You are on page 1of 11

ARTICLE IN PRESS

Int. J. Production Economics 104 (2006) 462472 www.elsevier.com/locate/ijpe

Inventory control of perishables in supermarkets


K. van Donselaar, T. van Woensel, R. Broekmeulen, J. Fransoo
Department of Technology Management, Eindhoven University of Technology, P.O. Box 513, Eindhoven 5600 MB, The Netherlands Received 16 April 2004; accepted 21 October 2004 Available online 13 September 2005

Abstract In the last decade large supermarket chains have started developing and implementing automated store ordering (ASO) systems in order to improve both the quality and efciency of the inventory replenishment decisions in their stores. These ASO systems have been developed mainly for non-perishables. The aims of this paper are 1. 2. 3. 4. to better understand the product, sales and supply characteristics of perishables in supermarkets, to dene and operationalise relevant concepts in controlling the inventories of perishables, to investigate how the intelligence in ASO systems in supermarkets can be further improved, to give academics empirically based insight in the relevance of the inventory models that are currently available in the literature, insight in relevant parameter settings and insight in which models are needed.

The research is carried out in cooperation with two Dutch supermarket chains and focusses on perishable items. To further improve the intelligence in the ASO systems, we will classify the items in a supermarket, based on the notion that different items need different inventory control rules. For every class of items, we describe the main logistic characteristics; moreover, for the perishable items we describe the inventory control rules which are currently being applied and we discuss the desired inventory control strategy for these items. r 2005 Elsevier B.V. All rights reserved.
Keywords: Inventory management; Perishables; Empirical data; Classication; Retail; Model; Distribution

1. Introduction 1.1. Research environment The annual sales in Dutch supermarkets are approximately 24 billion h/year. The two supermarket chains involved in this research project together have more than 500 supermarkets and have a joint market share of approximately 20%.

Corresponding author. Tel.: +31402472869; 31402464531. E-mail address: k.h.v.donselaar@tm.tue.nl (K. van Donselaar).

fax:

0925-5273/$ - see front matter r 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.ijpe.2004.10.019

ARTICLE IN PRESS
K. van Donselaar et al. / Int. J. Production Economics 104 (2006) 462472 463

They typically sell food products and personal care products. The main difference between the supermarket chains is in their marketing strategy. Both supermarket chains aim for high product availability and low costs, but one of them also aims for high customer service in terms of a broad assortment. Their stores are medium sized to big (often in the range from 500 to 1500 m2 selling space). From both supermarket chains we received large databases with point-of-sales (POS) scanning data and data on planograms (i.e. which item is on which shelf with how many facings and what is the maximum number of products that t there), product information (like article number, EAN code, shelf life, product dimensions, case pack size, etc.), promotions and delivery schedules. 1.2. Scope In this paper, we restrict ourselves to the mainstream inventory replenishment decisions for perishables in the supermarkets. With mainstream we exclude the following situations:

how, several key logistic parameters are calculated for both classes of items in Section 2. Next, in Section 3, a further classication is proposed within the class of perishables. After a description of the inventory control strategies currently applied in practice in Section 4, we will discuss which inventory control strategy is proposed for each class of perishables in Section 5. These proposed strategies are then compared with the models available in the literature. Based on the proposed control strategies, we will also give improvement suggestions for current automated store ordering (ASO) systems. The paper ends with the main conclusions.

2. Perishables versus non-perishables The main difference between perishables and non-perishables is the Shelf Life. The shelf life of a product is measured in days, counting from the day it is produced until the product becomes unacceptable for consumption or obsolete. This end date has a strict relation to the date mentioned on the product. Clearly, perishables are items with a short shelf life and non-perishables are items with a long shelf life. An item is considered perishable if it meets one of the following two criteria:

  

Promotions Product introductions and product declines Products that are being sold in a very short period only (like Christmas chocolates or onetime items, which are being pushed by the Purchasing Department)

Furthermore, we primarily focus on perishables that are delivered via the retailers DC, since data availability for these items is better compared with items that are delivered directly by the supplier to the supermarket. We exclude products that have a price which is weight dependent and different per customer (e.g. because they are being weighed in the store by the store clerks (like fresh meat) or the customers (certain vegetables)). We exclude these items since the transaction size (necessary to compare with sales in consumer units for other items) is not available in the database. 1.3. Structure of the paper To show that perishables are clearly different from non-perishables and to better understand

The high rate of deterioration at ambient storage conditions requires specic storage conditions at the store and/or at the consumer to slow the deterioration rate. Frozen food is excluded since the rate of deterioration is very low in the freezer. The obsolescence date of the product is such that reordering for the products with the same date is impractical (e.g. periodicals, like newspapers and (bi)weekly magazines).

To classify every individual item as perishable or not, we need to determine the threshold value for the shelf life. Examination of the data has revealed that all products which require a conditioned environment (e.g. as stated on the packaging by the manufacturer) have a shelf life that is equal to or less than 30 days. Consequently, we dene

ARTICLE IN PRESS
464 K. van Donselaar et al. / Int. J. Production Economics 104 (2006) 462472

perishables as items with a shelf life less than or equal to 30 days. If we apply this denition to the entire assortment (including here weight products and direct deliveries), we estimate that roughly 15% of the assortment are perishables versus 85% non-perishables. Based on this distinction between perishables and non-perishables, we show that both groups are really different in terms of several key sales and logistic product (group) characteristics:

  

  

average sales per week (in consumer units (CU)), coefcient of variation of weekly sales, potential delivery frequency, measured by the average number of days between two deliveries for the product group to which a product belongs (note a specic product may actually be delivered with a lower frequency), shelf life (in days), average case pack size (in consumer units), minimum stock norm (potentially including both commercial and logistic requirements; expressed in consumer units).

For six supermarkets (three for each supermarket chain), we determined for all items in the assortment the median for the above-mentioned characteristics (Table 1). As the data involved are importantly non-normally distributed, we use medians to describe the two groups under consideration. In this situation, the median is a better measure of the central location of each group rather than the mean as this measure is heavily affected by the impact of outliers. The statistical difference between perishables and non-perishables is tested using the median test
Table 1 Medians of sales and logistic (product) characteristics

(Mood et al., 1974). The null hypothesis of the median test states that the two groups (perishables versus non-perishables) have the same median, i.e. there is no statistical difference between both. The test statistic obtained is chi-squared distributed, which is compared to a critical chi-squared value with 1 degree of freedom (number of different groups minus 1 2 1: The critical value for a signicance level of 99% is 6.6394. Looking at the test statistic values, we conclude that we can reject the null hypothesis in all cases. The comparison in Table 1 shows that perishables typically have a weekly sales, which is approximately 50% higher than non-perishables. Combined with the fact that their median case pack size is approximately 40% smaller, we may conclude that the time between two orders for a perishable item is, roughly speaking, 2.5 times as small when compared with a non-perishable item. The fact that the median potential delivery frequency is only slightly different for perishables and non-perishables is due to the fact that this frequency is determined by the overall sales volume of the product groups, rather than by the characteristics of individual items. The overall sales volume for perishables (in CU) is relatively small and divided in multiple distribution channels (e.g. direct delivery for bread versus indirect delivery for most vegetables and fruits). Yet, the pressure on time forces the supermarkets to get perishables delivered with this high frequency. The coefcient of variation of weekly sales is slightly smaller for perishables, which makes sense since the average weekly sales for perishables is also higher and, as noted by Silver et al. (1998), often the relative demand uncertainty decreases if the average demand increases. Also, the minimum

Perishables Average weekly sales per product (in CU) Coefcient of variation of weekly sales Potential delivery frequency (in days) Shelf life (in days) Average case pack size (in CU) Minimum inventory norm (in CU) 11.9 0.345 1.2 21 6 3.3

Non-perishables 7.9 0.377 1.3 240 10 4

Test statistic 116.24 27.17 82.35 1050.00 523.50 47.49

ARTICLE IN PRESS
K. van Donselaar et al. / Int. J. Production Economics 104 (2006) 462472 465

inventory norm (in CU) for perishables is signicantly smaller than for non-perishables. Combining this with the fact that average sales are larger for perishables shows that the difference in minimum inventory norm, when expressed in days sales, is even bigger (the median for this norm is 1.3 days for perishables and 3.3 days for nonperishables). Based on this analysis, we can conclude that in terms of a number of sales and logistic (product) characteristics (shelf life, average weekly sales, coefcient of variation for weekly sales, case pack size, average time between two replenishment orders and minimum inventory norm) perishables are clearly different from non-perishables.

3. A classication within perishables In the remainder of this paper, we focus on how to control the inventories for the perishables. The control of these inventories will be aimed at reducing the major cost factor for perishables like waste. Waste stems from excessive inventories, which either have to be marked down just before the sell-by-date or thrown away after the sell-bydate. In both cases, the nancial consequences for the retailer are severe, assuming there is no return policy from the supplier.1 For items with a short shelf life, we distinguish three options to reduce the amount of waste:

  

reduction of leadtime and/or review period, demand substitution, limited assortments.

A typical example of an item with a very small shelf life is bread. Bread can only be kept fresh in the supermarket for one day. In order to maximise the freshness of the bread in the supermarket and to minimise waste, bread is typically delivered directly (from the industrial bakery to the supermarket) on a daily basis. For items with a slightly
1 An exception to this rule are items with a very low salvage value and/or a very high prot margin; for those items, the focus should be on maximising sales rather than minimising waste.

higher shelf life, cross-docking at the retailers DC may be an option. Both direct delivery and crossdocking aim to reduce the leadtime. If the supermarket manager wants to have a high service level for all types of bread, he needs high inventories for all of them and since the demand is uncertain, he will typically have plenty of leftovers at the end of the day. The alternative is that the store manager uses the customers willingness to substitute demand if his/her preferred item is out of stock. According to an expert, bakers typically aim for a low percentage of leftovers at the end of the day (12% of sales for independent bakers and a somewhat higher percentage for bakery departments in supermarkets). They typically aim to avoid leftovers for costly items, i.e. items with costly ingredients (like multiple-grain breads) or items with high added value (like cut bread (versus uncut bread) and breads which require a lot of space in the furnace). Another option, which combines reduction of leadtime and demand substitution, is to offer customers another type of bread, which is bakedoff in the supermarket at the end of the day, when the customers preferred bread is out of stock. The raw material for this bread is stored in the refrigerator and can be preserved for many days. Finally, if assortments are kept relatively small, the average daily demand per item is relatively large. As a result, demand will be relatively certain, limiting the total amount of waste in a product category. We note that, although for some perishables direct delivery or cross-docking is needed, this is not the generally preferred way of distribution. Multinational retailers in Europe often distribute their goods via a DC to the supermarkets (Bourlakis and Bourlakis, 2001). Supermarkets in The Netherlands also prefer shipments via the retailers DC in order to minimise the number of deliveries at their backdoor. Each delivery costs a considerable xed amount of time (i.e. independent of the quantity delivered) for both the store clerk receiving the goods and the transportation company delivering the goods. Moreover, the accessability of supermarkets decreases as city trafc regulations impose time windows that restrict the deliveries. Apart from costs, also other

ARTICLE IN PRESS
466 K. van Donselaar et al. / Int. J. Production Economics 104 (2006) 462472

important aspects (like quality and its impact on consumer buying behaviour) have to be considered when deciding on the distribution strategy. In general, with a more direct delivery strategy the quality of the products in the store will increase, which will have a positive impact on sales. To quantify this trade-off, detailed knowledge is needed per product group on the relationship between time and quality (how quickly the product deteriorates) as well as between quality and sales (how many more products will be sold as a function of increased quality). Usually, most customers buy the items they intend to consume in the coming week, once a week. Consequently, the remaining shelf life (RSL) should at least be a week to facilitate for this behaviour. The RSL is dened as the number of days a product is still available for consumption starting from the moment the product arrives in the supermarket. Remember that the previously dened shelf life starts from the moment of production. The (RSL) will be a function of the shelf life, the distribution strategy (including, e.g. decisions on direct delivery, cross-docking or delivery from stock at the retailers DC and the shipping frequency) and the inventory replenishment logic (e.g. push or pull): RSL f Shelf life; distribution strategy, inventory replenishment logic. Since many consumers only visit the supermarket once a week (Kahn and Schmittlein, 1989), supermarkets typically want to offer consumers products which can be consumed at least for one more
Table 2 Medians of sales and logistic (product) characteristic

week. As such, supermarkets try to reach an RSL norm of 7 days. If however the difference between shelf life and the RSL norm is only 1 or 2 days, then delivery from stock at the retailers DC often takes too much time. In those cases direct delivery (although not preferred) or cross-docking (although implying higher labour costs at the manufacturer) may be the solution to still meet the RSL norm. From the reasoning above it is clear that shelf life is the main factor which puts pressure on lead times and which drives the need for substitution. Therefore, we propose a further classication within the perishables, based on shelf life. The majority of perishable items for which we observed a cross-dock operation or direct delivery typically have a shelf life less than or equal to 9 days. Examples are: bread, newspapers, weekly magazines (direct deliveries) and owers, exotic fruits, ready-to-cook vegetables and some dairy products (cross-docking). Therefore, we distinguish between perishables with a shelf life p 9 days, called Days Fresh (DFs) and perishables with a shelf life between 10 and 30 days, called Weeks Fresh (WFs). After dividing the perishables into DFs and WFs, based on this shelf life threshhold, we calculated the same product and sales characteristics as in the previous paragraph for both classes (Table 2). Comparing the obtained test statistic values with the chi-squared critical value (1 degree of freedom), shows that we are not able to reject the null hypothesis for most variables under consideration. Consequently, based on the dataset we cannot conclude that there is a signicant

Days Fresh n 269 Average weekly sales per product (in CU) Coefcient of variation of weekly sales Potential delivery frequency (in days) Shelf life (in days) Average case pack size (in CU) Minimum inventory norm (in CU) n is the number of observations. 11.3 n 96 0.371 1.2 4 6 2.7 n 31

Weeks Fresh n 1606 11.9 n 1328 0.344 1.2 21 6 3.3 n 1043

Test statistic 0.04 1.01 4.59 38.80 15 5.53

ARTICLE IN PRESS
K. van Donselaar et al. / Int. J. Production Economics 104 (2006) 462472 467

difference between both medians. The main reason that we do not nd a statistical difference between DFs and WFs is due to the relatively low number of DF items in our integrated database having sales data, which indicate that this sample for DFs may not be representative for the whole group. In reality there are much more DF products being sold in the supermarket. The sales data for these items are available in local POS databases per supermarket. However, they cannot always be linked to a central product database since the central product database and the local POS database use different article numbers for the same item. Normally, there is a separate central database which provides a link between the two article numbers, but for the directly delivered items and for owers and some vegetables and fruits there is no such link. For owers e.g. the purchase orders are registered for each individual type of ower (like yellow rose) but the sales in the stores are registered only at a higher aggregation level (like h 2.99 owers). For vegetables and fruits the opposite may hold: the POS database sometimes contains more items than the central product database. This might not be a problem if the two databases could be linked via their product descriptions. Unfortunately, the product description in the two databases may be slightly different, preventing an automated link between the two databases based on description only. Moreover, the product description in the local POS database is not unique (the description white round bread may simultaneously be used for an article number which refers to a 400-gram bread and for an article number which refers to an 800-gram bread) or the same product may have different descriptions and/ or different article numbers in different stores due

to different local suppliers (in case of direct deliveries). Consequently, the only way to test the hypothesis that DFs are clearly different from WFs with respect to sales characteristics is to look into detail at the local POS data. This is done for one store only, since we had to select manually (based on the item description) all DFs. The POS le for this store contains 7173 items and 562 of them are DFs. We rst categorised these items into DFs and WFs. The median of the average sales and coefcient of variation for DFs and WFs are given in Table 3. The table shows that DFs have relatively large average sales and low coefcient of variation. Next, we categorised the DFs in this store into Bread, DF-media (newspapers and weekly magazines), DF-dairy (dairy-products with shelf life p 9 days) and Ready-to-cook vegetables (containing mostly cut vegetables and some pealed-and-cut fruits). The median of the average sales and coefcient of variation for these subcategories are also given in Table 3. The table shows that Bread and DF-dairy products have relatively large average sales. Ready-to-cook vegetables and DF-media have relatively low sales when compared to the other DFs. The DF-dairy products, which are being cross-docked, are mainly fastmovers. They are typically handled in crates of 20 CU or in rolling containers of 160 CU which are driven into the dairy section in the supermarket. This reduces the handling costs in the store signicantly. The coefcient of variation for Bread is unexpectedly high (given the high average sales). Further analysis revealed that the main cause for this is price promotions, which were still included in the POS data we analysed.

Table 3 Sales characteristics for different types of perishables Days Fresh n 218 Weeks Fresh n 721 Bread n 137 DF-dairy products n 20 62.4 0.25 Ready-to-cook vegetables n 28 13.2 0.42 DF-media n 33

Average weekly sales per product (in CU) Coefcient of variation of weekly sales

33.8 0.38

19.2 0.58

45.6 0.39

16.8 0.41

ARTICLE IN PRESS
468 K. van Donselaar et al. / Int. J. Production Economics 104 (2006) 462472

The general conclusion from these analyses is that perishables can be divided into DF items (shelf life p 9 days) and WF items (shelf life between 10 and 30 days), where DF items typically

supermarkets might use different rules for different product types. 4.1. ASO systems in current practice In the current ASO systems in both supermarket chains no distinction is being made between perishables and non-perishables. The ASO system is primarily designed and used for the non-perishables. The underlying logic is based on a (R, s, nQ)reorder policy with a dynamic reorder level s. The reorder level s is based on a demand forecast for the coming L R days L R being the sum of the leadtime and the review period). The demand forecast itself was based on a simple exponential smoothing technique. In special sales periods (like Eastern), the demand forecast was adapted. An important input for the demand forecast is the weekly pattern (see Fig. 1). The weekly pattern for sales in Dutch supermarkets; most sales (as a percentage of total weekly sales) are on Fridays and Saturdays. The ASO system gives an advice for every item on when and how much to replenish. Often the store clerk visually inspects the shelves typically looking at the fastmovers, strange order advices and items which are physically out-of-stock, and decide whether or not to follow this advice (since the retailers inventory data are known to be very inaccurate

  

need short leadtimes, often resulting in crossdocking at the DC or direct delivery from supplier to the supermarket, often have high average daily sales per item (which can be kept high by keeping assortments relatively small), and should have an inventory control rule which takes into account the substitution effects within a product category.

4. Comparison with current practice To describe the current practice in inventory control for perishables, we have to distinguish between inventory control via ASO systems and other inventory control. In both supermarket chains a recently developed ASO system was (being) implemented in a number of stores, while in other stores they still used an older (lessadvanced) system or relied on a store clerk with an orderbook to decide on the replenishments. In the last case, typically different store clerks in different

30,00% 25,00% 20,00% 15,00% 10,00% 5,00% 0,00% Mon Tue Wed Thu Fri Sat
Fig. 1. The weekly pattern for sales in Dutch supermarkets.

ARTICLE IN PRESS
K. van Donselaar et al. / Int. J. Production Economics 104 (2006) 462472 469

(Raman et al., 2001). This visual inspection is especially crucial for perishables because of the risk of getting too much waste. If we compare the reorder logic in the ASO system with our proposed inventory control for the perishables, we notice that the ASO system can be enriched by including

made for thousands of items a day. Analysing the impact of weather conditions or price on demand for each individual item in each supermarket in a thorough way will therefore be virtual impossible for a store clerk. Further research is needed here to see how these insights can be generalised and incorporated in ASO systems.

the substitution effect for items with a small shelf life (e.g. by checking the total amount ordered within a category with total average daily sales and differentiating the service levels of different items within a category) and/or a shelf-life-based upperbound on the maximum inventory level.

5. Proposed inventory control and comparison with literature We will discuss which inventory control strategy per group of product categories, is proposed and which part of the academic literature on inventory control systems is most relevant. The section ends with improvement suggestions for current ASO systems. 5.1. Bread and DF-media

4.2. Other inventory control systems in current practice As mentioned above, typically different store clerks in different supermarkets might use different rules for different product types. In one supermarket, e.g. the dairy products were ordered in a different way compared to the vegetables and fruits. Actually, even within the dairy products group, one clerk used two different systems depending on whether the products were ordered directly or via the retailers DC. When interviewing store clerks doing the ordering, they mentioned that for dairy products they recognised the danger of waste and therefore they paid extra attention to demand forecasting and inventory control; they kept a separate order and sales administration and they typically looked at sales data on comparable days in recent weeks to make a demand forecast. The clerk responsible for ordering the vegetables and fruits mentioned that he made a demand forecast based on his individual judgment, taking into account several factors like price, quality of the items (e.g. related to the country where the vegetables come from), weather conditions, promotions, time of the year, etc. Although the store clerks take into account a lot of factors, the problem is that each individual uses his own system and his own logic. The time available to make order decisions is very limited, since in a store these decisions typically have to be

Within DF items we note that certain items typically can be ordered only once every day (bread and newspapers) or once every week (weekly magazines) and this frequency exactly matches their maximum RSL. So the products which are on the shelves will never have different ages. Moreover, DF items may typically face substitution in case a stock-out situation occurs within the product category. Therefore the oneperiod multi-item newsboy problem with substitution applies here. There are only a few papers in the literature that are devoted to this problem. McGillivray and Silver (1978) and Parlar and Goyal (1984) assumed that products have identical costs. More recent papers allow for non-identical costs. See for example Smith and Agrawal (2000), who study the determination of a product assortment and associated inventory levels in case of demand substitution. Rajaram and Tang (2001) and Netessine and Rudi (2003) do not include the product assortment decision, but focus on the determination of the optimal order quantities. Netessine et al. study the difference between centralised and decentralised decision making in case of demand substitution. Rajaram et al. develop a heuristic to determine the order

ARTICLE IN PRESS
470 K. van Donselaar et al. / Int. J. Production Economics 104 (2006) 462472

quantities and investigate the impact of demand substitution on both order quantities and expected prots. In the literature most models for the situation with multiple items and demand substitution focus on the product assortment decision and assume that demand is independent of inventory levels. The paper of Mahajan and van Ryzin (2001) is an exception to this. They formulate a general mathematical model which can be used together with several consumer choice models. They also provide three different algorithms to solve the model. They applied the algorithms to a limited set of examples and show that the benet of using the most advanced algorithm compared to the classical Independent Newsboy algorithm (which is somewhat comparable to the methods used in ASO systems for dry groceries in practice) varies between 1% and 12% higher prots. Most authors stress that the challenge in application of their models and heuristics is in the empirical determination of the substitution pattern of customers. Smith et al. mention two easy-way-outs for this: either assuming demand allocation in proportion to the overall market shares of the available items or obtaining subjective estimates from retail buyers. 5.2. DF-dairy products, ready-to-cook vegetables These items may have products with different ages on the shelves. Moreover, since their shelf life is small (approximately one week), the danger for waste is clearly present. Therefore, the supermarket chains may not only apply direct delivery or cross-docking, but also aim for substitution between products in the same category. The peak demand is on Friday and Saturday and the RSL norm for Ready-to-cook vegetables is approximately one week. Therefore it seems to make sense, if the main focus is to minimise waste, to order most of these items just before Friday and to rely to some extent on substitution from Monday till Thursday. If, on the other hand, the main focus is to maximise customer service, the best way may be to either accept the waste or reduce the case pack sizes, e.g. by introducing case packs in which two or three slow moving items are combined.

On comparison with the literature, we notice that there is a large stream in the literature focussing on perishability and a small stream focussing on demand substitution (see also discussion above). The only paper, which combines the two is written by Parlar (1985). For an extensive review of the literature on inventory systems with perishable items we refer to Nahmias (1982), Raafat (1991) and Goyal and Giri (2001). Due to the perishability the optimal reorder rule no longer is of a simple form: rather than only taking into account the total inventory position, the reorder quantity depends on detailed information about the number of items in stock per age. This makes the analysis of these systems very complex. Adding the substitution aspect to it, makes it even worse to analyse. 5.3. Weeks Fresh items From the shelf life perspective direct delivery or cross docking is not a necessity for these items. Substitution and perishability may again play a role in controlling the inventories for these items, although the relative impact of both tends to be smaller compared with DF items. A general remark should be made about some key aspects that were encountered in practice, which are often not taken into account in the modelling. These include the weekly sales pattern and the inventory data inaccuracy (see Raman et al., 2001). In general, it is assumed that demand is constant over time and inventory data are 100% accurate. Incorporating these two aspects will make the mathematical tractability even worse. To circumvent the rst aspect, Kok and Fisher (2003) suggest the use of simulation-based optimisation. How to best circumvent the second aspect is unknown at the moment. Further research is needed here. Based on the inventory control strategies proposed above, the following improvement opportunities arise for ASO systems in practice. The ASO systems should include the ability

to check the aggregate order level for a group of substitutable items (this is the easiest indicator for the expected amount of waste; if one day

ARTICLE IN PRESS
K. van Donselaar et al. / Int. J. Production Economics 104 (2006) 462472 471

 

 

1000 breads are ordered while normally only 700800 breads are sold, the waste will probably be very large), to differentiate the target service level per substitutable item, to register the exact time of a stock-out for a substitutable item (if this is structurally getting earlier, it indicates demand for this item is increasing and the order quantity for this item could be increased), to register the daily amount of waste and/or markdowns per perishable item, to take into account the weekly sales pattern when forecasting demand for a perishable item.

markets). Since current automated store ordering systems provide little decision support for perishables, suggestions for improving these systems are given in this paper. In recent years, renewed attention is devoted in the literature to the multi-item newsvendor problem with substitution, which applies to the product categories Bread and DF-media. Other perishables however need more complex inventory control models, which simultaneously take into account aspects like the weekly sales pattern, perishability, substitution and/or data inaccuracy. Little research is done so far for these type of items.

Whether the ASO systems should also include other aspects like price elasticity, weather forecasts and seasonality when forecasting demand for the very near future for perishables, remains an open question that needs additional research.

References
Bourlakis, M., Bourlakis, C., 2001. A case study on the Greek multiple food retail sector. Supply Chain Management: An International Journal 6 (4), 189200. Goyal, S., Giri, B., 2001. Recent trends in modeling of deteriorating inventory. European Journal of Operational Research 134 (1), 116. Kahn, B.E., Schmittlein, D.C., 1989. Shopping trip behavior: An empirical investigation. Marketing Letters 1 (1), 5569. Kok, G., Fisher, M., 2003. Demand estimation and assortment optimization under substitution: Methodology and application. Working paper. Mahajan, S., van Ryzin, G., 2001. Stocking retail assortments under dynamic consumer substitution. Operations Research 49 (3), 334351. McGillivray, A., Silver, E., 1978. Some concepts for inventory control under substitutable demand. INFOR 16, 4763. Mood, A.M., Graybill, F.A., Boes, D.C., 1974. Introduction to the Theory of Statistics. McGraw-Hill, Kogakusha, Tokyo. Nahmias, S., 1982. Perishable inventory theory: A review. Operations Research 30 (3), 680708. Netessine, S., Rudi, N., 2003. Centralized and competitive inventory models with demand substitution. Operations Research 51 (2), 329335. Parlar, M., 1985. Optimal ordering policies for a perishable and substitutable product: A Markov decision model. INFOR 23 (2), 182195. Parlar, M., Goyal, S., 1984. Optimal ordering decisions for two substitutable products with stochastic demand. OPSEARCH 21, 115. Raafat, F., 1991. Survey of literature on continuously deteriorating inventory model. Journal of the Operational Research Society 42, 2737.

6. Conclusions Based on empirical data, we showed that supermarkets have different product groups, each having different logistics characteristics and therefore need different inventory control rules. The most obvious and statistically signicant difference is between perishables (items with a shelf life p 30 days) and non-perishables. The median for perishables typically shows higher average sales, a smaller case pack size, lower coefcient of variation of weekly sales, lower potential delivery frequency and a lower minimum inventory norm. Within the perishables a further distinction can be made between Days Fresh (DF) items (with a shelf life less than 9 days) and Weeks Fresh items. DF items typically need short leadtimes, often resulting in cross-docking at the DC or direct delivery, and often have relatively high average sales per item. Substitution is one way to help reduce the risk of waste for these items. Substitution can be especially benecial for items with a short shelf life (like bread) or for items with a case pack size, which is relatively large compared to the average demand during the shelf life (like slowmoving ready-to-cook vegetables in small super-

ARTICLE IN PRESS
472 K. van Donselaar et al. / Int. J. Production Economics 104 (2006) 462472 Silver, E., Pyke, D., Peterson, R., 1998. Inventory Management and Production Planning and Scheduling. Wiley, Toronto, p. 126. Smith, S., Agrawal, N., 2000. Management of multi-item retail inventory systems with demand substitution. Operations Research 48 (1), 5064.

Rajaram, K., Tang, C., 2001. The impact of product substitution on retail merchandising. European Journal of Operational Research 135, 582601. Raman, A., Dehoratius, N., Ton, Z., 2001. The Achilles heel of supply chain management. Harvard Business Review 79, 136152.

You might also like