ISSN17487595(Online)

Working Paper Series

Analysing the Impact of Supermarket Promotions: A Case Study Using Tesco Clubcard Data in the UK
Melanie Felgate Kent Business School Andrew Fearne Kent Business School Salvatore Di Falco London School of Economics and Political Science

Working Paper No. 234 January 2011

ANALYSING THE IMPACT OF SUPERMARKET PROMOTIONS: A CASE STUDY USING TESCO CLUBCARD DATA IN THE UK

Dr Melanie Felgate Research Associate Centre for Value Chain Research, Kent Business School, University of Kent, Canterbury, Kent, CT2 7PE, UK Tel: +44(0)1227 824766 Email: m.felgate@kent.ac.uk

Professor Andrew Fearne Professor of Food Marketing and Supply Chain Management Director of the Centre for Value Chain Research, Kent Business School, University of Kent, Canterbury, Kent, CT2 7PE, UK Tel: +44(0)1227 824840 Email: a.fearne@kent.ac.uk

Dr Salvatore DiFalco Lecturer in Environment and Development London School of Economics and Political Science, Houghton Street, London, WC2A 2AE, UK Tel: +44(0)20 7852 3778 Email: s.difalco@lse.ac.uk

 ABSTRACT The aim of this paper is to show how supermarket loyalty card data from a sample of over 1.4 million UK households can be used to analyse the effect of promotions on the sales of specific products and across different shopper segments. Regression analysis is used to test four hypotheses:

H1: Positive promotional impacts at the product level do not systematically result in category growth. H2: Promotional impacts are sensitive to the specific mechanic used H3: Promotional impacts are sensitive to specific product characteristics H4: Promotional impacts are sensitive to the demographic characteristics of specific shopper segments
The analysis compares the effects of different promotional mechanics upon different cuts and tiers of product across the fresh beef category in Tesco the largest supermarket chain in the UK. The results show that promotions do not always lead to category growth, there is considerable variability in the impact of different promotions mechanics across different products, and households respond differently depending upon their life-stage. The paper demonstrates the potential benefits of the analysis of loyalty card data for own-label products in destination categories to ensure that retailers and their suppliers are implementing the most effective promotional strategies. The evidence from the data analysis suggests this may not always be the case.

 INTRODUCTION This paper reports the use of loyalty card data from one of the biggest retailers in the world Tesco - to analyse the impact of promotions. While the case study is based on data from the UK, it demonstrates how other major retailers with similar loyalty card schemes (e.g. Kroger in the USA, and Casino in France) could make use of this powerful source of behaviour data to gain a better understanding of the effectiveness of promotions in key (destination) categories. The Tesco ClubCard is used regularly by 17 million households in the UK (approximately 40 percent of all UK households) and provides data not only on what is being sold on a national basis but also who is buying it, with shoppers segmented by television advertising region, lifestage, lifestyle (Tesco have their own lifestyle segmentation based on the distribution of products in a shoppers basket over a twelve week period) and geo-demographics. Panel data of this type or on this scale has not previously been used in any prior academic studies on the impact of promotions. Thus, this paper provides a contribution to the promotions literature by enhancing the empirical evidence base and demonstrating the value of analysis that is highly disaggregated by product type and shopper profile. The example used in this paper is fresh meat a key destination category for most supermarket chains, with a particular emphasis on fresh beef, which in the UK accounts for 56%of fresh meat sales. However, the model used could be applied to any product category in order to help retailers and food manufacturers make better (more informed) decisions with regard to promotional planning. In particular, by identifying how different types of shoppers respond differently to different types of promotions, retailers and manufacturers will also be able to target their promotional activity more effectively. The use of promotions in retailing has increased rapidly in recent times (Nielsen Wire 2009), yet more often than not promotions are being implemented with an inadequate understanding of which mechanisms are most effective, for which products and for which shopper segments. Despite this growth in the use of promotions, particularly for fast moving consumer goods, consideration of their impact and effectiveness amongst academics has been limited and identified as an important

 area for future research, as has greater use of supermarket panel data to provide insights into how different shoppers respond to price changes, including promotions (Grewal and Levy, 2009). .

Moreau et al (2001) argued that a lack of understanding by retailers about consumer perceptions of promotions can lead to weaknesses in their marketing strategies. Thus, as the use of promotions continues to grow, it is increasingly important to gain a more complete understanding of how consumers actually behave in response to different promotional activities. Previous studies have focused on the potential effects of promotions, including brand switching, category expansion and purchase acceleration (E.g. Chiang 1991, Chintagunta 1993; Foubert and Gijsbrechts 2007; Manning and Sprott 2007; Neslin and Shoemaker 1983; Putsis and Dhar 2001). However, the vast majority of these studies have relied upon small scale scanner datasets and small-scale experiments which many would argue cannot be considered representative of supermarket shoppers as a whole. This paper is unique in that it is the first to use loyalty card data from a large scale panel of over 17 million households to analyse promotions, and hence can be considered as more robust than previous academic research.

The use of promotions and their impact on retail performance Drawing upon the definitions put forward from previous researchers (e.g. Blattberg and Neslin 1990; Kotler 1988; Webster 1971) promotions are defined as marketing events limited in duration, implemented to directly influence the purchasing actions of customers with the underlying intention of achieving the objectives set out in the marketing strategy for the retailer and/or manufacturer. These objectives may include improving competitive position, brand extension, category expansion or increasing profitability. The use of sales promotions in the UK has increased significantly over the last decade, particularly in grocery retailing where competition between retailers has intensified. Nielsen Wire (2009) reported that in May 2009 a record 32 percent of all grocery sales in the UK were made up of products on promotion. In the US the figure is even higher, with a reported 42.8 percent of grocery

 sales made on promotion in September 2009, up from 40.8 percent in 2008 (Nielsen Wire 2009). Yeshin (2006) stated several reasons thought to have contributed to the increase in the use of promotions in the UK in recent years: The rising strength of retailers, with the top four supermarkets (Tesco, Asda, Sainsburys and Morrisons) now controlling almost 70 percent of grocery sales, resulting in fierce competition between own-label and branded products as well as for overall market share. A reduction in product differentiation A reductions in the level of brand loyalty

This has resulted in both UK supermarkets and their (branded) suppliers becoming increasingly dependent on promotional activity to drive sales growth. It is our contention that much (if not most) of this activity has occurred with limited analysis (and thus understanding) of the impact of promotional activity beyond the uplift in sales of the promoted products. Over the last few decades substantial inroads have been made in empirical research into the effects of retail promotions and what influences their effectiveness. For a promotion to be profitable for a retailer it must increase overall category sales, not just switch sales between brands. The type of mechanic used can influence how effective a promotion is. It is likely that there is not one type of promotion which is most effective for all products within a category, but rather that different promotions are more effective on different brands and products. Research by Wansink et al (1998) found that on average the sales volume increased by 125% with single-unit promotions, compared to 165% with multiple unit promotions. Research by Manning and Sprott (2007) studied specifically the effects of multiple unit price promotions on purchasing behaviour. The uplift in quantity purchased as a result of multiple unit price promotions was found to be dependent upon the magnitude of the quantity specified in the offer and the rate of product consumption for the specific product category.

 Both the length and the frequency of promotions can affect how responsive the consumer will be. The longer a promotion lasts, the less effective it will be, because over time the effect of the promotion upon sales will be reduced (e.g. Rao and Thomas 1973, Blattberg R. C. and Wisniewski 1987). The explanation for this may be that after a promotion has run for a certain length of time, consumers will come to expect that they can buy the product at the offer price and so will stockpile less and increase their inter-purchase time. Martnez-Ruiz et al (2006a) advised that promotions for non-perishable storable products should not exceed ten days, otherwise profitability will be reduced. The frequency of promotions will affect the consumers reference price (Kalwani and Yim 1992, Mayhew and Winer 1992) and hence can lower the height of the promotional spike in sales (Raju 1992). If a brand is discounted often, consumers will come to anticipate the promotion and will expect to always pay a lower price for the product. Factors related specifically to individual product categories such as brand share (Bemmaor and Mouchoux 1991), perishability and bulkiness (Bell et al 1999; Manning and Sprott 2007; Wansink and Deshpand 1994), and the occasion for which the product will be used (Meat and Livestock Commission 2002) can all influence the shoppers response to promotions. For example Bemmaor and Mouchoux (1991) found that as a result of price cuts, smaller brands experience a larger relative increase in sales, compared with larger more established brands. It was found that consumers will increase their usage for products which are perishable, as a result of promotions (Bell et al 1999). However, in categories of more staple items such as toilet paper and detergent, stockpiling takes place, increasing the inter-purchase time, meaning that the consumers move their purchasing forward, but do not increase their overall consumption (Wansink and Deshpand 1994). Most, if not all, product categories have a variety of brands and products suited to different uses or occasions and this can influence how the shopper responds to promotions within a category. Characteristics related specifically to the shopper and their household can also influence the effectiveness of promotions such as the household size, presence of children, age and income. The larger the family, the further the budget needs to stretch and the more likely they will be to buy into promotions (Bawa and Gosh 1999; Urbany et al 1996). There is conflict in the literature as to the

 effect age has on promotional response. Some believe younger shoppers are more likely to purchase products on promotion, either because they are more price sensitive than older shoppers (Ainslie and Rossi 1998) or because they are more likely to be influenced by stimuli such as displays and advertising and hence make decisions on impulse at the point of purchase than older shoppers (Inman and Winer 1998). However, it has also been suggested that older shoppers have more time to shop and therefore may be more likely to take advantage of promotions (Raju 1992). In reality it is quite likely that the response by different age groups will depend upon other factors such as the product being promoted and the mechanism used.

Hypotheses Our understanding of how promotional response varies within and between product categories and across shopper segments is crucial as it enables better targeting of promotions to ensure they yield the best results possible. Despite the amount of data now available to retailers through loyalty card schemes and point of sale transaction data, very little research has been undertaken to assess which promotions are most successful for which products and the impacts of promotions beyond the sales uplift of the promoted product. On the basis of our review of the existing promotional literature, four hypotheses were formulated for the purpose of this paper:

H1: Positive promotional impacts at the product level do not systematically result in category growth. H2: Promotional impacts are sensitive to the specific mechanic used H3: Promotional impacts are sensitive to specific product characteristics H4: Promotional impacts are sensitive to the demographic characteristics of specific shopper segments
These hypotheses will be tested using the methodology outlined below.

METHOD Multiple regression was used to estimate the impact of different promotional mechanics within the fresh meat category. Multiple regression enables us to evaluate the separate contributions of one or more variables acting jointly on a single dependant variable. Moreover, it is easily applicable to the analysis of promotions as a given product may use several types of promotion, each of which may have a different effect on sales of both the promoted product and substitute products. Regression analysis techniques have been widely used by other researchers to estimate the effects of promotions (see e.g. Bolton 1989; Mac and Neslin 2004; Sethuraman and Tellis 2002; Van Heerde et al 2004). Regression analysis is well suited to the analysis of large samples of data and for measuring the effects efficiently across different product sub-groups, identifying switching and substitution effects between products and differences across segments of shoppers. The method enables several independent variables to be modelled at the same time, taking into account interactions, and is a logical choice given the available dataset.

Data There are two main types of data used in the promotions literature to analyse the impact of promotions on purchasing behaviour: panel data and retailer scanner data. Panel data provides information at an individual household (or segment) level; for example by household size or by age. Popular sources of panel data include A. C. Nielsen and TNS Worldpanel. Examples of studies which have utilised panel data include Ailawadi et al (2007), Bell et al.(1999), Chintagunta (1993), Foubert and Gijsbrechts (2007) and Vilcassim and Jain (1991). All these studies suffer from the same limitation in that the panel data used is relatively small scale and therefore less representative of the whole population when trying to make inferences about the way people respond to promotions.

 Store-level scanner data pools all sales in a given store, or chain of stores over a period of time but does not contain information on specifically which type of household these sales relate to. Examples of studies which have incorporated store-level scanner data include Mac and Neslin (2004), Martnez-Ruiz et al. (2006b) and Raju (1992). Panel data is more detailed than store-level scanner data and can be more useful in many circumstances, for example to compare the effects of promotions on different categories of shoppers or on brand loyalty. However, panel data can be noisier than aggregate store-level data because in panel data sales will typically be generated from a smaller sample size of consumers, which is likely to result in more week to week variation. The lower noise in aggregate store sales data is a strong factor in favour of using it for many types of analysis. Store-level scanner data also has the advantage of containing a much larger sample than panel data, meaning there will be less variation in the sales data which should yield more robust results. For this study weekly supermarket purchase data was used from all Tesco supermarkets across the UK, collected via the Clubcard loyalty scheme, which covers approximately 80 percent of total sales. The dataset combines the benefits of both store-level scanner data and panel data, and is collected on a much larger scale than those datasets previously used for promotion research. At the time of data collection for this research approximately 40 percent of UK households, owned a Tesco Clubcard. Tesco is the largest grocery retailer in the UK, with a segmented retail strategy, serving the entire spectrum of shoppers from price sensitive to up-market, and through different retail formats such as on-line, convenience and supermarkets. Recent figures indicate Tescos market share to be at 30.9 percent of total grocery retailing in the UK (TNS, 2009, cited in Wall Street Journal [online], 2009). Thus, the sample of shoppers is considered to be as closely representative of UK supermarket shoppers as possible from a single dataset. The power of using retailer loyalty card data is the ability to generate purchasing data at a segmented level on a large scale. For the purposes of this research the database was used to create a cross-sectional panel data set, with sales data sorted by life-stage and TV advertising region. Within the dunnhumby database there are five life-stage segments and ten TV advertising regions, which

 made it possible to create a panel dataset based on fifty individual segments in total. The different life-stage segments are young families (all children under 10yrs), older families (at least one child over 10yrs), young adults (aged 20-39yrs), older adults (aged 40-59yrs) and pensioners. The different regions of the UK, used within the panel dataset, were Scotland, Wales and the West, the South West, the South East, the Midlands, East England, London, Yorkshire, the North East and the North West. To create the panel dataset, sales data for each life-stage segment for each region was collated together into a large dataset comprising fifty panels in total. The dataset comprised of weekly sales over an 86 week period from 29th May 2006 to 21st January 2008. In order to offset the affect caused by increases in distribution (due to the continued growth in Tesco stores) the sales value was divided by the number of stores selling the product each week. This created the dependant variable; sales value per store.

Example - Fresh Beef Category In the UK, the majority of fresh meat sold through supermarkets is private label with little or no branding, making it unique from most non-commodity product categories. In other grocery categories where branding is more prevalent, consumer attitudes towards private label promotions may differ from the fresh meat sector. Garretson et al (2002) argued that consumers may have different attitudes towards private label products (which are generally priced below national brands) and national branded products on promotion. While there is no branding, as such, within the fresh meat sector in the UK, products are positioned differently with a range of price tiers, from value products through to premium and organic. These different tiers of products can be considered in essence to be separate brands and the model enabled us to identify switching effects between these different brands of beef. As well as different tiers there are also different cuts such as roasting joints, ground beef, and steaks. For the data analysis individual beef stock keeping units (SKUs) were grouped into eleven subgroups categorised by cut, and by tier. The products were first sorted by cut: roasting joints, frying and grilling meats

 such as steaks and chops, and ground beef. These groups were then further disaggregated by tier: standard label (e.g. Tesco private label brand), premium (e.g. Tesco Finest brand), organic, value (e.g. Tesco Value) and healthy label (e.g. lean or Tesco Healthy Living range). Substitution is most likely between the different tiers (e.g. Organic, Standard) of the same cut (e.g. Roasting joints), rather than between different cuts, therefore the analysis was carried out within the different cuts of meat. Roasting joints were analysed together, as were fry/grilling beef and ground beef. This made it possible to observe switching effects of promotions between tiers of products, within each cut sub-category. For example, a promotion on standard roasting beef joints may influence sales of value roasting beef joints. Substitution between cuts is considered less likely since different cuts are used for different meal occasions.

The Model A multiple regression model estimated using the fixed effects method was used. Fixed effects estimation was chosen as this method is most appropriate for the analysis of cross-sectional panel data because it controls for heterogeneity across different types of shopper segments or panels. The following equation represents the model used for the regression analysis: SALESit = 0 + 1SPCit1 + 2MPCit2 + 3LPCit3+ 4MULTIit4+ eit In the model, SALESit represents the dependant variable sales value per store for a given product subgroup, i, in a given time period, t. The parameters of the model are 0, which represents a fixed unknown parameter, and a series of 0-1 dummy variables representing the different types of price promotion for product sub-group i in the time period t. Single unit price reductions were by far the most commonly used form of promotion in the fresh beef category, and these were split into three different levels of price reduction. Small price reductions of less than fifteen per cent off (SPC), medium price reductions of between fifteen and thirty per cent off (MPC), and large price reductions (LPC) of more than thirty per cent off the original price. The final independent variable represented

 all types of multi-buy promotions (MULTI). The error term, e, incorporates all the immeasurable factors which may also be influencing sales aside from promotions. The beta coefficients are reported in the results to show which promotions have the largest effects on which products, and amongst which consumers, segmented by life-stage. The coefficients represents the value by which sales will increase for the retailer during the promotion, which will help to inform the retailer about how much additional stock is required when planning promotions.

RESULTS AND DISCUSSION In total 24.4 percent of fresh beef sales occurred while promotions were taking place during the time period analysed. Table 1 shows the percentage of sales within the different beef sub-groups occurring while on promotion. . These percentages were calculated by summing the sales of each individual product within each sub-group during the weeks when a promotion was taking place on the given product, and then diving this by the total sales for each product.

Table 1: Percentage of Sales occurring on Promotion by Beef Sub-group over the 86 Week period from 29th May 2006 to 21st January 2008

Sub-group Total Standard Premium Organic Healthy Value

% Of Sales Occurring On Promotion Roasting Beef 39.96% 50.31% 27.88% 31.71% Ground Beef 39.0% 54.67% 9.88% 3.47% 19.91% 3.92% Fry/Grilling Beef 16.7% 17.61% 23.32% 12.44%

In total almost 40 percent of roasting beef sales occurred while products where on promotion, compared with 39 percent for ground beef and a much lower value of 16.7 percent for fry/grilling

 beef. Within the roasting and ground categories the highest percentage of sales on promotion occurred within Standard sub-group, yet for the fry/grilling beef category the highest percentage of sales on promotion occurred within the premium sub-group. We will now present the results of the model, testing each hypothesis in turn. The tables report the results from the regression analysis, with only those results which are significant at least at the 5 percent significance level being reported.

Hypothesis 1: Positive promotional impacts at the product level do not systematically result in category growth The first hypothesis states that positive promotional impacts at the product level do not systematically result in category growth. For promotions to benefit the retailer they must ultimately result in category growth, as profit is not generated through shoppers merely switching existing consumption between brands or products within a category. Table 2 reports the results of the regression analysis at the category level, showing the impact beef promotions had on the value of the beef category overall.

Table 2: Regression results at the total category level with respect to different Promotional Mechanisms
Coefficient (/week) Promotion Small Price Cuts Medium Price Cuts Large Price Cuts Multi-Buys R-Sq *p<0.05 229238.9* 0.1437 -187524.5* Fresh Beef Category

Promotions account for just 14% of the variance in sales of beef at the category level, which tells us that promotions are just one of several factors likely to be influencing sales. At the total category level the impact of promotions overall was found to be insignificant, however when divided

 into specific promotional mechanics, it can be seen that multi-buys added value to the fresh beef category while medium price cuts de-valued the fresh beef category. This result is important as it shows that not all promotions add value to the retailer, and indicates that it would be highly beneficial for retailers to make informed decisions when deciding promotional plans. However, this result does not necessarily mean multi-buys are the best promotion to use for all fresh beef products.

Hypothesis 2: Promotional impacts are sensitive to the specific mechanic used The second hypothesis states that the impact of promotions will depend upon the specific mechanic used. This has already been touched upon above, where it was seen that multi-buys had a positive effect on the fresh beef category overall, while medium price cuts de-value the category. However it is likely that the effectiveness of different promotional mechanics will vary when analysing the category at a more disaggregated level. Table 3 reports the coefficients from the regression analysis, showing the overall effect on sales at the cut level as a result of the different promotional mechanics.

Table 3: Regression Results for the Roasting Beef, Ground Beef and Fry/Grilling categories with respect to different Promotional Mechanisms
Coefficient (/week) Promotion Roasting Beef Small Price Cut (<15%) Medium Price Cut (15-30%) Large Price Cut (>30%) Multi-buy 120.44** 9.78** 270.65** 89.26** Ground Beef Fry/Grilling Beef

R-Sq

0.132

0.106

0.380


**p<0.01 *p<0.05

Promotions have the biggest influence on sales within the fry/grilling beef category, since the R-Squared value tells us 38 percent of the variance in sales is attributable to promotions. The Rsquared values are much lower within the ground and roasting Beef categories. Within the roasting category, large price cuts were found to have the greatest significant positive impact on sales value, indicating that sales value per store will increase by 120 during the promotion. Within the ground beef category multi-buys were the only promotion to have a significant impact overall, while medium price cuts were the only promotion to have a significant impact on fry/grilling beef sales. Small price cuts did not have a significant impact on sales across any of the beef cuts. These results support the findings of a qualitative study by the Meat and Livestock Commission (2002), which indicated that price discounts would be more effective than multi-buys at driving sales of key occasion meats such as roasting joints and fry/grilling meats. Multi-buys were found to be the only effective form of promotion within the ground beef category, which adds further backing to the research carried out by the MLC, which found that multi-buys would be most effective on core everyday proteins such as ground beef which shoppers buy weekly out of habit for everyday meals during the week. Here we have identified that promotional response is sensitive to the mechanic used, with different cuts responding better to different mechanics. The results here show that promotions are not only sensitive to the mechanic used but also to the cut of beef. Hypothesis three tests in further detail how promotional response varies by specific product tier such as value or premium.

Hypothesis 3: Promotional impacts are sensitive to specific product characteristics The third hypothesis states that the impact of promotions will vary depending upon characteristics specific to the product. The results, when testing hypothesis two, have already demonstrated that different promotional mechanics work better on different cuts of beef. Here we

 explore this in more detail, further disaggregating the data within the specific cuts to how promotional response varies between different tiers of product. Table 4 reports the regression results for the roasting beef sub-groups with respect to different price promotions within the roasting category

Table 4: Regression Results for the Roasting Beef sub-groups with respect to different Price Promotions within the Roasting category
Standard RoastingBeef 130.1661** 94.39361** 138.0552** Coeficient(/week) Premium OrganicRoasting ValueRoasting RoastingBeef Beef Beef 9.872005** 4.495259** 4.430451** 14.29107** 2.342127** 2.265968** 5.520764** 10.01471** 8.998592** 17.43943** 1.529242* 0.061 0.381 0.119

Promotion PremiumMediumPriceCut OrganicMediumPriceCut StandardMediumPriceCut StandardLargePriceCut StandardMultiBuy RSq **p<.01 *p<.05

0.174

Within the roasting sub-group, promotions had the greatest impact on sales of organic roasting beef, as promotions were found to account for 38 percent of the variance in sales of organic roasting beef. The r-squared values were much lower for the other sub-groups. There were found to be differences across the tiers in terms of which promotions had the greatest impact. Large price cuts on standard roasting beef had the greatest positive impact on sales of standard beef. Conversely medium price cuts on premium roasting beef were found to have a significant negative impact on sales of standard roasting beef, as well as value roasting beef. This suggests shoppers are trading up to premium tier products away from lower tier products such as standard and value when they are on promotion, which adds weight to theories of asymmetric brand switching (see e.g. Krishnamurthi and Raj 1991). However, there is evidence of consumers trading down from premium tier products as a result of large price cuts and multi-buy offers on standard

 roasting beef, which goes against the theory of asymmetrical brand switching, whereby shoppers will trade up a tier but not down. There are some results which are hard to interpret, for example any kind of promotion within the roasting beef fixture appears to have a positive impact on sales of organic roasting beef. One theory behind this may be that the promotions attracted customers to the fixture initially, but some decided to buy organic rather than the product on promotion perhaps because the product looked more appealing.

Table 5 reports the regression results for the ground beef sub-groups with respect to different price promotions within the ground category

Table 5: Regression Results for the Ground Beef sub-groups with respect to different Price Promotions within the Ground beef category
Coeficient (/week) Standard Premium Organic Ground Beef Ground Beef Ground Beef -335.089* 18.16696* -25.0329* 170.707* 18.83326* 16.26799* 28.81646** 378.365** 16.3145** -17.19021* R-Sq **p<0.01 *p<0.05 0.164 0.26 0.113 Healthy Ground Beef 401.6603** -105.194**

Promotion Premium - Medium Price Cut Healthy - Multi-buy Organic - Multi-buy Standard - Multi-buy

0.544

Within the ground beef sub-group promotions have the biggest impact in the healthy subgroup, where promotions account for 54 percent of the variance in sales. Multi-buy promotions on healthy ground beef have a significant positive effect on sales of healthy ground beef, but consumers will switch to organic when on promotion. Within the standard subgroup, multi-buy offers on have a large positive impact on sales, but consumers will switch from standard to premium when premium

 ground beef is on offer. Sales of organic ground beef are negatively affected by promotions on standard and premium products. Harder to explain is the positive impact on premium ground beef of multi-buy promotions on healthy and standard ground beef. However it may be that consumers, who only want one unit of the product, switch to premium because they do not want to partake in the multi-buy promotions on healthy or standard ground beef that they otherwise would buy. This would therefore suggest that multi-buy offers have a negative impact on some shoppers, who will actively switch to a substitute product as a result. Similar affects are observed elsewhere in the sub-group, for example standard multi-buy promotions apparently increase sales of premium ground beef.

Table 6 shows the regression results for the fry/grilling beef sub-groups with respect to different Price Promotions within the fry/grilling category

Table 6: Regression Results for the Fry/Grilling Beef sub-groups with respect to different Price Promotions within the Fry/Grilling category
Coeficient(/week) Standard Premium Organic Fry/GrillingBeef Fry/GrilingBeef Fry/GrillingBeef 165.1634** 12.06273** 78.9555** 24.25732** 140.2161** 26.59584** 6.521285* 91.78317** 11.73351** 54.66264** 82.41607** 27.19562** 36.62011** 84.8706** 14.97713** 141.8864** 6.438946** 17.52011** RSq **p<.01 *p<.05 0.567 0.465 0.341

Promotion PremiumSmallPriceCut PremiumMediumPriceCut PremiumLargePriceCut OrganicSmallPriceCut OrganicMediumPriceCut StandardSmallPriceCut StandardMediumPriceCut

Of all the beef sub-groups, promotions had the biggest impact on sales for fry/grilling beef. The R-squared values are generally much higher for fry/grilling sub-groups than within the ground and roasting beef categories. Promotions accounted for about 57 percent of the variance in sales of

 standard fry/grilling beef. Interestingly the results imply that it was promotions within other fry/grilling sub-groups which had as much of an impact on sales of standard products, as promotions within the standard sub-group itself. The promotions having the greatest positive impact on sales of standard fry/grilling products were small and large price cuts on premium beef, and medium price cuts on standard beef. Small price cuts on organic beef were the only promotion to negatively impact upon sales of standard beef. The reason so many promotions within other tiers of fry/grilling beef had such a positive impact on sales of standard beef may be due to the heavy presence of promotions within the fry/grilling category overall which led to several promotional clashes across tiers. Sales of premium fry/grilling beef were negatively affected by promotions on organic fry/grilling beef, and vice versa sales of organic beef were negatively affected by premium promotions. This indicates organic and premium products are close substitutes and shoppers will swap one for the other when a promotion occurs. The results provide evidence to support hypothesis three. There were found to be differences in promotional response across cuts and price tiers of product. The results also support hypothesis two, in that there were differences in response depending upon the promotional mechanic used.

Hypothesis 4: Promotional impacts are sensitive to the demographic characteristics of specific shopper segments One of the benefits of panel data over and above scanner data is the ability to observe how behaviour varies between consumer segments. Loyalty card data particularly leans itself towards such analysis as retailers can create detailed segmentations of their shoppers. The literature review revealed that it was thought characteristics such as age and household size can impact upon promotional response. Therefore the dataset was also used to determine how different shoppers responded to the promotions based upon their lifestage. Tables 7A-C report the results of the regression analysis by

 life-stage segment for the standard roasting beef, standard ground beef and standard fry/grilling beef sub-groups with respect to the various promotions. The

Table 7A: Regression Results by Consumer Life-stage for the Standard Roasting Beef sub-group
Coeficient(/week) Promotion PremiumMediumPriceCut OrganicMediumPriceCut StandardLargePriceCut RSq **p<.01 *p<.05 OlderAdults 147.9276** 82.17182** 191.0175** 0.1582 YoungAdults 119.2251** 90.74004** 116.3995** 0.1637 OlderFamilies 174.8232** 140.5603** 189.5294** 0.1969 YoungFamilies 161.3345** 134.0492** 120.0747** 0.1739 Pensioners 47.51983** 24.44671** 73.2548** 0.1089

Of the five life-stage segments, sales value per store for standard roasting beef in response to a large price cut increases the most for older adults and older families, and the least for pensioners. Sales of standard roasting beef fell in response to medium price cuts on premium roasting beef across all segments. Sales value increased the most in response to the price cuts on premium roasting beef for Families, both older and younger.

Table 7B: Regression Results by Consumer Life-stage for the Standard Ground Beef sub-group
Promotion PremiumMediumPriceCut HealthyMultibuy StandardMultibuy RSq **p<.01 *p<.05 OlderAdults 221.1436** 248.365** 246.5703** 0.1419 YoungAdults Coeficient(/week) OlderFamilies YoungFamilies 439.846** 670.0788** 556.3392** 0.1263 666.7191** 0.1332 0.1021 Pensioners 95.0449** 122.1197**

363.3129** 0.1282

Standard multi-buy promotions had a large and positive effect on sales of standard ground beef within all segments except Pensioners. The segment for which sales value increased the most in response to the standard ground beef multi-buy promotion was young families. Similarly the young families were the segment for which sales of standard ground beef fell the furthest in response to the price promotions on premium ground beef. Sales value of standard ground beef amongst both older adults

 and pensioners actually increased in response to multi-buy promotions on healthy ground beef. This indicates that shoppers within these segments were put off by the multi-buy offer on healthy ground beef, which perhaps they normally purchase, and instead switched to buying standard ground beef. Older adults and pensioners perhaps do not want to purchase multiple packs of ground beef since they do not have families to feed, so switch their spend to a product not on offer.

Table 7C: Regression Results by Consumer Life-stage for the Standard Fry/Grilling Beef sub-group
Promotion PremiumSmallPriceCut PremiumMediumPriceCut PremiumLargePriceCut OrganicSmallPriceCut OrganicMediumPriceCut StandardSmallPriceCut StandardMediumPriceCut RSq **p<.01 *p<.05 OlderAdults 150.0513** 73.65662** 138.6866** 72.80364** 54.00508** 78.66549** 114.4792** 0.3162 YoungAdults 194.8505** 67.02216** 122.4337** 100.5544** 106.0432** 78.75664** 115.811** 0.3071 Coeficient(/week) OlderFamilies YoungFamilies 212.9796** 231.7815** 113.3383** 109.122** 207.6007** 174.8731** 135.1511** 137.3172** 124.3196** 124.1518** 144.0243** 108.3398** 222.2258** 223.0028** 0.3365 0.3031 Pensioners 36.15398** 31.63837* 57.48661**

33.91312* 0.2923

Looking at the promotions specifically on standard fry/grilling beef, sales value increased the most amongst Older families in response to small price cuts, while spend amongst young families increased the most in response to medium price cuts. Spend by families also increased by the most in response to small price cuts on organic beef; with sales of standard fry/grilling beef declining the most for both young and older families. Promotions in the fry/grilling category had the smallest effect on sales amongst pensioners.

CONCLUSIONS The main purpose of this paper was to demonstrate that supermarket loyalty card data can be used to generate unique empirical insights into the effectiveness of promotions. Retailers and manufacturers need to move away from the notion that one promotion fits all and instead focus their efforts on targeting promotions effectively and using those mechanisms which will yield the best

 results for the particular product or category in question. The use of loyalty card schemes amongst retailers worldwide is increasing, which means more detailed analysis of promotional impacts, of the kind reported here is becoming increasingly possible. One of the main points arising from the results is the considerable variability in the impact of different promotion mechanics between and within the different sub-groups. This illustrates the point that one promotion does not fit all and promotional strategies should take greater notice of the effectiveness at the individual product level, to avoid devaluing the product category. In the example used for this paper, there were found to be differences in which promotions work best within each cut and tier within the beef category, which could not be captured by analysis of highly aggregated data. Indeed the r-squared values in the examples used in this paper were generally higher for the analysis carried out at the disaggregated level, than at the total category level. Much of the promotional research, not just in the meat category, has used highly aggregated data on product categories to measure promotional impacts where in fact very different results may have been observed if the data used had been at a more disaggregated level. The research has also demonstrated how promotional response varies across life-stage segments. Generally spend amongst families increased the most in response to promotions, and for pensioners the least. This result is perhaps unsurprising, but some interesting points have arisen. For example pensioners and older adults appear to be put off multi-buy promotions within the ground beef category, to the extent that they switch their purchases to standard ground beef in response to multibuy offers on healthy ground beef. Further research across other grocery categories using panel data is necessary for retailers to have a broad understanding as to how different segments respond to enable them to plan and target promotions more effectively. There is a huge gap in the literature at present for such research, particularly using large scale panel datasets. The findings in this paper have also contributed further to the literature on brand switching. Existing literature is unanimous in the belief that if a lower tier brand is promoted it does not attract customers from high-tier brands, but the promotion of higher quality, premium priced brands impacts

 significantly upon weaker brands (see e.g. Kumar and Leone 1988; Krishnamurthi and Raj 1991; Mulhern and Leone 1991; and Martnez-Ruiz et al. 2006a). While the results from the ground beef sub-group backed up this theory with sales of standard ground beef falling considerably in response to promotions on premium ground beef, some other results were conflicting. For example, while medium price cuts on premium roasting beef were found to have a significant negative impact on sales of standard and value beef, there was evidence of consumers trading down from premium tier products as a result of large price cuts and multi-buy offers on standard roasting beef. These findings suggest that while the theory of asymmetric can be applied in some circumstances, it does not apply for all products categories. This paper significantly contributes to previous literature in the subject of promotions through analysing the impacts using a data set unique in both its size and scope, and which has not previously been used for such a purpose. Loyalty card panel data is now more accessible than ever to many of the major retailers, but so far it is not being used to the best of its potential to inform decision making as to which promotions will be most effective. As can be seen from the results, some promotions have a greater impact than others, but retailers need to understand these differences in order to implement promotions effectively. There is much scope to expand upon this research through adding additional variables to the model, such as seasonality and merchandising, as well as using other product categories.

LIMITATIONS AND AREAS FOR FURTHER RESEARCH There were several limitations to this study which can be built upon and developed with future research. The r-squared values were relatively low, which tells us that there are many other factors influencing sales at the same time as promotions. For example point of sale displays, advertising and merchandising taking place at the time of promotion. If some products on promotion are being positioned on aisle ends or being heavily advertised, then this is likely to increase the impact of the promotions since more people will be aware of the promotion. Research by Sethuraman and

 Tellis (2002) found there to be a positive relationship between the advertising and the response to promotions, but other research has contradicted this (Nijs et al 2001). Ailawadi et al (2009) stress the need for further research in this area to enable retailers to effectively account for manufacturer advertising in their promotion budgeting decisions. Since this information was unavailable, it was not possible to consider how the positioning or advertising of promoted products influenced the impact on sales in this paper; however this is a variable which can be included in the model in the future and could yield very insightful results using such a large scale panel dataset. A further limitation to the research carried out is that it does not consider any additional benefits that promotions within the red meat category brought to the retailer. Even if a promotion does not increase the value of the category or a particular promoted sub-group, it is still possible that the retailer is benefiting from increased footfall as a result the promotions and shoppers are spending money elsewhere in the store. However, these possible indirect effects of promotions to the retailer do not benefit the primary producers who inevitably lose out, particularly if promotions are de-valuing products within the overall category. The next step following this research is to further develop the model through adding additional variables to attempt to capture more of the variables influencing sales in addition to promotions, such as seasonality and in-store merchandising. It is hoped this will strengthen the model and further demonstrate the value in such analysis by retailers worldwide using their own loyalty card data to create more sustainable and profitable promotional strategies. The model used here is just one of several which could be used to analyse the impact of promotions with supermarket loyalty card data. Further research will enable us to test other models and compare the results to ultimately identify which model is most appropriate to analyse the promotional impacts.

 REFERENCES Ailawadi, Kusum L., Karen Gedenk, Christian Lutzky and Scott A. Neslin (2007). Decomposition of the Sales Impact of Promotion-Induced Stockpiling. Journal of Marketing Research, 44(August), pp450-467 Ailawadi, Kusum L., J. P. Beauchamp, Naveen Donthu., Dinesh K. Gauri and Vankatesh Shankar (2009). Communication and Promotion Decisions in Retailing: A Review and Directions for Future Research. Journal of Retailing, 85(1), pp42-55 Ainslie, Andrew and Peter E. Rossi (1998). Similarities in Choice Behaviour across Multiple Categories. Marketing Science, 17(2), pp91-106 Kapil, Bawa and Avijit Ghosh (1999). A Model of Household Grocery Shopping Behaviour. Marketing Letters, 10(2), pp149-160 Bell, David R., Jeongwen Chiang and V. Padmanabhan (1999). The Decomposition Of Promotional Response: An Empirical Generalization. Marketing Science, 18(4), 504 Bemmaor, Albert C. and Dominique Mouchoux. (1991). Measuring the Short-Term Effect of Instore Promotion and Retail Advertising on Brand Sales: A Factorial Experiment. Journal of Marketing Research, 28(May), p202-214 Blattberg, Robert C. and Scott. A. Neslin (1990). Sales Promotion Concepts, Methods, and Strategies. NJ: Prentice-Hall. Blattberg, Robert C. and Kenneth. J. Wisniewski (1987). How Retail Price Promotions Work: Empirical Results. Working Paper (43), University of Chicago, Chicago IL Bolton, Ruth N. (1989). The Relationship Between Market Characteristics and Promotional Price Elasticities. Marketing Science, 8(2), pp153-189 Chiang, Jeongwen (1991). A Simultaneous Approach to the Whether, What and How Much to Buy Questions. Marketing Science, 10(4), pp297-315

 Chintagunta, Pradeep K. (1993). Investigating Purchase Incidence, Brand Choice, and Purchase Quantity Decisions of Households. Marketing Science, 12(2), 184-208. Foubert, Bram and Els Gijsbrechts (2007). Shopper Response to Bundle Promotions for Packaged Goods. Journal of Marketing Research, 44(November), pp647-662 Garretson, Judith A., Dan Fisher and Scott Burton (2002). Antecedents of Private Label Attitude and National Brand Promotion Attitude: Similarities and Differences. Journal of Retailing, 78(2002), pp91-99 Grewal, Dhruv and Michael Levy (2009). Emerging Issues in Retailing Research. Journal of Retailing, 85(4), pp522-526 Inman, Jeffrey J. and Robert S. Winer (1998). Where Rubber Meets the Road: A Model of In-Store Consumer Decision Making. Working Paper, Marketing Science Institute (October), Report No. 98-122. Kotler, Phillip (1988). Marketing Management: Analysis, Planning, Implementation and Control. 6th Ed, Englewood Cliffs, N.J. Prentice-Hall International Krishnamurthi, Lakshman and S. P. Raj (1991). An Empirical Analysis of the Relationship between Brand Loyalty and Consumer Price Elasticity. Marketing Science, 10(2), p172-183 Kumar, V and Robert P. Leone (1988). Measuring the Effect of Retail Store Promotions on Brand and Store Substitution. Journal of Marketing Research, 25(May), pp178-185 Mace, Sandrine and Scott A. Neslin (2004). The Determinants of Pre- and Post Promotion Dips in Sales of Frequently Purchased Goods. Journal of Marketing Research, 41(Aug), pp339-350 Manning, Kenneth C. and David E. Sprott (2007). Multiple Unit Price Promotions and their Effects on Quantity Purchase Intentions. Journal of Retailing, 83(4), pp411-421 Martnez-Ruiz, Maria P., A. Moll-Descals, Miguel A. Gmez-Borja, and J. L. Rojo-lvarez (2006a). Assessing the Impact of Temporary Retail Price Discounts Intervals Using SVM Semi-

 parametric Regression. International Review of Retail, Distribution and Consumer Research, 16(2), pp181-197 Martnez-Ruiz, Maria P., A. Moll-Descals, Miguel A. Gmez-Borja, and J. L. Rojo-lvarez (2006b). Using Daily Store-Level Data to Understand Price Promotion Effects in a Semi-parametric Regression Model. Journal of Retailing and Consumer Services, 13, pp193-204 Mayhew, Glenn E. and Russell Winer (1992). An Empirical Analysis of Internal and External Reference Prices Using Scanner Data. Journal of Marketing Research, 19(June), p62-70 Meat and Livestock Commission (2002). The Shopping Decision Tree: Understanding the Consumer. Moreau, Page., Aradhna Krishna and Bari Harlam (2001). The Manufacturer-Retailer-Consumer Triad: Differing Perceptions Regarding Price Promotions. Journal of Retailing, 77(2001), pp547-569 Mulhern, Francis J. and Robert P. Leone (1991). Implicit Price Bundling of Retail Products: A Multi-product Approach to Maximizing Store Profitability. Journal of Marketing, 55(Oct), p63-79 Neslin, Scott A. and Robert W. Shoemaker (1983). A Model For Evaluating the Profitability of Coupon Promotions. Marketing Science, 2(Fall), pp361-88 Nielsen Wire, (2009): Promotions Spur Growth in UK Grocery Sales. [Online] [Accessed 21st July 2010] Available from http://blog.nielsen.com/nielsenwire/consumer/promotions-spur-growthin-uk-grocery-sales/ Nijs, Vincent., Marnik G. Dekimpe, Steenkamp, Jan-Benedict E.M., and Dominique M. Hanssens (2001). The Category Demand Effects of Price Promotions. Marketing Science, 20(1), pp122 Putsis Jr, William P. and Ravi Dhar (2001). An Empirical Analysis of the Determinants of Category Expenditure. Journal of Business Research, 52, pp277-291

 Raju, Jagmohen S. (1992). The Effect of Price Promotions of Variability in Product Category Sales. Marketing Science, 11(3), p207-220 Sethuraman, Raj. and Gerard Tellis (2002). Does Manufacturer Advertising Suppress or Stimulate Retail Price Promotions? Analytical Model and Empirical Analysis. Journal of Retailing, 78(2002), pp253-263 Urbany, Joel E., Peter R. Dickson, and Rosemary Kalapurakal (1996). Price Search in the Retail Grocery Market. Journal of Marketing, 60(2), pp91-104 Van Heerde, Harold J., Peter S. H. Leeflang and Dick R. Wittink (2004). Decomposing the Sales Promotion Bump with Store Data. Marketing Science, 23(3). Vilcassim, Naufel J. and Dipak Jain (1991). Modeling Purchase Timing and Brand Switching Behaviour Incorporating Explanatory Variables and Unobserved Heterogeneity. Journal of Marketing Research, 28(Feb), p29-41 Wall Street Journal (2009). Tesco Grocery Mkt Share Falls To 30.9% From 31.1% On Yr TNS. [Online] [Accessed 23rd September 2009]. Available from: http://online.wsj.com/article/BTCO-20090915-708310.html Wansink, Brian. and Rohit Deshpand (1994). Out of Sight, Out of Mind: Pantry Stockpiling and Brand-Usage Frequency. Marketing Letters, 5(1), pp91-100. Springer Netherlands Wansink, Brian., Robert J. Kent, and Stephen J. Hoch (1998). An Anchoring and Adjustment Model of Purchase Quantity Decisions. Journal of Marketing Research, 35(February), pp71-81 Webster, Frederick E. (1971). Marketing Communication: Modern Promotional Strategy. John Wiley & Sons Publishers Yeshin, T. (2006). Sales Promotion. P71, Thompson