Download this case study

Fast Track to Profits: Using Simulation to Improve Store Operations

A Case Study with a Fortune 500 Company

________________________________________________________________________________________________________________________________________________

In retail, excellent inventory management makes it easy for customers to find what they want, minimizes undesired markdowns, and allows store associates to spend their time providing higher valued service. Any deviation from perfect execution has negative impact on sales and customer satisfaction. But, at some point, the extra effort and expense required to further improve is offset by the costs. Deciding whether or not to invest in better inventory management is difficult for two reasons. First, retailers lack data upon which to base decisions, and second, retailers lack tools to predict how changes will impact performance. Item-level radio-frequency identification (RFID) is a technology that delivers unprecedented access to data about store operations, and the Impinj Store Performance Simulator is a web-based tool that helps retailers predict how changes to store operations will impact performance and profits. Through a case study with a Fortune 500 company, we show how RFID and the Impinj Store Performance Simulator are complementary and point the way toward a new data-driven approach for maximizing retail profits.
________________________________________________________________________________________________________________________________________________

The Impinj Store Performance Simulator (SPS)

SPS is a retail store simulator that uses a method known as Discrete Event Simulation to play out shopping and store operations activities while keeping track of the cumulative results (see Using the Impinj Store Performance Simulator below). Developed to bring low-risk, low-cost experimentation to retail operations, SPS scenarios consist of over 25 inputs that describe how a specific store, or a product category within a store, operates. Inputs that describe SPS scenarios are organized into:

• > Operational parameters such as opening hours, supply chain responsiveness, and customer service availability
• > Inventory management variables like cycle counting activity, replenishment frequency and performance, and shrink rates
• > Merchandising data including numbers of SKUs, target inventory levels, shopper traffic, and seasonality
• > Financial information about selling prices, variable and fixed costs, and wages

SPS simulates the performance of a store for a given scenario over the course of a year while recording more than 35 performance metrics. Within SPS, the two primary actors are i) shoppers who come into a virtual store hoping to find something to buy; and ii) store associates who manage inventory and help shoppers. If shoppers find what they want (with or without help from an associate) they buy it. If not, they leave empty handed. Store associates try to keep shelves properly stocked by periodically and opportunistically moving merchandise from the back of the store to the front, and by periodically reordering from the distribution center. When associates are available, they help shoppers find items, which can include fetching items from back-stock or locating misplaced items. All of these activities are recorded in SPS on a daily basis along with other performance metrics for the store, including:

• > On-shelf availability, inventory assortments, and out-of-stock rates
• > Front, back and markdown inventory levels
• > Shopper traffic levels and store associate activities
• > Unit sales and units sold from back along with financial data

In addition to capturing daily performance metrics, SPS also produces an income statement that summarizes the store's performance for a full year of simulated operations in terms of turnover rates, unit sales, profit and profit margin. With SPS, users describe and compare scenarios to gain insight into how changes to input settings impact operations.

Though we developed SPS with RFID in mind, it is not just for RFID analysis. In fact, one of the most powerful aspects of SPS is that it allows retailers to quickly and cheaply compare many ways of improving store operations. For example, with SPS retailers can experiment with different front-stock and back-stock ratios, faster and slower supply chains, and different staffing levels for customer service or stock management.

One thing that SPS does not do is help retailers assess the impact of different marketing and merchandising strategies. In fact, SPS makes the assumption that the merchandising and marketing efforts will bring the desired and expected shoppers into the store, and that those shoppers are eager to purchase something. Because the focus of SPS is on store operations, all shipments from the distribution center in SPS are on-time and accurate. However, SPS could be extended to account for variability in distribution center availability and accuracy.

SPS Case Study

This opportunity to apply SPS arose when a Fortune 500 company suspected that they had a store-level inventory management problem and initiated a pilot project to measure and improve it. The goal of the pilot project was to increase on-shelf availability (without using RFID) for a non-seasonal category of apparel that should never be out-of-stock throughout the year. Working with the pilot team, we developed scenarios for SPS and tested predictions against actual store data with and without pilot project changes. Then we drilled into the simulation results to gain insight and understanding. And finally, we used SPS to make predictions about the additional benefits of implementing RFID. This is our attempt to relate some of the sense of discovery that we experienced in using SPS to understand the results of a real world experiment, and to explore future possibilities. To establish a baseline for the pilot project, on-shelf availability1 was measured in the store through weekly, manual cycle counts prior to making any changes to store operations or personnel training. Then, during the pilot phase, the weekly cycle counts continued as store associates were trained to adhere more closely to replenishment pick-lists for this category. At the same time, we configured the SPS scenarios to match as closely as possible to our understanding of how the stores operated both before and after the changes. Figure 1 shows the ranges of on-shelf availability results that were measured in the store compared to the ranges predicted by SPS. Each data set represents actual or simulated performance over a three month period. This data shows that SPS makes accurate predictions for on-shelf availability under the baseline and extra-effort scenarios. But, as with all simulations, accurate predictions do not validate the simulator once and for all. The success of simulation comes from a continuous process of testing and refinement of simulation scenarios against real world data until the users have confidence that the simulator is telling them something meaningful about the real world. And even then, the process continues. We now drill deeper into the simulation results to share just a few of the many "aha!" moments that we shared with the project team while using SPS. Figure 2 is a chart produced by SPS showing how on-shelf availability changes throughout the year under a simulated onslaught of shoppers. On day one, the store is fully stocked with 100% on-shelf availability. The baseline scenario in SPS shows that availability degrades to between 78% and 85% percent over the course of the year. When SPS runs the extra-effort scenario, which includes more cycle counting and increased replenishment accuracy, on-shelf availability stabilizes between 90% and 95%. This means that, for every 100 stocked SKUs, the extra effort made about six more SKUs available for shoppers at any given time-yet with the same total amount of merchandise on the selling floor. While it is intuitive that improved on-shelf availability is a good thing, SPS allows us to quantify the benefits in several ways. Consider "selling from the back" as an example. It is possible, in real life and in SPS, for store associates to fetch missing items for customers from back stock. Figure 3 shows how many items SPS records as sold from back-stock each day in the month of February for the two scenarios, which differ only in how inventory is managed. Selling from the back is better than not selling at all, but there is a cost both in terms of service labor and customer satisfaction. According to SPS, the 6% improvement in on-shelf availability between the two scenarios reduces selling from back by about half because more shoppers are finding what they want on their own. Unfortunately, sales associates are not always available when needed and back-stock availability is also not perfect. Taking these facts into account, SPS predicts that overall conversion rates2 are improved as shown in Figure 4 when extra effort is put into inventory management. The net result is captured in the income statement produced by SPS shown in Figure 5. Due to the higher conversion rates, inventory turnover is also higher. Unit sales have increased by almost 7%, leading to an operating profit increase of 35%. The high operating profit increase is due to the simple fact of life in retail, which is that incremental sales on the same fixed costs drive profitability. The payoff is big for those extra units sold. Putting extra effort into stock management proved to be about as effective in real life as it was in simulation. But the retailer in this case readily admits that frequent manual counts are neither sustainable in the long term, nor scalable across all product lines. Improved on-shelf availability has a significant value, but if it's too time-consuming, it won't happen. Furthermore, it is not reasonable to expect store associates to sustain improved replenishment performance without some mechanism for ongoing evaluation and feedback. This, too, must be built into the process. Using SPS, we modeled what it would be like to operate the same store using RFID for inventory management. The use of RFID was reflected in SPS in the following ways: cycle counts are performed once per week at a rate of 10,000 items per hour instead of 500 items per hour for manual cycle counting in the pilot. Cycle count data, coupled with POS activity and continuous RFID monitoring of the back-to-front transition keeps the simulated perpetual inventory (PI) database up to date. And, unlike the baseline scenario, the simulated PI has separate records for front-stock and back-stock which is used to drive replenishment and re-ordering. Figure 6 shows the on-shelf availability results comparing the retailer's extra-effort approach to the use of RFID instead. On average, SPS predicts that RFID will deliver an additional five percentage point increase in on-shelf availability over the extra-effort approach, even with an 85% reduction in labor devoted to cycle counting. The labor savings coupled with increased unit sales leads to a predicted 58% increase in operating profit for that category compared to baseline, or 21% improvement over the extra-effort approach, after accounting for the cost of the RFID tag. Not all scenarios show such a significant upside purely for improved inventory management. In many cases, retailers adopt RFID for a market basket of benefits including on-shelf availability, omni-channel fulfillment, working capital reductions, labor savings, and loss prevention.

Figure 1: SPS makes accurate predictions for on-shelf availability under the baseline and extra-effort scenarios. Figure 2: For every 100 stocked SKUs, the extra effort made about six more SKUs available for shoppers without adding inventory. Figure 3: A 6% improvement in on-shelf availability reduces selling from back by about half because more shoppers are finding what they want on their own. Figure 4: A higher conversion rate means faster inventory turnover and more unit sales. The extra-effort scenario shows a 7% increase in unit sales. Figure 5: Putting extra effort into stock management proved to be about as effective in real life as it was in simulation. Figure 6: SPS predicts that RFID will deliver a five percentage point increase in on-shelf availability over the extra-effort approach, even with an 85% reduction in labor.

The Availability Effect

When shoppers find exactly what they want, right away, they are happier and have more time and energy to spend shopping. In SPS we call this the availability effect, which is defined as the chance that a shopper will keep shopping if they have success³ on the first try. If we run the baseline scenario with and without a 5% availability bonus, SPS predicts a unit sales difference of only 3%. This is due to the fact that low availability hurts the retailer on both the first and second try. However, SPS also predicts that the same 5% availability effect accounts for 4.8% of sales when using RFID-a 1.8% improvement over the baseline. Higher availability leads to more first-try success, and to more second tries which are also more successful. At 10% net operating margin and 50% gross margin that 1.8% sales increase delivers a 9% net profit increase. The retailer's pilot project confirmed that the sales increases for the selected category did not come at the expense of other categories. In fact, the pilot project data showed an increase in cross selling due to better availability.

After accounting for the availability effect, the RFID scenario delivers more than 12% higher unit sales over the simulated baseline, all with reduced labor. If service level is defined as the chance that a customer gets assistance if they need it, then higher on-shelf availability means that higher service levels can be achieved with less actual labor. Better service is another kind of availability effect that can be quantified with SPS.

A Surprise

An assumption built in to the RFID scenario is that store associates will do a better job of adhering to a pick-list during replenishment. This is justified because RFID uniquely enables continuous monitoring and feedback. It is possible to improve only what you can measure. Yet, SPS shows that improved on-shelf availability when using RFID does not depend much on improved replenishment accuracy by store associates. Running the RFID scenario with and without improvements in replenishment accuracy shows only a 0.5% difference in on-shelf availability averaged over the year. This was a surprising result to us, but the reason is actually straightforward: RFID driven replenishment and reordering is self-correcting.

In POS driven systems, errors accumulate throughout the time between physical inventories, resulting in what are often called "frozen" or "hidden" out of stock conditions⁴. Figure 7 shows how hidden out-of-stock conditions accumulate to a certain point⁵ in the POS driven replenishment approach (baseline). But, with RFID, hidden out of stocks are continuously identified even with relatively low replenishment compliance. Simply put, if you didn't replenish it correctly yesterday, you'll be asked to do it again today. RFID is a real-world technology in that it automatically compensates for inevitable mistakes and lapses in practice.

Using the Impinj Store Performance Simulator The Impinj Store Performance Simulator is available on-line at http://sps.impinj.com as shown in Figure 8.

The left half of the display is the input side, the right half is the output side. The input side shows all of the input parameters that are being evaluated and has a column for each of two scenarios that are being compared. The differences between the two scenarios are highlighted in yellow. The drop-down menu at the top of each column is for selecting scenarios to load and compare. The default scenarios shown in Figure 8 are called "Baseline" and "RFID" and are prepopulated with store data (note: this scenario is intended to represent a denim category in a department store and is different from the scenarios described in the whitepaper). The red "compare" button causes SPS to run a simulation for each scenario and display the results on the right side of the display. The output side initially shows a comparative income statement on the top and a chart comparing daily operating profits for the store on the bottom, but users may select from a wide variety of other charts by choosing from the dropdown menus. Finally, users may download a spreadsheet containing all of the data for the two simulations. The trial version of SPS allows users to load prebuilt scenarios and change some of the input parameters. Please contact Impinj for access to the professional edition which allows users to edit and save new scenarios in which all inputs can be controlled.

Summary

On the surface, this project was about the benefits of better on-shelf availability. The retailer was able to determine, with a lot of hard work, that their on-shelf availability could be improved and that those improvements would pay-off handsomely if they could be sustained. But on another level, they learned something about what it means to run a data-driven retail operation. For instance, it would have been interesting to compare SPS predictions about sales from back-stock shown in Figure 3 with the real world data. Without RFID, that data is simply not available. Nor is it possible to know if store associates are replenishing the shelves as often or as accurately as they should. RFID provides the data that allows retailers to embark on a program of sustainable continuous improvement with data-driven incentives and accountability across the operation.

In our experience, the process of developing scenarios for SPS led to previously unasked questions about what was really happening in the stores, which in turn led to valuable insights and some surprises that have been validated. We have only been able to share a fraction of those insights here. Using SPS we have explored a range of interesting scenarios for this retailer, including more frequent reordering and shorter lead times from the DC, reduced back-stock coupled with more reliable re-supply (just in time supply chain), increased service levels, and even trade-offs between fixed RFID read points and more frequent use of hand-held scanners for cycle counting. The retailer is now in the process of planning an RFID pilot in conjunction with an overall program of continuous data-driven improvement.

Simulation is a proven approach to reducing cost and risk in the development of products, systems, and processes. Simulation is why Angry Birds is so much fun, the reason that new cars rarely fail crash tests, and it's why Intel can design processors with millions of interacting components. And now, we think it is how retailers can gain the confidence to make important changes to their operations and get the results that they expect, with RFID as the key enabling technology.

About Impinj, Inc.

Impinj is the leader in UHF Gen 2 RFID, the technology widely used in retail. As an RFID technology company, Impinj provides the chips and hardware that retail solution providers use in their offerings. Impinj works closely with leading retailers to ensure that the costs and benefits of the technology equate to high ROI. For more information about Impinj and RFID for retail, please visit www.impinj.com/retail/.

About the Author

Dr. Larry Arnstein is Impinj's Vice President of Business Development. Larry has led Impinj's efforts to support the retail industry's adoption of item level RFID for several years including a leadership role in the VICS Item Level RFID Initiative. Larry helps Impinj work closely with brand owners and retailers to ensure that RFID technology delivers business goals. Prior to joining Impinj, Larry received a PhD degree in Computer Engineering from Carnegie Mellon University and was Assistant Professor in the Computer Science & Engineering Department of the University of Washington. He can be contacted at larry.arnstein@impinj.com.

________________________________________________________________________________________________________________________________________________

¹ On-shelf availability is defined as the percentage of stocked SKUs that are represented by at least one item on the selling floor (front stock).

² In SPS, conversion rates are high because all SPS shoppers intend to buy something if they can find what they want. In real stores, many store visitors are not ready to buy, which makes actual conversion rates lower relative to total traffic entering a store. In SPS, the relative differences between predicted conversion rates for different scenarios are more important than the absolute number.

³ Success is defined as the shopper finding what they want without help from a store associate, which happens when the desired item is available in the right location on the selling floor.

⁴ Hidden out-of-stock is when the store systems believe that an item is available to the shopper when in fact it is not. Such items might not be replenished or re-ordered until a store associate discovers the problem. Two major sources of hidden-out-of stocks are failure by store associates to comply with POS driven pick-list replenishment, and shrink.

⁵ In SPS it is possible for store associates to discover unknown out-of-stocks during normal course of business, so we use the term "hidden" rather than "frozen." During simulation, as in real stores, the number of hidden out-of-stocks can go up or down depending on the rate that they are created and discovered.

SPS includes many concepts illuminated in the research by Dr. Bill Hardgrave and colleagues at the University of Arkansas Sam M. Walton College of Business. Several papers on RFID in retail and supply chain can be found at http://itri.uark.edu.