Store Operations and Retail Data
This chapter considers store operations and the data required to support it. It is often an afterthought, as most of the focus of retail analytics relates to selling to customers, but running an efficient operation is also an area that merits attention, and where analytics can be of assistance.
Setting Up the Store for Success: Strategic Uses of Data
The author selects some examples of the way in which analytics can support store operations.
Handling merchandise, stocking shelves, maintaining the facility, assisting customers, and various tasks must be done to operate a store. As such, a retail store employs dozens or hundreds of full-time and part-time workers, and the need for many of them will fluctuate with the level of business done by the store.
Many store managers can estimate how many people are needed to perform specific tasks, such as running the registers at the front of the store, on any given day and hour. Where costs are managed, many chains give a specific store a fixed number of budgeted employee hours based on the previous year's volume.
It's extremely vague, and the practical consequences is that a store manager has scheduled in advanced a fixed number of employees to work a given shift, and must often make do with them. If there are too many employees on the clock, many of them will stand idle, which is a burden on a store that operates on a thin margin. If there are too few, the store will not be able to provide adequate service.
Given that many stores are open 24/7/365, scheduling has become a complicated task. It becomes more complicated when stores impose arbitrary rules, such as there may be no more than three customers in line before opening another register - to do so, there must be an employee at the ready to take over a cash register.
Cross-training floor associates to run a register can help, and a manager can sometimes call in a worker who expected to be off-duty, but neither is an ideal solution to the scheduling problem. Self-checkout lanes, which enable a single employee to monitor a set of six to eight registers, have been attempted at many stores, with mixed results.
Information technology has been applied to labor forecasting, but its accuracy depends on the ability to predict need. In most instances, basked data is the primary focal point - it indicates when purchases occur, and implies the amount of work needed to check out the customer, restock the inventory, and provide assistance in selecting certain items.
Importance of Accurate Labor Forecasting
The author refers to the "fun facts" section of Wal-Mart's web site, which lists average employee salary, total number of employees, average hours per week, and total number of customers served annually. (EN: I looked at the site and was unable to find this page. I expect it may have gone away - whether it was to safeguard sensitive information, or in recognition that these "fun facts" aren't much fun at all, except to statisticians.)
- Wal-Mart employs more than 1.3 million store associates in the US (1.9m worldwide).
- Average salary was just over $10 per hour
- Average work week is 30 hours
- Putting the first three together, this constitutes a labor expense of around $570 million
- There are more than 8,000 stores worldwide
- Most stores are open around the clock
- There are more than 125.000 checkout registers across the chain
- The chain serves 176 million customers
Granted, this is a very large example and the numbers are dramatic, but it underscores the cost of labor and the complexity involved in accurately predicting the number of hands needed to run registers, fill shelves, and attend to other duties. Getting it right is essential to serving customers adequately and reducing the wasted cost of idle employees.
Returning to the POS data, every customer transaction has a date and time stamp along with a list of every item the customer purchased. This historical data is used to forecast and predict future volume: how many transactions will be performed as well as how much merchandise is being removed from the shelves, both at any given time.
Even more granular detail can be analyzed, such as the number of self-service checkout transactions occur, how long it takes to check out a customer based on the number of items purchased, etc.
As a side note, a study by the National Retail Federation looked into impulse purchases at registers - which is influenced by the amount of space available for merchandising at the register, but also the amount of time a customer spends waiting to be checked out. Faster checkout pleases customers, but dramatically decreases impulse purchases.
Another granular consideration for a supercenter is exactly which registers to open. Given the fact that customers tend to shop one side of the store (grocery or general merchandise) per trip, it matters which specific registers are open: it would not make sense to have all staffing on the GM side during periods where most customers are shopping groceries. (EN: And given the size of the checkout bank, it's not an exaggeration to suggest a customer might have to walk 100 yards from one end to the other and are more likely to wait in line at a busy register than march to the far end where clerks are standing idle).
Adjusting forecasts for seasonal behavior is also critical. During the Christmas season, stores are mobbed and there are dramatic swings in consumer volume - the registers may be light for two hours, then suddenly mobbed for half an hour, then light once again. The same is true, to a lesser degree, when there are sales events during the regular calendar year.
Seasonality is further impacted by location - beach stores do a high volume business in summer and practically nothing in winter - and special events such as a football game at a nearby stadium can create a spike in traffic. Stores located near major events such as a playoff game (Super Bowl, NBA championship) or the Olympic games can be highly dramatic, and the influence can be difficult to predict.
(EN: This all seems very one-sided, focused on the retailers need for workers - but I'm also concerned about the worker's need for work. To the employee, a job must constitute a stable income: being worked overtime one week, part-time the next, and having no work at all the week after is unacceptable to the employee. Part-time, seasonal, and contract labor are common but poor solutions, as they generally involve temporary workers who are not trained and acculturated. None of this seems to be considered by theorists, and the assumption appears to be that there are trained and capable people who are happy to work overtime one day or stay home and earn nothing from one day to the next. This is a significant problem for the employees, as well as for the retailer.)
Consumer Differentiation at the Point of Sale Register
Quote: "The point at which the consumer pays for her merchandise is said to be the most important moment of truth between the merchant and the consumer." (EN: My sense is this places too much emphasis on a small part of the shopping experience. Customers don't routinely bail at checkouts, leaving the piles of goods they have selected behind and redoing their shopping elsewhere unless something goes catastrophically wrong. But it may influence their decision to return, so it is significant in the long term - not "the moment of truth" but still important.)
Most register systems (NCR, IBM, and Fujitsu) wire the POS terminals into a central information system, which itself is designed to provide data to other systems that inform management. Their primary design remains as an accounting system, tabulating sales, and extending to inventory management, but they are being leveraged for a wider array of purposes because the data they collect is valuable to store operations in general.
The author mentions using POS terminals to calculate accumulated discounts - to demonstrate to the member the money he has "saved" off of full retail prices as a result of discounts and coupons by shopping at a given retailer. His own experimentation suggests that this increases the frequency of trips in about 30% of customers, which was surprisingly significant.
He mentions a few other marketing-related analyses, such as tracking the purchase of specific items to tell which customers cherry-pick the store to buy only items that are on sale or discounted, and only shop when there is a sale, as opposed to creating stable and loyal customers through discounting. It's also apparent that certain sales events (Dollar Days) draw a lot of cherry pickers who do not become stable customers.
Heating and Cooling
An unexpected but intelligent use of basket data pertains to the need for heating and cooling to be adjusted to the number of customers in a store at a given time. Body heat can have a significant effect, and a shopper who is uncomfortable seeks to escape the "too hot" or "too cold" environment of a store, and buys less.
Because of the time it takes for an air conditioner or heater to fully heat or cool a large, open space, manual adjustment is insufficient: if the thermostat is turned down when it gets too hot, a large space will not be heated or cooled until the number of customers has changed, making it "too" far to one or the other extreme.
Using basket data, the store can predict the number of people who will be on-site at any given time, the heating and cooling systems can be programmed in advance to make adjustments to the temperature and hold it within a comfortable range.
It's also suggested that many regional power companies offer discounts to customers who can predict their demand - as heating and air conditioning of all the stores in a given area also constitutes fluctuations in their own operations, which can be better managed if they are predictable.
There are few stand-alone stores in the present day: most belong to chains that are regional or national in scope. Data from each store is collected and aggregated at headquarters, then disseminated back through regions, divisions, areas, and ultimately to the individual stores as orders to be followed and goals to be met.
(EN: This has often been a problem in retail - decisions are handed down to the store, which are contrary to the specific behavior that management and employees can witness on a daily basis. The "stupid ideas" imposed by "the corporate office" are a source of constant complaint, stress, and problems on the staff of a given store. Perhaps with better data, the decisions can be less stupid, but I sense that disempowering the front lines will always be an issue, no matter how much data is collected and how finely tuned the headquarters staff thinks its algorithms are.)
It's mentioned that regional and divisional managers will frequently make inspection visitors to stores in their demesne, and grill managers and key staff who are expected to know specific details about their store: daily sales estimates, benchmarks based on historical data, and goals. The ability of to manage "by the numbers" is critical to keeping your job, and to the store's continued existence.
Aside of that, there is an upside to operating a store in a chain, such as aggregating the data to learn from similar stores - though the presumption that two stores are similar must be questioned, as similarity depends not only on sales volume, but on store format, customer demographics, location, and other factors. All of this must be considered before two store may be considered to be similar.
Another advantage is the ability to do store-to-store transfers. If one store is selling product faster than another, merchandise can be moved from one to another more quickly than returning and reshipping through a distribution center.
(EN: What's not mentioned is the ability to do the same with employees if the stores are within a reasonable distance of one another.)
Intra-store transfers, which were once handled by informal communication between managers, can now be handled through data systems that monitor sales and inventory. Abnormal demand (high and low) can be tracked and transfers automatically arranged.
An alternative to transferring merchandise is changing prices, based on the economic model of supply and demand: raising the price of an item that is moving briskly or lowering the price of one that is moving slowly is reckoned to mitigate the fluctuations in demand.
In the modern store, it's as simple as changing the price in the store's system and dispatching a clerk to update the shelf label and signage. Some of the more technologically advanced retailers use electronic signage and display, so even that task is eliminated.
The author does concede negative impact on the consumer. While few customers will be impacted by a change that occurs during their shopping excursion (the price changes after they selected the item but before they have made the purchase), those few that are impacted will experience what the author calls "customer confusion." (EN: My sense is that it should not be so casually dismissed - a customer who notices a change that is not in his favor will be strongly displeased, and rightly so.)
Ideally, inventory management should progress to the point where every store gets sent exactly the quantities it needs and deliveries should be frequent enough to adjust - but merchandise transfers and liquidation markdowns remain in practice as error-correction mechanisms until inventory management can be perfected (if ever it indeed can).
Replenishment and POS Sales
Register data is strongly connected to the need to replenish inventory on the sales floor. It seems obvious, but each item a customer purchases and removes creates a need to replace it on the shelves; the shelves deplete the storeroom; the storeroom depletes a distribution center; the distribution center must obtain more goods from suppliers.
Generally, this is done by triggers: when a certain quantity is sold, orders are placed to obtain more. The inventory counts must be adjusted for damaged or stolen merchandise. The author suggests that setting up such triggers requires "a tremendous amount of data" but doesn't elaborate. (EN: In my mind, it's a data row that indicates the number of items that is decremented each time an item is purchased - but if the system is to be "smart" it may also consider time of purchase to be even more predictive.)
Managing inventory for replenishment is based on the assumption of a steady amount of normal business. This must be adjusted for seasonality, sales events, trends, weather, and other factors that will cause temporary fluctuations or even permanent adjustments (increasing lower limits when goods are in demand, decreasing them when they will no longer be in high demand).
Retail Career Path
(EN: The author speaks of career path in retail, which is a quaint and antiquated concept in the present world. People no longer stay with a company for the entirety of their careers, and companies no longer develop and promote employees through the ranks. It's nice in theory, and presents a sense of the way that things could work, and arguably should work, but in reality do not. Such is the problem with analysts and theorists in general.)
The entry-level position in most stores is typically the stockperson, who handles freight, stocks shelves, sets up displays and signage, and performs various other maintenance tasks. In some stores (such as grocery and general merchandise), cashiers may also be entry-level positions, but in others (department stores) it tends to be a position a stockperson matriculates into.
From there, a person may be promoted into a supervisory position over similar employees, and then to department management. Depending on the number of employees in a store, there may be a few levels of management until an employee with management potential tasks on ad assistant manager and store manager position. A store manager may be rotated through various stores before taking on a district, area, or regional management position.
Other career tracks involve specialization: a person may take a role in merchandising, advertising, buying operations, or other role and can likewise advance from smaller stores to larger ones, then to district and eventually corporate departments.
It is suggested that, in retail, store-level experience is valued over academic training: while some positions in a retail operation may be staffed with candidates with experience in other industries, most positions tend to be staffed with employees who have worked in the store environment.