Methods for Product-Mix Decisions

Based on the demand- and cost-side product information presented above, more accurate product selection and shelf space allocation decisions can be made. Even though the PROFSET model introduced in this dissertation is a product selection model and not a shelf space allocation model, a concise literature overview of both types of models will be provided in this section. The reason is that product selection and shelf space allocation are closely related to each other. One can generally distinguish between three types of models.

The first type of models only treats product selection without dealing with shelf space. A second type of models only treats optimal shelf space allocation (e.g. [65, 66]), provided that the product selection problem is already carried out. Finally, still other models treat product selection and shelf space allocation simultaneously (e.g. [20, 44, 131]). In the first and second case, distinct models are used, first to select an optimal set of products, and second to allocate shelf space to the selected products, whereas in the latter case the problem of product selection and shelf space allocation are treated within one overall model.

Product-mix and shelf space allocation models in general can, however, again be divided into two classes: heuristic models and optimization models.

5.3.3.1 Heuristics for selection/shelf space allocation

As a result of the complexity and the cost to collect all relevant product related demand and cost information, many retailers rely on rather simple heuristics (rules of thumb) to select products or allocate shelf space, based on a number of quantitative criteria, also called indices of SKU productivity [43], such as unit sales, dollar sales, rotation speed, gross margin, contribution per m², DPP and many others. Furthermore, these measures typically do not account for product interdependence effects and therefore they do not reflect the dynamics in the store. Nevertheless, these indices serve as the input for product selection and shelf allocation methods where the focus is on simplicity and ease of use.

Some popular commercial systems include the PROGALI model [182], which allocates shelf space in proportion to total dollar sales. In the OBM model [203], shelf space is allocated proportional to the product’s gross profit. Other systems for shelf space allocation have concentrated on minimizing costs of inventory and handling, such as COSMOS [85], SLIM [76] HOPE [100] and ACCUSPACE [172]. CIFRINO [75] and McKinsey [193] combine both product revenues and costs to allocate shelf space in relation to DPP. However, none of them incorporates demand elasticities. Finally, SPACEMAN developed by ACNielsen, is also worth mentioning. On top of shelf space allocation, SPACEMAN visualizes shelf space allocation into store planograms.

With regard to product selection, the method of ‘product portfolio analysis’ is also worth mentioning. Although originally the idea of portfolio analysis was proposed in the context of multi-product firms, it seems like an interesting approach to evaluate the retailers existing portfolio of products along a number of important performance dimensions like sales, market share, profitability, growth potential, etc. A well-known product portfolio instrument is the Boston Consultancy growth-share matrix. This matrix is built along two dimensions, namely the product’s market share and the stage of the product in the product life cycle. The product’s market share reflects the cost advantage that a

-128-share generate more cash. The stage of the product’s life cycle is usually measured by its sales growth. The idea is that for high growth products, more cash will be needed to consolidate the growth, in contrast to low growth products. Swinnen [257] offers a critical review of the BCGS matrix within the context of a supermarket chain. He argues that it is very uncertain whether the BCGS approach is of practical usefulness. First of all, he argues, the cost advantage for products with relatively high market share is only guaranteed under the assumption that the experience curve holds and this has not yet been studied for distribution firms. Secondly, the objective to generate cash may not be the primary objective of supermarket chains, whose main resource appears to be the availability of shelf space. Thirdly, the product-portfolio approach is a rather strategic approach and it is therefore best suited for decisions on the level of product categories and thus not on the SKU level. To conclude, although the existing product portfolio approach seems less suited for product selection decisions in supermarkets, the BCG matrix may present an interesting approach to evaluate different product groups according to particular dimensions that are relevant to the supermarket retailer, such as image and contribution per m².

Finally, ABC analysis rank-orders manufacturer brands according to the perceived price and quality levels, advertising efforts, brand reputation and distribution coverage [275]. ‘A’-brands can be described as having a high perceived price and quality, a wide distribution coverage and a premium brand reputation. They primarily serve to support the retailer’s store image by taking advantage of the excellent image of the brand, often as the result of large investments in advertising by the manufacturer. ‘B’-brands have less reputation, less geographical spread and a lower perceived price and quality reputation than ‘A’-brands. They primarily serve to cover the lower-end of the product assortment. Finally, and mainly as the result of increased price competition during the last few years, ‘C’-brands are perceived as having a very low price and quality. They are usually distributed by a single retail chain and in most cases the manufacturer does no longer invest into advertising such

products. The categorization of manufacturer brands into ABC groups may therefore serve as a guidance to select particular products into each category.

5.3.3.2 Optimization models for selection/shelf space allocation The objective of constrained optimization models for product selection is to use mathematical or operations research procedures to find a product mix that maximizes a particular objective (like profit) subject to a number of constraints (like limited shelf space) using quantitative data. However, the term

‘optimization model’ may be somewhat misleading in so far that, as a result of the complexity of the product selection problem, the overall best model probably does not exist. In fact, each model aims at finding the best solution relative to the objectives, constraints and product data available.

Probably one of the first models for product selection was the model by Anderson and Amato [20]. They introduced a mathematical optimization model for simultaneously determining the optimal set of products to choose from a large set of available products, together with the amount of display space to allocate to each selected product. Their model is similar to the model presented in this dissertation in so far that our model also aims at selecting the most profitable subset of products out of a larger set of available products.

However, in contrast the PROFSET model, their model does not take into account cross-selling effects between products. On the other hand, they include brand-switching behaviour into their model, which we do not.

Unfortunately, they did not test their model on real data.

In 1979, Hansen and Heinsbroek [131] developed a product selection and shelf space allocation model taking into account the space elasticity of sales and a number of constraints related to the minimum amount of shelf space allocated to selected products. Their model, however, does not include substitution and complementarity effects between products. They argue that the information needed to estimate the demand interdependencies for supermarkets was not available at that time and that the large number of such

-130-impossible in practice. The objective of their model is to maximize profit, taking into account the unit margin of products, product demand in function of allocated shelf space, the unit cost of space and the cost for replenishment of the shelf stock.

The absence of cross space elasticities in most product selection models has motivated Corstjens and Doyle [83] to present a model for optimizing retail space allocations where both main and cross-space elasticities were considered.

They used cross-sectional data to estimate the elasticities for a small set of 5 product categories, including chocolate confectionary, toffee, hard-boiled candy, greeting cards and ice cream. The results of their model show that ignoring cross-elasticities may lead to a major suboptimalization in the allocation procedure. Later, the Corstjens and Doyle model was also implemented by Swinnen [257] on cross-sectional data from twenty-seven stores belonging to one supermarket chain in Belgium. The study implemented the model on ten product groups and found significant positive cross-space elasticities between canned tomatoes and spaghetti, and between Knorr Royco dry soup and tomato soups.

Bultez et al. proposed the S.H.A.R.P [65] and S.H.A.R.P II [66] model. They tried to maintain the parsimony of the Corstjens and Doyle model and also incorporated both direct and cross-space effects and modelled costs as a function of sales per unit space.

The issue of product selection was also studied by Borin et al. [44]. They also included main and cross-space elasticity effects into their model.

Furthermore, they argued that the effect of stockouts and assortment decisions should also be considered in a profit optimization problem. Indeed, a stockout of a particular item may influence the sales of substitute items in the same category (known as stockout demand), and an item’s sales may also increase due to the non-selection of a particular product as a result of switching behaviour by consumers (known as acquired demand). Given the huge number of considered products in our model, stockout demand and acquired demand are not included in our optimization framework.

5.4 PROFSET: An Optimization Framework for