose the retail relationship of diapers to six packs, for example, is proven through statistical analysis. How does a store now exploit this fact through better merchandising? There are two opposing choices. One choice is to place the diapers conveniently near the beer (next to or opposite the cooler, we assume). The other is to separate the two as far as possible. The first approach supposes that the shopper is in a hurry and has no other purchase desires and, in fact, may choose that store because it offers a faster shopping experience, assuming others are as convenient to stop at on the way home. The second approach attempts to capture and capitalize on the impulsive nature of a shopper who enjoys seeing other merchandise between the two sought-after objects and might choose to spend more time (and money) in the store, en route from the diaper rack to the cooler. (This is the more like the classic supermarket separation of the milk isle from the bread isle.) However, neither strategy could be known to be better than the other without more data. To know exactly which strategy has the best chance of boosting sales, the store owner needs to know the duration of shopping visits and the paths taken through the store. Was the shopper deliberate and focused, or did the new snack display en route cause a path pause? How many beer and diapers customers were in fact male?

Armed with new sources of digital video data, the information is finally becoming available.