1. Grocery Retail Strategies Fall Flat If Not Backed by the Right Supply ChainsFood retail is a tough and turbulent market. Grocery has never been easy, but the current business transformation is more dramatic than anything we have seen in decades, especially with COVID-19 further accelerating many trends shaping the market. Show
Grocery retailers need to simultaneously address several major trends:
These trends all present challenges and opportunities, but it’s clear that supply chain management will lie at the heart of the successes—and failures—we see in the future of grocery retail. All food retailers need to make tough choices today about where to place their business bets. Whichever strategy a retailer chooses, they have a slim chance of success if they don’t develop their grocery supply chain to match it. Successful food retailers need to master both the hard discounters’ lean, highly efficient grocery supply chains as well as the agile, responsive supply chains needed for fresh products. In addition, many of them will need to manage the complexity of operating multiple store formats while offering several fulfillment options. To achieve this, retailers need the right planning tools at their disposal. Furthermore, they need to understand how to apply them.
In this best practice guide, we will highlight key approaches for increasing both responsiveness and efficiency in grocery supply chains. You will be hard pressed to find a single retailer employing all of these best practices. Rather, we encourage you to prioritize the most feasible and impactful development areas from your own perspective. 2. Harness the Power Of AI to Optimize Your Grocery Supply ChainToday’s grocers collect massive amounts of data on transactions and interactions with consumers both on- and offline. That’s precisely why grocery retail is the perfect match for artificial intelligence (AI), which makes it possible to leverage that data into faster, more accurate decisions. This is an invaluable asset in an industry where retailers must control millions of goods flows and accurately match supply to demand at hundreds or even thousands of locations on a daily basis. Technology companies can be eager to position their AI algorithms as “intelligent” by making them as human as possible—even giving them human names like Siri, Alexa, Einstein, or Watson. Keep in mind, though, that AI is not a person. AI is not even a singular “it.” We are still far from general artificial intelligence that would be able to creatively solve ill-defined problems. We are, however, making great progress in specialized AI that solves well-defined problems (such as algorithms for image recognition) and combinations of several types of specialized AI (such as self-driving cars).
Retailers do not need “AI”—they need to employ several AI algorithms in their analytics toolbox to supplement “old hat” technologies like statistical analysis and rules-based heuristics. Machine learning algorithms, for example, consider hundreds of potentially demand-influencing factors when forecasting retail sales (Section 3.2), something a human demand planner could never achieve. The application of AI in grocery retail is not limited to demand forecasting, though. Retailers can reap even greater benefits by leveraging AI to optimize the full range of their operations – from order optimization (see section 4.1.1) to workforce optimization (read the whitepaper) and optimized markdowns (read the whitepaper). 3. Demand Forecasting is the Engine Running Your Grocery Supply ChainDemand forecasting is the engine running your grocery supply chain. Yet, despite the technology available, a great number of notable grocery retailers have yet to truly embark on their journey of data-driven forecasting. 3.1. Granular, Data-Driven Forecasting Is A Must for Grocery RetailersIn a 2020 survey of North American grocery retailers, only 52% of respondents reported that they could produce forecasts on the day-SKU-store level (SKU = stock-keeping unit). The rest wanted to do day-level forecasting, but simply weren’t able to. Few retailers today are able to forecast online orders picked in their stores separately either, which inevitably leads to capacity management issues, especially after the significant growth of online demand fueled by the COVID-19 crisis. Granular forecasting is not just a best practice—it’s a must-have in today’s grocery retail. Without detailed forecasts, it’s impossible to correctly position inventory in the supply chain to maximize sales and minimize waste. Granular forecasts are also the planning foundation for both resource and capacity management, and thus should be considered a prerequisite for profitable operations. 3.2. Machine Learning Delivers High Value In Grocery Demand ForecastingWhat started with a few forerunners like RELEX has become mainstream over the years: machine learning for retail demand forecasting. Currently, retail technology vendors either apply machine learning or are rushing to update their legacy systems to offer it. Machine learning gives a forecasting system the ability to learn automatically and improve its predictions using data alone, with no additional programming needed. Because retailers generate enormous amounts of data, machine learning technology quickly proves its value. Of course, machine learning algorithms are not new—they’ve been around for decades. But never before have they been able to access as much data or data-processing power as is available today. Though grocery retailers may have struggled to update their forecasts quickly in the past, large-scale data processing and in-memory technology now enable millions of forecast calculations within the space of a single minute. Machine learning makes it possible to incorporate a wide range of demand-influencing factors and relationships into your grocery sales forecasts. This is enormously valuable, as weather data alone can consist of hundreds of different factors that can potentially impact demand. However, a word of caution is called for here: it really does make a difference how machine learning is implemented. Although grocery retailers are able to collect massive amounts of data, their data is often quite limited on the store and SKU level. Slow-moving products may not provide enough sales transactions to study; master data on past promotions and product displays may be lacking; and as products move in and out of range, data may not be able for the exact SKU they’re attempting to forecast. Machine learning forecasting must be set up correctly to make it less vulnerable to data issues, which would cause a “garbage in, garbage out” scenario. Implemented the right way, machine learning effectively addresses common challenges with retail data to deliver benefits across all facets of grocery demand forecasting: 1) capturing recurring demand patterns caused by weekdays and seasons, 2) forecasting the impact of promotions, price changes, and other internal business decisions, 3) predicting the impact of local footfall, events, weather, and other external factors, and 4) even detecting when unknown factors (such as an unrecorded change in how a product is displayed) may be impacting demand. In simple scenarios like forecasting a predictably recurring demand variation, machine learning delivers only equivalent or slightly better accuracy than traditional time series-based demand forecasting. When dealing with complex situations, though, such as overlapping promotions or sales cannibalization, machine learning-based forecasting clearly outperforms traditional forecasting approaches. 3.3. Typical Demand Forecasting Challenges for Supermarkets, Discounters, and Convenience StoresNext, we will discuss how you can overcome some of the typical forecasting challenges that supermarkets, discounters, and convenience stores face. 3.3.1. Predicting Demand for New Products and StoresBecause machine learning relies on finding patterns in historical sales data, new products with no historical sales data can prove a challenge. Fortunately, further routines are available to improve the management of new product introductions as well. When introducing a new product, the most common approach is to assign it a reference product to use as a blueprint for its sales pattern until the new product has accumulated sufficient historical data of its own. However, in grocery retail, the number of new products per year can be massive. This makes manually identifying and setting reference products infeasible, or at least highly inefficient. It’s far more efficient to use a system that can automatically compare product attributes (e.g. product group, brand, pack size, color, or price point) to assign the most relevant reference product. The same approach can, of course, be applied to find suitable reference stores for new store openings. 3.3.2. Forecasting The Impact Of Promotions And Price & Display ChangesYour own business decisions as a retailer are also an important source of demand variation, from promotions and price changes to adjustments in how products are displayed throughout your stores. Despite the fact that retailers plan and control these changes themselves, many in the industry remain incapable of accurately predicting their impact. In the 2020 study of North American grocers, 70% of respondents indicated that they cannot consider all relevant aspects of a promotion—such as price, promotion type, or in-store display—when forecasting promotional uplifts. But they wish they could. Machine learning allows retailers to accurately model a product’s price elasticity, or how strongly a price change will affect that product’s demand. Price elasticity alone, however, does not capture the full impact of a price change. A product’s pricing in relation to other products in its category often has a large impact as well. In many categories, the product with the lowest price captures a disproportionally large share of demand. Machine learning-based demand forecasting makes it quite straightforward to consider a product’s price position, as shown in Figure 8 below. Machine learning does more than simply leverage price data, though. With machine learning forecasting, grocery retailers can accurately predict the impact of promotions by taking into consideration factors including, but by no means limited to:
3.3.3. Considering Cannibalization and Halo Effects in ForecastingIt’s quite common for a promotional uplift for one product to actually drive down sales for another. For example, if a supermarket carrying both the “HappyCow” and “GreenBeef” brands of lean organic ground beef puts the HappyCow product on promotion, more people will buy it—but it’s likely that some of the baseline demand for GreenBeef will shift to HappyCow. If they don’t lower the demand forecast for the GreenBeef product, planners are at a high risk of stock-piling, leading to waste. For most center store products, such as canned food or cereal, cannibalization is not a big problem. If demand decreases temporarily, a replenishment order for the cannibalized product will simply be triggered later than usual. However, when working with fresh products and especially products that have a limited number of direct substitutes, forecasts must consider the impact of cannibalization to avoid excess stock and spoilage. Manually adjusting the forecasts for all potentially cannibalized items is just not feasible in most retail contexts because the number of products to adjust is simply too large. Usually, the patterns are quite specific to individual stores’ assortments and shopping patterns as well. Machine learning algorithms’ ability to automatically identify patterns and adjust forecasts accordingly adds enormous value when addressing sales cannibalization. The flip side of cannibalization, of course, is the halo effect, or when promoting the HappyCow product also drives sales for related products outside of the “ground beef” class. Hamburger buns, for example, have an obvious and predictable correlation with ground beef. Unfortunately, the halo effect’s impact can be so diffused across the assortment that identifying every impacted product becomes more or less impossible, even with machine learning. Think onions, potato chips, beer, watermelon, taco meal kits, salad fixings, oyster crackers, corn on the cob, Worcestershire sauce, soy sauce, or any number of other items shoppers might associate with ground beef-based dishes. But even if forecasting systems can’t identify all possible halo relationships, they should still make it easy for planners to adjust forecasts for the relationships they know to exist. 3.3.4. Estimating The Impact of Weather and Other External Factors on DemandExternal factors such as the weather, local concerts and games, and competitor price changes can have a significant impact on demand. It often feels intuitively simple to understand how something like the weather, for example, might impact sales: high temperatures increase ice cream sales, rainfall increases demand for umbrellas, and so on. When looking at a retailer’s entire assortment offering, though, it becomes more complicated. The use of local weather data and forecasts to increase demand forecast accuracy is a great example of machine learning’s power. Machine learning algorithms can automatically detect the relationship between local weather variables and sales for individual products in individual stores. Machine learning algorithms can be used in this same way to take advantage of a wide range of data sources beyond weather alone, helping retailers identify the relationships between external variables such as local sporting events or concerts and local sales for specific products. In grocery retail, the most useful external data sources include:
3.3.5. Dealing with Unexplained Changes in DemandIn brick-and-mortar retail, local circumstances—such as a direct competitor opening or closing a nearby store—may cause a change in demand. Unfortunately, data on the factor causing this change may not be recorded in any system. Sometimes, retailers’ own internal decisions also go unrecorded, such as adding a product to a special off-shelf display area in a store. Fortunately, machine learning can help in these situations. Machine learning algorithms can tentatively place a “change point” in the forecasting model, then track subsequent data to either disprove or validate the hypothesis. This allows forecasts to adapt quickly and automatically to new demand levels. Consider the example in Figure 12 below, which shows the sales impact when store staff created a table display in addition to the regular shelf space for a product. Though nobody recorded this change in the master data, the system was easily able to track the demand impact as a factor of how the product was displayed in the store. 3.3.6. Incorporating Planner Expertise In Demand ForecastingIf you want to remain competitive in food retail, machine learning is something you have to adopt, but you also need to understand its limitations. Automating the majority of your demand forecasting isn’t just desirable—it’s actually quite feasible with machine learning. But the COVID-19 crisis demonstrated clearly that there will always be circumstances in which system-generated forecasts will be off (although some systems can recover more quickly than others). The pandemic was a particularly extreme shock to the system, but in a business as dynamic as retail, there is always a risk that forecasts based on how things used to be will fail to accurately capture how things are now or how they will be in the future. No demand planning solution, no matter how advanced, can ever completely escape forecasting errors. That’s why it’s so important that the experts on retail demand planning teams are able to fully understand forecast errors. If their system provides transparency into how it forms the forecast, retail experts are able to quickly understand and correct any errors they see in it. Too many retailers rely on “black box” forecasting systems that can take in all sorts of data to produce accurate forecasts, but lack transparency. Black box systems may well kill your business—or at least your planning efficiency—for a couple of reasons. First, occasional extreme forecast errors can inflict far more damage on performance than smaller, more frequent errors. Second, when demand planners can’t make sense of forecast errors, it erodes their confidence in all forecast calculations, leading to increased double-checking and manual forecasting and undercutting the entire goal of harnessing computer power to automate your forecasting. That’s why best practice retailers understand the value of transparency. Even when the system does the heavy lifting, human planners must always be able to both understand and control how their forecasts are generated. 4. Refine Grocery Replenishment for Improved Availability, Waste, and EfficiencyThe quality of a grocery retailer’s store replenishment process has a direct impact on its top line revenue and bottom line profits. High quality, forecast-driven grocery store replenishment consistently translates into the following benefits:
Yet, in a 2020 survey of North American grocery retailers, only 24% of respondents had implemented some level of forecast-driven store replenishment, and only 7% had implemented it extensively. Store replenishment is definitely an area where many supermarkets’, discounters’, and convenience stores’ operations are currently quite far from best practice. 4.1. Fresh Food Replenishment Requires Detailed Planning and ExecutionFor fresh products, well-managed store replenishment is central to finding the optimal balance between the risk of lost sales margins caused by stock-outs and the risk of waste or markdowns eating already slim margins. Even though traditional supermarkets have decades of experience dealing with fresh products, many still do not excel in this area. Their supply chains are reactive enough to support frequent deliveries, but their replenishment planning is not up to scratch. According to the North American grocers surveyed, the annual value of spoilage was on average more than $70 million, and up to several hundred million dollars annually for the largest companies offering a wide range of fresh products. A reduction of 10-40% would translate into somewhere between $7–28 million in annual savings. This is not only feasible, it is something that modern food retailers are expected to do in order to reduce their carbon footprints and make their businesses more sustainable. (Read more about reducing carbon footprints in supermarkets, convenience, food service, and e-grocery here). 4.1.1. Balancing Waste and Lost SalesFor so called ultra-fresh products, meaning short shelf life items that need to be sold that same day, 100% on-shelf availability means that there will always be waste or markdowns unless the forecast is consistently flawless on the day, store, and product level. This means that very granular control is needed to find the optimal balance between the risk of stock-outs and the risk of waste. Other fresh products face a similar challenge, just a bit less pronounced. Demand for a product in a specific store typically varies between different weekdays. For some stores and products, this weekday variation in fresh replenishment can be very dramatic. This means that the same safety stock does not fit all weekdays when dealing with short shelf life products. Roast beef, for example, tends to sell a lot more leading into the weekend than coming out of it. For roast beef – even when the day-level forecast is accurate – a static safety stock level leads to 1) excess inventory after the weekend, with increased risk of waste, and 2) perilously low safety stocks during the weekend, with an increased risk of stock outs. To find the right balance between the risk for waste and the risk for stock-outs, safety stocks need to move up and down in step with the expected sales volumes and forecast errors for the different weekdays. Good retail planning systems do this kind of granular safety stock optimization automatically. In fact, the best retail planning systems take optimization even further by not only enabling dynamic safety stocks, but optimizing each order based on cost-benefit calculations that balance the risk of waste against the risk of out-of-stocks. Such machine learning algorithms minimize the total of lost sales margin and cost of waste. The cost function needs to be adjustable in terms of how much weight it places on on-shelf availability vs. waste to allow for considering the strategic roles of key categories and items as well as whether there are many or limited opportunities for substitution within the product category.
When managing store replenishment of fresh products, it is very important that all calculations and optimizations are done automatically. It is an impossible task for any human to keep track of all factors influencing demand, such as weekday variation (e.g. seasons, weather, and promotions) as well as all factors influencing replenishment (e.g. delivery schedules, batch sizes and day-level probabilities of waste and stock-outs) for hundreds or thousands of items per day in a store, let alone hundreds of stores. However, it is equally important that the forecasting and replenishment system does not turn into a black box. Actionable analytics allow supply planners to easily detect and remedy exceptions such as historical or projected waste or poor availability. Examples of typical exceptions in fresh food replenishment are:
Automation radically reduces the time spent on routine tasks in store replenishment planning. At the same time, it multiplies the impact of your most knowledgeable process experts. If store replenishment has not been automated, your best supply chain analysts have limited leverage. They can review successes and failures in the rearview mirror and try to turn a few of their findings into action in the stores with help from the field training team. When store replenishment is automated and replenishment planning centralized to a knowledgeable team, your planning experts can make a visible difference in hundreds of stores, almost immediately, simply by fine-tuning replenishment settings. 4.1.2. Stores Have Turned into KitchensWith consumers increasingly looking for convenience, food-to-go and meal solutions are on the rise. Many stores are turning into kitchens where sandwiches, hotdogs, and salads are made. Traditionally, products manufactured on-site have been considered special items that have to be managed manually in the stores. However, with the growing demand for ready meals, the importance of on-site production has grown much more pronounced and more critical to food retailers’ profitability. The replenishment process of ready meals is not all that different from replenishing other products sold in a store. It is just a bit more complicated. The demand for the end products—the meals—needs to be translated into the ingredients used to manufacture the end products. The replenishment calculations need to be done for each ingredient accounting for each ingredient’s lead time and on-hand stock. Essentially, the process is the following:
Sometimes, the ingredients included in a recipe are composed of other ingredients, such as special mayo or mustard produced on-site. In those cases, similar calculations need to be conducted for several levels of recipes. A horrible task for any human being, but quite manageable for a computer. 4.1.3. High-frequency Replenishment for Ultra-Fresh ProductsFor ultra-fresh products, many retailers have chosen to deliver them to stores multiple times per day to guarantee freshness. Similarly, items produced on-site are typically prepared in several batches during the day. This applies especially to the growing category of in-store bakery products, which ideally should still be warm when the customer picks them up. In addition, the new trend of food retailers opening small stores in urban locations has made several replenishments per day a must due to lack of in-store storage space. Placing more orders per day or designing the optimal bake plan per day requires factoring in both weekday-related and within-the-day variation in demand. For some products, the within the day or so-called intraday demand pattern will follow the general customer footfall for that day; for other products, such as lunch items, demand is more influenced by how the items are planned to be consumed. Again, keeping track of both weekday and intraday demand patterns manually is quite a complex and error-prone process. Yet, many retailers still rely on their store associates to figure this out on their own. This is a high-stakes gamble, as ultra-fresh products inevitably have a big impact on how consumers judge the quality of fresh products in a store. Best-in-class retail planning systems can figure out the optimal split between multiple orders or production batches per day as well as adjust the quantities as needed, automatically. 4.1.4. Adding Science to the Art of Managing Fruits and VegetablesFruits and vegetables are often last in line when store ordering is automated. Obviously, produce faces the same challenges caused by short shelf life and variable demand as other fresh product categories. In addition, the varying supply and quality of fruits and vegetables demand additional flexibility from the planning system in use. The regions from which fruits and vegetables are sourced constantly change as crops are harvested in different parts of the world at different times. Even growers in the same region may have timed their crops slightly differently. Furthermore, as there is always some uncertainty in the availability of good quality product, food retailers usually try to ensure that they always have several vendors for the same product. From a consumer perspective, a lemon is a lemon, but the supply chain may need to deal with tens of different product codes for lemon, each associated with a different vendor. Effective management of fruits and vegetables requires the planning system to be able to seamlessly switch between planning levels as needed:
The forecasting and replenishment process for fruits and vegetables is highly laborious to manage manually but can be effectively automated. The key prerequisite is clear guidelines for which products are to be included in the stores’ assortments and which vendors are to be used for sourcing at any given time. As in any automation process, high-quality master data is essential. 4.2. Optimized Center Store Replenishment Is Key to Supply Chain EfficiencyFresh products need to be delivered to stores in perfect sync with demand. Center store products and other products with longer shelf-lives, on the other hand, offer more opportunities for an optimized flow of inventory in the supply chain. Optimized replenishment of center store products is key to lowering costs in stores and throughout the grocery supply chain. Retailers that have mastered the replenishment of non-perishable products benefit from a much more level flow of goods through their distribution centers, enabling a much quicker return on investments in warehouse automation and reducing the risk of capacity bottlenecks having a negative impact on on-shelf availability. Furthermore, as grocery retailers’ store staff spend a lot of time and effort shelving products, optimized replenishment helps retailers to reduce operational costs in their stores. 4.2.1. Synching Replenishment and Shelf Space for Cost-Efficient OperationsGrocery retailers have traditionally operated in a very siloed manner with very little communication between the merchandizing teams responsible for store planograms, the supply chain teams responsible for store replenishment, and the store operations teams responsible for in-store work processes. This must change. The space allocated to each product in a store has a big impact on both the results and costs of the store replenishment process:
Although surprisingly rare, full integration between space and replenishment planning is an important best practice for increased operational efficiency:
The space assigned to each product is of vital importance to how efficiently the replenishment process can function, so it is important to deliver continuous feedback to merchandizing. Good analytics tools will help you identify products and stores where there is a mismatch between space and sales, i.e. products and stores for which incoming deliveries do not fit directly on the shelves or products and stores where visual minimums lead to waste or markdowns. Ideally, space planning should always be based on the detailed store-, product-, and day-level forecasts as well as information on replenishment cycles and main replenishment days available from replenishment planning:
We have seen forecast-based optimization of shelf space translate into up to 30% lower distribution and in-store replenishment costs. 4.2.2. Smart Replenishment For Efficient Store Operations And More Level Goods FlowsTypically, every large grocery retailer replenishes all or at least most of its stores every day from its distribution centers. This is because fresh products demand frequent deliveries and because the overall inventory flows are substantial enough to warrant daily deliveries. If all replenishment opportunities are used for all product groups without discretion, two problems will follow:
Instead of automatically using all available order or replenishment opportunities for all products, the best practice is to define main replenishment days for longer shelf life products. This means that replenishment of some center store product groups is concentrated to specific weekdays instead of being scattered throughout the week. Replenishment planning, such as the optimization of safety stock and calculation of order quantities, will be based on delivering the goods on the specified main replenishment days. However, to ensure the highest possible availability, replenishment orders are also triggered for the other available replenishment days to avoid stock-outs if there are unexpected demand peaks. In practice, this means that instead of ordering detergents every day, fast moving detergents are primarily replenished on, for example, Mondays and Thursdays, and slow-moving detergents on Thursdays. For detergents, the other replenishment days from the distribution center to the store are only used in case there is a risk of stock-out in the store. The use of main replenishment days allows for significantly more efficient in-store replenishment without hampering on-shelf availability. More consolidated deliveries make it more efficient for store personnel to replenish store shelves, especially when the main replenishment days are set based on what product categories are displayed in the same aisle or zone of a store. We have seen reductions of 20% in the time spent stacking shelves following the introduction of main replenishment days. As with many other processes, the use of main replenishment days can be further optimized when the basics are in place. By using AI to optimize main replenishment days across the whole store, the inbound goods flow to a store can be leveled out across weekdays. In many stores, weekends can be very busy, with lots of customers doing their weekly shopping while large quantities of fresh products are being delivered to the stores. Setting main replenishment days for center store products to the quieter weekdays balances the incoming goods flow and makes personnel planning in the stores easier. (Read this case study for additional details.) 4.2.3. Dynamic Pack Sizes to Meet Dynamic DemandOne powerful tool to increase store replenishment efficiency is to optimize the use of different pack sizes. In many cases, stores can choose to order case packs, pallet layers, or full pallets from the distribution center. Larger batches are more efficient to handle both in the stores and at the distribution centers, but clearly the deliveries need to match the available space and demand in the stores. Otherwise inventory will pile up in the stores and reduce rather than increase efficiency by congesting backrooms and causing multiple trips between the backroom and shop floor to replenish shelves. Especially for retailers operating stores of different sizes, optimizing replenishment pack sizes per product and store has a direct impact on handling costs. However, doing it only once as a concerted effort does not suffice as demand changes over time and, for some products, also with the seasons. During the high season, a pallet might be most efficient while outside the peak, smaller case packs may be more efficient. The retail planning system needs to be able to automatically optimize which pack size to use per product, store and order. This means that whenever an order is placed, the system always checks all available pack sizes—typically varying from the case pack to full pallets—and selects the most efficient pack size in relation to forecasted demand. To attain the full efficiency gain, the supplying warehouses need to be able to estimate the demand for the different pack sizes. Otherwise they may end up in a situation where they use individual packs to put together pallets for the stores, rather than having full pallets flow through the distribution system. This is possible when the store projections (see Section 5) used as the basis for distribution planning reflect the stores’ forecasted use of different pack sizes. 5. An Integrated Supply Chain Driven by Customer DemandTraditionally, store replenishment and inventory management at the regional distribution centers or central warehouses have been separate processes, driven by separate demand forecasts. In a 2020 survey, we found that 31% of large US grocery retailers still base their distribution center forecasts on historical data of outbound deliveries from these distribution centers. This is akin to driving a car while looking at the rearview mirror. According to the same survey, only 29% of respondents have chosen the more forward-looking approach of basing their distribution center forecasts on store demand forecasts. Granted, this is a better approach than only looking at outbound deliveries. There are, however, some important disadvantages to using store demand forecasts to drive planning at the distribution centers:
It is quite ironic that many of the situations considered most difficult to tackle in the distribution centers, such as building up stock in stores for promotions or new product introductions, are situations fully in the hands of the retailers themselves. The best practice is to base distribution center forecasting on the stores’ projected orders, which reflect both pull-based demand as well as planned, push-based stock movements. In the 2020 survey, 40% of responding North American grocery retailers had implemented this. To achieve seamless integration of store and distribution planning, the planning system needs to be able to calculate projected store orders per product, store, and day several months or even a year into the future, reflecting current and known future replenishment parameters as well as the demand forecast. These calculations, of course, require significant data processing capacity, which is likely to be one explanation for the surprisingly low adoption rates. In practice, the stores’ order projections consolidate data on their current inventory, safety stocks and visual minimums, delivery schedules (including main replenishment days), as well as any planned inventory movements, including everything from stocking up to build promotional displays to shifting orders to level out the capacity requirements in distribution. Table 1 presents some examples of situations in which the value of basing forecasting at the distribution centers on projected store orders is especially notable.
When the order projections are aggregated across all stores, they form a very accurate, customer-driven forecast for the distribution centers. Additional benefits of the supply chain integration include supply chain transparency supporting capacity planning, supplier collaboration (discussed in Section 7) as well as straightforward handling of cross-docking, pick-to-zero, and shortage situations. 5.1. Plan Once and Execute Automatically Throughout the Supply ChainWhen planning at the distribution centers is based on the stores’ projected orders, the impact of planned activities, such as promotions or pre-season allocations, are immediately visible throughout the entire supply chain. To reap the full benefits of this transparency, all planning data needs to be made available to the planning system as soon as a promotion plan, assortment change, price change, or any other relevant decision has been made. A planning system that supports time-dependent master data is a key enabler of proactive planning. Below are just a few examples of how time-dependent master data enables you to register valuable information immediately when it becomes available. This, in turn, allows your replenishment planners to rely on the planning system to automatically trigger the necessary actions at the right time with very little manual work.
An integrated supply chain set-up removes the need for double-planning work. The impacts of planned changes in store replenishment are automatically reflected in the projected store orders forming the demand forecast for the distribution centers. This means that as soon as the required store stock-ups for promotions are planned, they will be visible in the distribution center forecast on the right dates and in the right quantities. Of course, having the right functionality in your planning system is a key enabler, but the real challenge is getting the whole organization to work more proactively. Ensuring that decisions are made early enough, but not too early to unnecessarily reduce flexibility in a dynamic market, requires that everyone in the organization has a basic understanding of how the supply chain works and what the relevant lead times for different types of decisions are. 5.2. Multi-echelon Optimization of Goods FlowsAn integrated supply chain makes it possible to manage multi-echelon inventory flows efficiently, with minimum waste and a high level of automation. When all data on demand forecasts, available stock, delivery schedules, lead times and batch sizes for all supply chain echelons is available in the same planning system, it enables seamless optimization of inventory flows throughout the supply chain. Cross-docking is an inventory strategy aimed at maximizing transportation efficiency while minimizing handling costs. Cross-docking is often applied to bulky products, such as drinks, to attain lower storage and handling costs. It can also be used to cut lead-times for short shelf life products. In a cross-docking set-up, goods are delivered from the supplier to a cross-docking facility where the goods are put not into storage, but moved from the inbound truck to an outbound truck for distribution to stores. There are some requirements for cross-docking to work efficiently: 1) Suppliers need to be able to deliver full truckloads to the cross-docking facilities, 2) the delivery units, such as pallets or roll-cages, need to be ready for immediate movement to the outbound trucks without additional handling, and 3) the outbound trucks need to get a high fill-rate to keep transportation costs down. The planning system, thus, needs to optimize both inbound and outbound flows to and from the cross-docking facilities as well as understand the total lead time from supplier to store. Another example of an inventory policy that requires integrated supply chain planning is pick-to-zero. In this inventory strategy, orders to the suppliers are based on the stores’ replenishment needs. However, rather than fixing the quantities to be delivered to each store, the supplier delivery is reallocated to the stores upon receipt based on the latest inventory and forecast information. This allows for adjusting the delivery quantities per store in case the supplier could not deliver in full or in response to potential unexpected demand peaks in the stores after the original replenishment need was calculated. As a result, supply matches demand more accurately than when using the traditional cross-docking approach. The pick-to-zero approach can be seen as a way of shortening the order-to-delivery lead times to the stores, as the store-specific quantities are finalized not when ordering from the suppliers but when preparing the goods for store distribution. When supply chain planning is fully integrated, exceptions can be resolved in an optimal and automated fashion. Let’s look at inventory scarcity due to, for example, an incoming shipment being delayed. Instead of fulfilling store orders on a first-come, first-served basis, the available inventory can be automatically allocated to stores to maximize overall on-shelf availability or in accordance with a tactical prioritization of the stores. In the best case, on-shelf availability is not even affected. In a similar manner, inventory batches nearing their expiration dates can proactively be forced out to the stores that have the best chance of selling the products at full price. 6. Efficient Inventory Management in Distribution CentersReplenishment of central warehouses and distribution centers is sometimes seen as more of an art than a science. It is true that longer lead times, especially when ordering overseas, and lack of control over external suppliers introduce complexities. Yet, at least in principle, replenishing central warehouses or distribution centers is not that different from replenishing stores. When replenishing stores from their own distribution centers, retailers can optimize order fulfillment as they find best. When ordering goods from suppliers, though, there may be complex restrictions regarding minimum order value or quantity. In addition, there may be volume-based discounts or other rebates which, when efficiently exploited, can have a significant impact on margins. Many retailers have not been able to put this kind of supplier contract or price information into their planning systems, making it necessary for the operative buyers to invest significant time double checking orders. When replenishing stores, the active goods flows (combinations of products and stores) for any larger retailer is typically measured in millions or tens of millions, which means that automation is crucial. For central and regional warehouses, the number of order lines is much smaller and the value per order line much higher, making the economic impact of each order line more pronounced. This has both enabled and encouraged a lower degree of automation in operative buying compared to store replenishment. We have found that setting up operative buying processes in a structured way with good system support can also result in very high levels of automation at the distribution centers. This does not, however, necessarily mean that best practice retailers have a significantly leaner buying team. A key result of increasing the automation of routine tasks is that operative buyers have more time to proactively deal with potential capacity, delivery or quality issues and to analyze the performance of the current assortment, suppliers and supplier agreements for continuous improvements. 6.1. Total Cost Optimization of Inbound FlowsAs the inbound goods flows to distribution centers are more consolidated than the outbound flows, there are more opportunities for order optimization when replenishing distribution centers than when replenishing stores. It is important that the planning system can perform order optimization on multiple levels to reach the most cost-efficient outcome. Some examples of order optimization on different levels are:
Although it seems simple, the process of pooling orders for multiple products to fill load carriers or meet supplier order limits can be quite the test for your planning system’s flexibility. To meet supplier requirements and benefit from lower transportation costs or supplier discounts without accumulating excess stock, you typically need to be able to:
In addition to letting the planning system do the heavy lifting when it comes to supplier order requirements, the best practice is to constantly evaluate these restrictions and their impact on the flow of goods. Multi-year contracts in a dynamic market or fixed order restrictions for products with seasonal demand may turn out to be costly or infeasible as demand changes. To support this, the ideal planning system should highlight all order suggestions more/less than needed as a result of these constraints, as well as show the difference from the actual need. Furthermore, it should provide analytical support to help the operative buyers make rational decisions concerning whether the benefit, such as rebate, of meeting a supplier restriction is greater than the resulting increase in inventory carrying cost and risk of obsolescence. 6.2. Smart Buying Takes Advantage of Good PricesRetail costs are dominated by the cost of goods sold. The operative buying team needs to take responsibility for efficiently exploiting rebates to improve gross margins. In theory, smart buying when prices are changing is quite straightforward:
To be able to truly benefit from price changes, you also need to factor in your inventory carrying cost, time your orders correctly relative to when the price is changing, and potentially split the investment buy—the additional quantity you are buying above what you would need to meet demand—into several orders. To further complicate things, there may be other factors that have an impact on the optimal order quantity. For fresh products, shelf life is always a factor. It makes absolutely no sense to stockpile inventory that will end up as waste, or to harm your reputation by putting goods with unattractive expiration dates in stores. Furthermore, in situations where storage space is scarce, the cost of inventory may suddenly jump to a whole new level if you exceed the capacity limits of your current warehouses. When your storage space is very full, you would need to find additional space outside of your current warehouses for additional goods, quickly turning your investment buy into a very unprofitable move. The best practice is to feed your planning system with time-dependent price data to let the system optimize when and in what quantities to buy when prices are changing. In this way, you can take advantage of even minor price changes effectively, as the operative buyers do not need to spend time manually figuring out the optimal order quantities. It is important to keep in mind that restrictions, such as shelf life for perishable items or capacity limits on storage space, need to be considered. If your planning system is not able to deal with such restrictions automatically, the suggested investment buys will need to be double checked by the buying team. It is not unusual to have supplier contracts include a rebate triggered by the buyer’s annual order value exceeding a set quota. Again, keeping track of supplier quotas, placed orders, and forecasted orders is very hard to do manually. Intelligent planning systems support smart buying decisions by suggesting additional orders to get the rebate when feasible and by not suggesting any additional orders that would result in counterproductive stock-piling. 6.3. Batch-Level Inventory Management of PerishablesIt is currently impossible to know the exact expiration dates of on-hand inventory in stores. It can even be hard to get a decent estimate if there are several batches of a product simultaneously on the shop floor, as some consumers work hard to find the freshest products available. However, distribution centers are a different story. In distribution centers, modern warehouse management systems ensure that inventory is shipped on a first-in-first-out basis. In addition, they keep track of the exact expiration date for each batch in stock. Making good use of batch-level expiration data in inventory management reduces waste and improves your service level:
6.4. Real-time Data for Buying Fresh ProductsFor perishable products, a very high inventory turnover both in stores and in the supplying distribution centers is a must. This means that the supply chain is very vulnerable to quality issues, delivery problems or unexpected peaks in demand. In situations where store requirements exceed available inventory, quick reactions are of essence. In many cases, suppliers of short shelf life perishables make several daily deliveries to the same distribution centers. This is partly to guarantee freshness and partly to level out volumes. Several daily supplier deliveries make it possible for a retailer to accommodate actual demand by placing the orders as close to the different ordering deadlines as possible, making use of the latest demand and inventory data. However, to be able to identify demand surges, the planning system needs to be tightly coupled with the underlying transaction systems and have access to real-time data. Of course, the planning system also needs to be able to process the data quickly enough to turn the latest data into orders as accurately as possible. Similar quick reactions and within-the-day calculations based on real-time data are valuable when fruits and vegetables, which are prone to supply and quality issues, are received in the morning. Because the actual available inventory may differ from what is planned, it makes sense to re-allocate stock based on the latest forecast and stock data from the stores rather than fulfilling store orders in an arbitrary order. 7. Planning for Optimal Capacity and Resource Utilization Throughout the Grocery Supply ChainIn a dynamic business like retail, capacity bottlenecks can emerge in almost any part of the supply chain in response to a range of events, from holidays or unusual weather to promotions or big assortment updates in stores. To identify and proactively resolve these bottlenecks, retailers need to understand how the forecasted demand will impact inventory, capacity, and resource requirements throughout their supply chains. 7.1. Retail Sales & Operations Execution (S&OE)The S&OE process aims to ensure that retailers can fulfill short-term demand for the upcoming 0–3 months as cost-effectively as possible. The starting point is a very granular demand forecast at the SKU-channel-day level (see Section 3.1). From there, planners can use supply chain projections (see Section 5) to get a detailed understanding of inventory, capacity, and resource requirements throughout the supply chain. This end-to-end visibility into retail operations drives many S&OE benefits, including:
When dealing with millions of goods flows, tens of thousands of employees, hundreds of vendors, frequent promotions, and regular price and assortment changes, there are bound to be exceptions to any retailer’s plans. Many of these exceptions require immediate attention and quick adjustments to avoid or minimize any negative impact on S&OE. However, by turning to AI and advanced optimization, retailers can actually resolve most of these deviations without any human intervention. Best-in-class, autonomous issue resolution increases the speed and accuracy with which a retailer can manage exceptions by:
The best-performing retailers are able to combine human expertise with technology, quickly adapting to new situations and implementing new business priorities at scale. A good case example of this is German drug store retailer Rossmann, who took only two days to stand up entirely new planning configurations that prioritized the delivery of essential products to stores during 2020’s COVID-impacted demand shift. 7.2. Retail Sales & Operations Planning (S&OP)If the goal of retail S&OE is to resolve unforeseen capacity and resource challenges in the short-term, then retail S&OP, by contrast, looks further into the future. The goal of retail S&OP is to ensure sufficient capacity and resources to support future growth targets, planned changes in the distribution network, major seasons, and more. Preparing for major holidays provides an important use case for retail S&OP, with the winter holiday season being the most important (and challenging) in most markets. Holiday seasons are generally marked by higher-than-normal demand that increases steeply until the holiday peak. After that peak, demand usually drops back to normal—or even dips below normal levels for a while. But the retail S&OP process is about more than supply chain efficiency alone—it’s about maximizing profitability. S&OP should result in:
Retail preparations for a major holiday like Christmas usually start around six months before the season begins. The first step is to agree on the constraints: will there be changes in delivery schedules/lead times or supplier capacity constraints due to the holiday season? After agreeing to the constraints, the next step is reviewing sales and delivery plans to identify potential bottlenecks, which can emerge anywhere in the supply chain throughout the holiday season. Potential bottlenecks might include overly large deliveries adding to store congestion on busy days; days when warehouse staff face more order lines than they can pick; days when a warehouse receives more frozen products than it can store; and so on. Using spreadsheets, it would simply be impossible model the supply chain—with all its complexities—accurately enough to proactively identify bottlenecks. Even building a simplified model would be enormously time-consuming and prone to error. The only way to detect the types of moving capacity bottlenecks described above with any degree of certainty is to use supply chain projections. Once they’ve identified potential bottlenecks, retailers should use “what-if” scenario planning to examine and eliminate them. Bottom-up scenario planning allows retailers to see exactly how changes in delivery timing, replenishment schedules, or forecasted sales volumes would impact goods flow. Typically, food retailers need to deliver long shelf-life products to stores earlier to free up capacity to effectively manage fresh products in their high season. Different strategies and scenarios for leveling out the goods flow ahead of a holiday season include:
Retailers should use their software’s scenario planning capability to identifiy the scenario that best meets their goals and come to an agreement with their suppliers on it. This enables them to lock their plan well in advance of the season, so they can then focus on execution and corrective actions. An effective S&OP process leads to more level capacity utilization throughout the season. Furthermore, planners know of and can plan for any remaining peaks remain beforehand, rather than scrambling to manage costly surprises as they pop up. 7.3. Efficient Supplier CollaborationSupplier collaboration has been a point of discussion for decades, but surprisingly few retailers have implemented it extensively. To establish fruitful collaboration, both parties need to both put in effort and receive measurable benefits from the process. Unfortunately, because this has rarely been the case, many collaboration initiatives fail. While technology doesn’t solve the challenge of supplier collaboration, it can ease the pain. For example, most collaboration projects spend the majority of their effort just on collecting data from various sources, but the right planning system can minimize that work. Rather than trying to fix everything in one go, we recommend building your supplier collaboration processes bit by bit. A good starting point is sharing order forecasts with your suppliers because it’s a lean way to collaborate. If your planning system can calculate supply chain projections, then the purchase order forecast, which tells your supplier what you plan to buy from them in the weeks and months to come, is already readily available to you. A good system can send automated reports that share that information with your suppliers. You can also share relevant information on planned promotions, upcoming events, or other changes, helping your suppliers understand the reasoning behind your purchase order forecast. Retailers may also share demand forecasts or point-of-sale (POS) data with their suppliers, but the most essential information is what you expect the supplier to deliver and when. A more collaborative way of working requires both parties to recognize the value that investing their time and effort will bring. Whereas simply sharing a forecast is one-way communication, Collaborative Planning, Forecasting, and Replenishment (CPFR) is true two-way communication. A good planning system helps by providing reliable projections of future purchase orders, analytical tools to understand potential changes and issues, as well as a platform or portal for the collaboration. Ideally, suppliers can simply be given access to a retailer’s view of their projected demand, plans for purchase order placements, and data on promotions, seasons, events, etc., then add their own view. Combining the supplier’s holistic view into their categories and products with the retailer’s understanding of its business and marketing activities in this manner ultimately results in a more accurate overall plan. Best-in-class planning systems can support this kind of collaboration by providing a platform that can take in multiple forecast types, alert users to any differences, allow users to edit plans, and finally, disaggregate the agreed plan to whatever level of detail is needed—whether stores, products, or days—to support operational execution. 8. Conclusion: Team up with the Machines to WinRetail is in turmoil, and it is unclear what the impact of the different sales and delivery channels, store formats or even retailer players will be. In 10 – 15 years, we will probably look back at this time in amazement and wonder “How did we not see this coming?” Some predictions about the future of food retail are, however, easy to make:
To summarize, retail supply chains need to become more responsive and finely controlled than ever before to meet the demand for fresh products with minimum waste. At the same time, retail supply chains need to become more efficient by optimizing inventory flows from multiple perspectives—store operations, distribution, picking and warehousing—to meet the price pressure. This is only possible by teaming up with the intelligent machines. The world of food retail is too complex to be managed with notepads and intuition. This has, of course, been true for a long time. The breaking news is that not only are the simplest jobs being automated, but significantly more advanced planner roles are being filled by machines, too. More importantly, intelligent automation will not only replace manual work, but take planning to a level of granularity never before seen. Will there then be any role for humans in this brave new world? Yes, there will be plenty. Three important roles are:
So please, do not hold your breath waiting for AI to revitalize your retail business or even solve your supply chain challenges. But please do start phasing in the use of machine intelligence where most feasible and impactful. This collection of best practices is a good place to start. |