Retail Market Measurement0001

Part

241

2: PROCESSES

Chapter 8

Retail marl{et measurement James Brooks Tim Bowles INTRODUCTION The purpose of this chapter is to explain how market information based upon retail sales is derived and used by manufacturers and retailers to make better business decisions. Since the emergence of mass marketing, manufacturers of fast turnover packaged goods have wanted information about the performance of their brands, both against key competitors, and the market as a whole. For the fifty years prior to 1980, such information came from two main sources: consumer panels and retail audits. In the retail audit, information on sales through a sample of shops was collected regularly by auditors working for the research company. Consumer panels approached the task by recruiting a sample of consumers who kept a record of all their packaged goods purchases, usually in the form of a purchase diary. In both cases, the information collected, from shops or consumers, could be analysed and statistical methods used to project from the sample to what had been sold through all shops, or bought through all households. Estimates of total market size, and the share of major brands, could then be provided for product categories ranging from detergents to savoury snacks. The principal advantage of retail audit measurement was accuracy. In a sample of affordable size many more purchase observations could be obtained from a sample of shops than from a sample of consumers, in the same data collection period. This advantage was particularly important for tracking infrequently purchased products or brands. The retail audit also avoided potential error arising from human memory and accuracy in recording purchases, which can affect consumer panel results.

242

Tim Bowles, James Brooks

Consumer panels, while limited in the ability to provide accurate market measurement, are a rich source of information on the consumer; who buys as well as what is bought. Through consumer panel data one can gain insight into the household profile for purchases of different products, and how brand choices are influenced by the characteristics of buyers themselves. The two main sources of market tracking information are therefore complementary. The process of collecting information on retail sales has undergone a revolution in the last twenty years. This revolution was driven by the progressive adoption by retailers of electronic cash registers equipped with scanners which can recognise bar-codes printed on product packaging. EPOS (Electronic Point of Sale) equipment is now widespread in the United States, Western Europe, Japan and other developed economies, where it has largely replaced the traditional audit as a source of information on retail sales of packaged goods. This availability of scanner information is a by-product of the retailers' drive for efficiency through the use of EPOS equipment, as the basis for stock control and other administrative systems. The impact of EPOS developments is enormous, in that information is available at an unprecedented level of detail for products and brands, for every shop in a retail chain, and for any time period, however short. Whereas the challenge for the retail audit was to collect reliable data, the challenge in the scanning era is to manage the data and reduce them to usable reports for management. While scanner penetration of retail outlets in developed countries is now widespread, goods in many countries are still mainly sold through traditional, privately-owned shops. Even in some developed markets, the traditional retail trade is still significant. International manufacturers and retailers of consumer goods can therefore expect to have to deal with both traditional retail audit and scanner-based information for the foreseeable future, in order to gain an understanding of their markets. In the remainder

of this chapter we will explain the basis of retail market measurement as it is practised today, and how the data are analysed to support business decisions.

A BRIEF HISTORY OF RETAIL DATA SUPPLY The science ofretail sales tracking was invented by Arthur C. Nielsen in 1933, when he set up his Drug Index service in the United States to measure drugstore sales. This development was followed in 1934 by the creation of the Food Index to measure sales in the grocery trade sector. The first European developments occurred in 1937, when the British Market Research Bureau set up a panel of 1,000 grocers in major towns in the United Kingdom. Soon afterwards Nielsen's company took the first step in an aggressive programme of European expansion when it set up the UK Food Index in 1939.

Retail market measurement

Over the following fifty years ACNielsen followed the global expansion of its multi-national clients to establish a powerful network of companies providing retail audit services in more than 120 countries. Competition to Nielsen was mainly local and the company became the dominant worldwide supplier except in certain niches such as specialist markets. (The company also became active in media audience research with the supply of radio and TV ratings). EPOS scanning equipment, and the associated adoption of bar-codes on products, was developed during the 1970s. The Article Numbering Association (ANA) was formed in 1977 as an initiative by European manufacturers and retailers. This organisation has now expanded across the world and currently comprises seventy-nine national product numbering associations representing eighty-six countries. This system of bar-coding, described in a subsequent section, is now employed by 600,000 consumer goods companies. The appearance of EPOS data in the late 1970s created the conditions for the emergence of a serious competitor to Nielsen in the United States: Information Resources Inc. (IRI). IRI recognised the potential for scanner data and, in 1979, created a revolutionary test market service, based upon scanner data, called BehaviorScan. IRI then exploited its early capability in the analysis of scanner data to launch a national retail tracking service in the United States, competing directly with Nielsen, called InfoScan, in 1986. At the time of writing ACNielsen and IRI are the dominant suppliers of retail tracking data in the United States, and IRI is challenging Nielsen's past dominance in Europe and other parts of the world. BASIC PRINCIPLES The underlying principle of retail market measurement is that sales data are taken from a number of retailers and combined to represent a larger trade sector: individual trade sectors can then be combined in order to represent the total market. This has meant that retailers can provide their sales data to be pooled into a tracking service whilst maintaining the confidentiality of each participating retailer's sales. Figure 1

data provided

data reported

243

244

Tim Bowles, James Brooks

For example, the top five grocery retailers in a country might be combined to form a trade sector called key account retailers. The data released to the client would be the 'Top five retailers' trade sector, which preserves data confidentiality for the individual retailers participating in the service. In developed markets, for the main trade sectors, this information is obtained

directly from retailers in the form of tapes containing EPOS records of sales through all, or a sample of their outlets. Tapes typically contain data for sales week-by-week although, in theory, data could be collected day-by-day or hour-by-hour. Where EPOS data are not available, sales information has to be derived through a retail audit. This process involves going into stores to count the stock within a store both on the shelf and in the stock room. Comparing this information with the stock holding on the last visit, together with deliveries in the time period, means that sales can be derived: sales

=

opening stoe k

.. + dehvenes

-

closing stoe k

Because a retail audit involves shop visits by trained auditors it is quite an expensive procedure. For this reason most retail audits have collected and reported data on a bi-monthly basis. In comparison with more frequent scanner data, retail audit data are therefore more limited in their ability to track the effects of marketing actions such as advertising and promotions.

The issue of retail data confidentiality Retailers have, in the past, been reluctant to provide details of their own individual sales for distribution to manufacturers or competitive retailers, concerned that they could thereby sacrifice competitive advantage. This is the reason they have insisted that research companies should only report pooled data, at total market and trade sector level. Manufacturers have therefore had to rely upon consumer panels to provide estimates of the shares of category sales held by different retailers. This is vital information when manufacturers are negotiating with retailers to increase their share of scarce shelf space. The agencies which analyse and supply retail data have long argued for more open data exchange, so that they can provide 'named account' or 'key account' data which show the individual sales performance of different retail chains. By the end of 1997 most retailers in the United States and in The Netherlands and a minority of retailers in other European countries had agreed to the release of their own, identified sales data in the marketplace. This trend is likely to continue to a point in the future where there will be total transparency of data exchange between retailers and manufacturers, but the process may take some years.

245

Retail market measurement

THE BAR-CODE The allocation of bar-codes is supervised by the Article Numbering Association (ANA). The international standards for the EAN (European Article Number) were developed as tools for improving business efficiency through: o o

providing a system for identification of products, services, etc. standard bar-codes to represent information which can easily be read by scanners.

It is this product identifier which is the focal point for scanning-based market measurement studies.

retail

The bar-code typically consists of thirteen digits, which are constructed in the following manner: 50 12345 12345 7 The first two digits signify the country of origin. The next five digits are a manufacturer's identifier which is allocated by the ANA. The next five digits are the product code, which is assigned by the manufacturer. The final digit is a check digit, which is calculated by an algorithm applied to the previous numbers. This check digit ensures the correct reading of the bar-code at the point of sale. This is achieved by the scanning device reading the bar-code and re-calculating the check digit, which is then compared with the check digit in the bar-code; if they do not match then the item will be rejected by the scanner. The process ensures full data iptegrity. Occasionally small products do not have the space for a thirteen digit bar code. In these instances an eight digit bar-code may be used. Due to the limited number of these codes, they are issued directly by the ANA. Although bar-codes are numbers, they are represented by a series of black lines on a white background which can easily be read by scanning technology, as in Figure 2. Figure 2

0"0'51111"12881""

7

Tim Bowles, James Brooks

246

This convention for the allocation of bar-codes ensures that: o o

bar-codes are unique the bar-codes are non-significant, i.e. it is the EAN which is the key to access the database

o o

they are consistent across countries, i.e. international they are read securely.

To give an indication of the scale of the data processing exercise, in the three years to 1997, IRI (in the United Kingdom alone) has processed in excess of one million different bar-codes. Bar-codes are allocated for each variant of a product. For example, a lemon flavoured product will have a different bar-code from the same product which is orange flavoured. Even special offer packs (e.g. 10% extra free) will have an individual bar-code. This allocation system enables maximum product detail to be available to data users, who can select the level of detail they require in analysis. Figure 3 Audit

Scanning

Accuracy

Audits rely on manual counting of stock and stores recording deliveries. As such they are subject to human error. They also measure store throughput. Shrinkage (theft, damage etc.) is recorded as a sale.

Scanning data are a true representation of what has actually passed through a retailer checkout and therefore are the most accurate measurement of sales.

Cost

As audits rely on personnel visiting stores to count stock, costs are relatively high.

Sales data can be supplied directly by the retailer thereby reducing the need for store visits and therefore the costs.

Timing

Due to the cost being linked to the frequency of store visits, data are typically available montWy or bimontWy.

Data are usually collected weekly. Daily data are now also becoming available.

Back data

Due to the costs, data are usually collected when a client commissions

a

Data are stored on tape, therefore back data are more easily accessible.

study, meaning that there are no back data available for previously unreported categories.

Client reports

Syndicated

Delivery time

Typically four weeks after period end.

Ten to twenty days (may be shorter in some cases).

Outlet coverage

Cost constraints mean that sample data are collected and projected to represent a total retailer's sales.

Census data (data for all stores) are available.

Product detail

Data may be collected at 'audit code level' to reduce the number of

Data are collected at EAN level.

trend reports.

individual items being counted, e.g. different scents of shampoo may be combined.

Customised reports driven by software applications.

-

247

Retail market measurement

AUDIT AND SCANNING DATA The primary purpose of retail sales tracking is to provide a reliable measurement of sales, covering all significant types of retail outlets. Scanning information is rarely available from smaller outlets, such as small shops, kiosks and market traders. Sales through such outlets may nevertheless be significant for certain product categories. Retail audits are often the only practical way of collecting data from such outlets, which do not have scanning equipment. A comparison of audit and scanning data, on key criteria from the user's point of view, is made in Figure 3. The availability of scanning data varies considerably by country. The chart below demonstrates the percentage of food sales which pass through scanning checkouts, varying from the high 90s to as low as 10%.

Figure 4 PERCENTAGE SCANNING PENETRATION 100 90 80 70 60 50 40 30 20

10 o

i ·~~ j j f

.[~~~.~~

J ~

J

~

j

<

(/)

~

...•

~

~

Source: ACNielsen estimates 1996

It was expected that the introduction of scanning would reduce the costs of the provision of retail market measurement services. This has not been the case for a number of reasons which are identified below: o

retailers' requirement for payment to access their data either in cash or information services (and in some cases both)

o

the requirement to continue store visits to collect information on in-store merchandising conditions (i.e. factors which influence sales)

248

Tim Bowles, James Brooks

o o o o

the requirement to continue audits in trade sectors and/or countries with low scanning penetration the scanning has vastly increased the size of databases as a greater level of product detail has become available the requirement for continual investment in information technology to support the advanced analysis of scanning data the maintenance of a product dictionary for an ever-changing universe of products and product variants.

With the development of consistent bar-codes and more portable technology, it is becoming more common to replace paper-based in-store audits with handheld scanners. With this technology, the fieldworker scans the product and then enters the volume of stock, together with the selling price. Historically audits have been conducted at an item level (e.g. flavours may have been merged to reduce the number of items for which data were being collected) but audit data can now be collected and reported at the same level of product detail as scanning data. Using the bar-code also ensures that audit and scanning data can be aligned perfectly in an integrated database. STRUCTURING THE DATABASE Retail market measurement is unlike many other forms of market research. The end product is not a research report but a database which is delivered to clients usually in electronic form. The database will typically be updated with new data each reporting period (weekly, four weekly, monthly etc.). This database is then used by clients on a continuous basis to monitor and investigate both their own and competitors' performances. Both Nielsen and IRI, the principal data suppliers, also offer proprietary software which their clients can use to select and manipulate data from the database. The databases are constructed with two primary purposes in mind: o

to monitor the trend of sales for different products, over time and in various trade sectors

o

to investigate how various other measures of causal conditions in the store influence sales; such measures can include distribution levels for the product, price level and the presence or absence of certain types of promotion.

It is sometimes helpful to think of the database as a cube in which each cell is a combination of specific: o

trade sectors, i.e. the combination of retailers

o o o

time periods measures, e.g. unit sales, volume sales, rate of sale products, e.g. which products are being reported.

249

Retail market measurement

Figure 5 Product

Trade Sector

Time

Structuring a database in this fashion means that it is possible to cut the data in different ways to meet the analysis needs of the end user. Some examples of how the data cube may be 'cut' are shown below: Figure 6

Financial Managers

Product Managers

Regional Managers

Ad hoc

Given that the dimensions for structuring the database are a critical part of its utility, the next sections will concentrate on how these dimensions are ,constructed, and how sales and distribution measures are defined.

Trade sectors In defining the trade sectors to be monitored

within a retail tracking study,

there are three key considerations: o o o

client requirements for trade coverage and segmentation of the trade structure of the retail trade in the country willingness of the retailers to contribute their data within different levels of trade sector aggregation.

There are likely to be several stages in defining the trade structure: 1. 2. 3. 4.

Client requirements Defining the universe Selecting the sample Projecting the stores to represent the universe.

250

Tim Bowles, James Brooks

Client requirements The requirements of clients in terms of trade coverage can vary significantly. For manufacturing clients, coverage of the outlets which distribute their products is likely to be the critical factor. As a result of this, clients who manufacture goods which are distributed through a wide network are likely to require coverage which represents the total market. On the other hand, clients who manufacture products for distribution through a network of specialist retailers are more likely to be satisfied with coverage of that specific trade sector. The relative importance of different trade sectors varies considerably by product field and by country. Table 1 shows the relative importance of different trade channels for health and beauty products across a number of different countries (i.e. the percentage of sales each trade sector represents). In practice it is difficult even to develop a cornmon classification of retail outlets, some of which are specific to a few countries. I 2 70 2 22 44 5 7 40 59 25 20 35 302 Greece 30 300 9 3 60 20 10 10 10 31 Poland 25 Republics Kingdom Turkey 7 3 kiosks and Pharmacies Perfumeries

Table Slovak 1 Source: IRI Czech and

In the above example the requirements for a health and beauty manufacturer would be different across different countries, and this needs to be reflected in the definition of the trade sectors used in the study. Achieving the right segmentation within the trade structure is particularly important as in many countries retailers have increased or are starting to increase their presence in 'non core areas'. A good example of this is the health and beauty market in the United Kingdom. Over the past five years, the traditional grocery chains have invested heavily in this area and have continually increased their market share at the expense of the traditional

Retail market measurement

chemists. A trade sector which combined grocery stores with chemists would not have been able to monitor this change over time. The requirement to define trade sectors tightly is sometimes at odds with the needs of the retailers. A number of retailers are very protective of their data confidentiality and prefer them to be aggregated with data from a large number of retailers. This a situation which the research agencies need to manage continually. A further issue is the availability of data by trade sector; for example in the United Kingdom the grocery stores can be monitored through the provision of scanning data whereas the drugstores/chemists require a combination of audit and scanning data to provide complete coverage of this trade sector. Having identified the trade structure them in a manner typically involves grouping hierarchy. An example for the

sectors of significance, the next stage is to which is logical for the client. This process like stores together and placing them in a health and beauty market is shown in Figure 7. Figure 7 TOTAL

GROCERS

&

ALL

OUTLETS

DRUG STORES

CO·OPS

IND. GROCERS

&

DRUG

Once the broad trade sectors to be covered have been identified, the next step is to establish the universe. Establishing the universe Having established the trade sectors which the client wishes to measure, the next key task is to define what the universe will be, i.e. which store types do we aim to cover. Sources of information to establish the relevant universe vary significantly and can include: o o o

government statistics or census on the number of stores information sourced directly from the retailers original research on the universe structure.

Obtaining universe information on the centralised retailers which participate in retail market measurement services is relatively straightforward as they are able to provide universe details for their own stores, i.e. number of stores, sales per store. The difficulty arises in defining the universe for the less sophisticated end of the retail trade. This is a particular issue in countries such

251

252

Tim Bowles, James Brooks

as the eastern European countries where market controlled pavilions are significant outlet types.

stalls and government

Where information is not directly available from the retailers or government sources (even when it is, it can be less than reliable or out of date) the research agency will need to carry out an establishment survey to determine the number and size etc. of retail outlets which are in the universe for a given trade sector. The establishment survey for any retail measurement service is a significant piece of work and underpins the accuracy of the service as a whole, because it will be used as the basis for projecting from the research sample to the total market estimate. The establishment survey can be carried out by taking a random sample of postal districts in the country being measured. Researchers will then count the number of shops of each type within that postal sector. They will also collect information such as store size, products sold, number of checkouts and average sales. These characteristics can then be used to aid the selection of a representative sample. This random sample of postal sectors can then be projected to represent the total country. The key output of an establishment survey is the number of stores within each trades sector. An example is shown in Table 2. 0% 20% 10% Retailer 100% 0% 9%Table 2 Universe sizeScan 100% 9,630 22,430 14,860 4,370 33,200 26,800 2,400 2,900 57,500 l,800/ /Source Survey retailer Survey retailer Survey

The Table also shows how the universe information was obtained, whether directly from the participating retailers or from the establishment survey itself.

Retail market measurement

Selecting and projecting the sample The sample of stores used to represent the universe will be dependent on the level of detail required for each trade sector. The sample needs to be constructed for each individual trade sector for which data are to be reported. In the trade structure shown in the previous section, this would entail constructing individual samples within each of the trade sectors such as multiple chemists and major multiple grocers. The individual trade sectors can then be combined in order to give an aggregate measurement of total market performance. For trade sectors where there is 100% scanning and a high level of retail cooperation, it is possible and in many cases desirable to process census data, i.e. process data for all stores and not just for a sub-sample projected to represent all stores. This has the advantage of being 100% accurate in terms of monitoring sales. Whilst this route may be possible it may not be pursued, due to cost constraints either in terms of the fees levied by the retailers for the data or the costs incurred in processing such high volumes of data (a chain of 500 stores selling in excess of 20,000 lines represents a very large amount of data to be processed!). Where census data are not being used, it is necessary to select a sample of stores to represent the universe very accurately. The aim is clearly to select a profile of stores which matches the profile of the universe. It is then a relatively straightforward process to project the sample stores to represent the total universe. The first stage in defining the sample is to establish which criteria are to be used to balance the sample. These criteria will be factors which influence the sales within a given store and may include: '0 store size o location o sales volume o o

presence of certain departments (e.g. fresh foods, pharmacy) age of store.

The larger the number of factors used to stratify the sample, the higher the accuracy which can be obtained in projecting to the universe. A higher number of stratification factors will also, however, necessitate a larger sample SIze.

Once the sample has been identified, the sample can (using the relationship between the sample and the universe) be projected to represent the total universe. Where available this relationship can be established based on actual sales. Where it is not available it can be based on the number of stores.

253

254

Tim Bowles, James Brooks

In its most simple form, the relationship between the total sample and universe will be applied to each individual store. For example if the universe sales are $1,000 million and the sales at the sample stores are $120 million, then the sales of the total sample will be multiplied by 8.3 ($1,000/$120). In the above scenario each individual

store in the sample is given the same weighting. It is however possible to assign different weightings to different stores in the sample. This technique is more likely to be used in situations where it is difficult to select the sample with a profile identical to that of the umverse.

Time periods The time periods reported within a study are driven by the requirements of the client, together with the data sources which are being used. Scanning data are typically collected and processed on a weekly basis, therefore it is feasible (and most commonly done) to report weekly data on a four weekly basis (i.e. every four weeks, four individual weeks of sales will be delivered). This enables the client to align the data more closely with the activity in the marketplace or any other conditions which may impact sales. In addition weekly databases also allow the client to align the data to time periods which they may use internally for reporting data. Whilst it is possible to deliver weekly data each week, this is not yet common practice due to the costs involved in such delivery. Where audit data are used to represent a trade sector it is very unusual for data to be reported weekly. Since auditing costs are a major issue, it is typical to report audited trade sectors on a four weekly, monthly or bi-monthly basis. Where trade sectors combine audit and scanning data to provide broader trade sector coverage, data will usually be reported at the lowest possible denominator, i.e. if weekly scanning data are being combined with fourweekly audit data, the trade sector will typically be reported four-weekly. This is illustrated in Figure 8. In some cases where the less frequent audit data represent a small proportion of the data which are to be combined, the audit data can be weighted and split into weekly blocks weighted to the profile of sales shown in the weekly scanning data. This can only be done when:

1. the audit data are a small proportion of the total 2. the audit stores are similar in profile to the scanning stores. It should be stressed that this is not an alternative to scanning data but is a method for integrating relatively small volumes of audit data into scanning data.

255

Retail market measurement

Figure 8 Trade sector 'B'

Trade sector' A'

Combined trade sector

Week] 4 week

4 week

period

period

4 week

4 week

period

period

+

Week 8

Products A significant element in structuring a retail market measurement database is how the products are organised within the database. A typical database for one product category can have in excess of 5,000 product codes. Failure to organise the products effectively will result in a database that is not of use to the client - imagine trying to find a single product code in a list of 5,000 items! The structure of the product dimension of the database will be dependent on the individual market and the client. The first stage in organising the database is to define the attributes which will be used to segment the data. Typically these attributes are the product characteristics that the consumer uses in a purchasing decision (or how the client segments the market for marketing purposes). Examples of product attributes are: o o o

pack type, e.g. bottle, can or jar fragrance, e.g. lemon, strawberry size, e.g. 200 ml, 300 ml.

In addition to the product characteristics the brand and manufacturer are usually treated as attributes to help in structuring the database. The attributes can then be used to construct a product hierarchy that can in turn be used to structure the order of the products in the database. An example product hierarchy for shampoo is shown in Figure 9. In some cases it is necessary to group different attribute values together in

order to prevent the database becoming unwieldy, and to enable the user to emolov the attrihlltes to strncture the analysis An examnle is :lir frp."hpnpr"

256

Tim Bowles, James Brooks

with different scents such as lavender, rose and honeysuckle, all being defined as 'floral', thereby removing unnecessary detail.

Figure 9 AN EXAMPLE PRODUCT HIERARCHY FOR SHAMPOO Total category - shampoo Type e.g. medicated versus non-medicated Supplier e.g. Henkel, Procter and Gamble Brand e.g. Timotei, Wash and Go Hair Type e.g. permed, greasy, normal Size e.g. 200 - 250 ml, 251 - 350 ml Bar-Code

The products are then sorted according to the product hierarchy allowing structured analysis of the data. For example, as well as examining performance of specific lines or products, it is also easy to analyse the performance of total sectors; for example how fast are 2-in-l shampoos growing versus standard shampoos. This prevents the user having to sort through individual lines and allocate them to different sectors. An example of how this may be done is shown for the stir fry sauces market (Figure 10).

Figure 10 100% 90% 80% 70% 60%

D Other Sauce III Stir Fry

50%

IIlITex·Mex

40%

lIT]

30%

DOriental

Traditional

• Indian • Italian 10% 0%

Retail market measurement

Collection of attributes By necessity, attributes are usually collected in the stores involved in the study. A field worker will identify the product that needs coding and then record all the attributes for that individual product. In order to ensure consistency, all attributes must be observable and tangible. Subjective attributes, e.g. premium versus standard, are not typically used as this requires judgement by the field workers and may make allocation inconsistent. At the initial database construction, all products will need coding and this can be a large task. Coding of products which are no longer being produced needs some judgement to be exercised by a market expert. Recent developments involve field workers video-recording products and coding attributes from the video image. This helps to ensure consistent attribute coding, but also means that databases can be re-coded if a new attribute emerges as being significant in the future. As new products are launched, they will also need to be incorporated into the database. Usually the first the research agency knows about a new product is when it appears in the data which are supplied by/collected from the retailer. Speed is of the essence in collecting the attributes in order for the product to be incorporated and accurately reported within the database. Given the fact that it takes time to collect the attributes, products may not immediately appear in the database. For this reason it is usual each period to re-process the previous period's data, to ensure that new product sales are fully included in the database. Clearly the objective of the agency is to introduce the product as quickly as is possible in order to minimise these data changes with respect to the previous period. The client can assist by providing example products before they are launched. This ensures that the product is placed in the database prior to it appearing in the shops.

Measures To some extent, data measures are the most important part of a retail market measurement study, as these are the data facts which are being reported for the combination of a specified time period, trade sector and product. The availability of measures varies considerably with the source of data being used. A scanning database can potentially have in excess of 160 measures. This coupled with weekly data for 5,000 products can generate a huge volume of data for the client. The key to data utility is in distilling the measures which are of value to the client to ensure that they are not swamped. The core measures which form the basis for all other measures can be defined in the following groups: 1. sales 2. distribution / rate of sale 3. price 4. promotional incidence and response.

257

258

Tim Bowles, James Brooks

Each of these measures is taken in turn in the following sections. Sales measures Typically, three types of sales will be used: o o o

unit sales: the number of packs sold value sales: the value (local currency) of products sold volume sales: expressed in volume equivalency, e.g. kilograms, litres etc. (value sales will typically be reported net of any promotional discount).

Sales measures are the basis of calculation for many of the share measures which are used within a study. The most commonly used share measure which allows a manufacturer to track his relative performance within the market is 'category sales share'. This can be defined as: Sales of product(s) Category sales share = --------x Sales of total category

100%

A manufacturer may choose to use this measure with an individual line, brand or total supplier. Similarly, retailers may use this type of measure to monitor their share of the market by dividing their sales with the sales of the total market. Distribution measures Distribution is the reach a product has within a given trade sector and can be derived in one of two ways: o o

numerically weighted by sales. Numerically

Numeric distribution refers directly to the percentage of stores carrying a product. For example in a universe of 2,000 stores, if a product is listed in 620 stores its numeric distribution would be 31 %. Given that not all stores are the same size (for example, compare a small convenience store with a hypermarket) the client may wish to weight the distribution according to the relative size of the different stores, i.e. reflecting the importance of different stores in calculating the product distribution. The stores may be weighted according to sales of all products stocked (All Commodity Volume: ACV). Distribution is then calculated by adding the ACV of all stores which sold the product during the period and dividing that by the total ACV of the trade sector which is being analysed, i.e.: Total ACV of all stores which sold the product x 100% ACV weighted distribution = ---------------Total ACV of all stores in the trade sector

259

Retail market measurement

This is illustrated with the example in Table 3 (yes/no indicates particular store sold that product in that week). Table 3 Yes £ISm 342 Store £3Sm £20m No

if that

Yes No BOrn I Store

30+20+1S 100% = 65% ACV weighted distribution = -3-0-+-2-0-+-1-S-+-3-S)<

However, All Commodity Volume is not always the most appropriate way in which to analyse weighted distribution. In cases where a trade sector contains a mix of store types, so that the ACV s represent completely different product mixes, e.g. where chemist chains are reported with grocery stores for shampoo, it may be appropriate to weight the distribution according to category sales. In this scenario, exactly the same formula and methodology is used as above, but ACV sales are replaced with the sales of the total category, e.g. total shampoo sales. When using the distribution measure, care should be taken over the time 'periods and products selected. The lower the level of detail used in defining the product and time period dimension, the more validity the distribution measure will have. For example, using the distribution measure for a total category is likely to result in a high distribution, in many cases 100%. Similarly, using distribution over a long time period increases the likelihood that a store will have sold a given product and therefore increase its level of distribution. Weighted rate of sale Distribution measures are often combined with sales measures to give an assessment of how well a product sells in a store where it is stocked. The measure, weighted rate of sale, calculates the average sales in an average sized store. This measure takes out the effect of a product being distributed across different store types and calculates sales for a particular product in an average sized store:

260

Tim Bowles, James Brooks

Weighted rate of sale = Distribution

Sales for product * x Total number of stores

*either ACV weighted or category weighted distribution depending on the data base.

To illustrate the calculation, consider the following example, which shows the value sales for Product A in each of five stores during each week in a fourweek period: £1£700 Store £ISm £40m £ISk Ok 5423 £4S0 £2S0 £2Sm £Sk

-

£800 £400 £300

£30m ££900 10k Table 4 £200 £SOO

£100

Store 1

ACV weighted

distribution

Total ACV of all stores which sold the product

= ------------------

00 x 1 %

Total ACV of all stores in the trade sector

30+1S+]S

x 100% =48%

30 + 40 + IS + ] S + 2S

This means that an average sized store selling Product A had sales of £1,000 of the product in the four-week period. Rate of sale measures are used most commonly in understanding the relative performance of products in stores where they are stocked. This can help the identification of products with high potential as well as those which are performing poorly. Using rate of sale in conjunction with distribution can enable the client to identify high potential products, i.e. those with a high rate of sale and low distribution, and potential products for de-listing by a retailer (those with high distribution and low rate of sale). Price measures (see also Chapter 20). As sales for the stores contained within the sample are a combination of sales from different stores, the price which is reported from a retail market

Retail market measurement

measurement survey will be an average price. An average price can be reported per unit or per volume (e.g. per litre). The price per unit is simply the value sales divided by the unit sales and reflects the price paid by the customer. It is at its most useful when reported at a product line level, as it can be ntisleading if reported at a higher level. For example the introduction of a small pack size within a brand can cause the average price per unit of the brand to fall, but it does not mean that the brand has reduced its price. Further pricing measures are also available as part of the promotional elements of retail market measurement which are covered in the next section. Promotional measures are of such significance that they will be covered in an independent section. PROMOTIONAL ANALYSES The aim of the promotional element of retail market measurement is to link the incidence of in-store causal activity (i.e. in-store promotions) with the sales achieved. This enables two key measures to be derived: o o

incidence of promotional activity (i.e. reach and depth of activity) response to promotional activity (i.e. the incremental sales generated).

Due to the fact that weekly store level, EAN level data are required for promotional analysis (as will be explained later) this form of analysis is restricted to trade sectors for which there are scanning data.

Methodology Causal data are collected at weekly level by individual bar-code from stores which are within the scanning universe. These causal data are then linked to the sales data in order that the sales can be split into base and incremental sales. Base sales are those which would have been expected without promotional support, whilst incremental sales are those which can be attributed to in-store promotions. Causal data are typically collected by a fieldworker going into a sample of stores and scanning the products which are on promotion with a hand-held scanner. The details of the promotion are then keyed into the scanner. This ensures that the promotion details are collected at the EAN level and can therefore be directly matched to the sales data provided by the retailer. Figure 11 shows the sales for an individual bar-code and the incidence of causal data for the same bar-code.

261

Tim Bowles, James Brooks

262

Figure 11 600 500 400 300 200 100

o

.""""

1'1'1'"

1-

Actual sales

I'"

I"

•

,"",'"

1'1'1'1"""

Promoted weeks

The next stage is to use an algorithm to calculate the 'base sales'. One method is to take the store level data and create an exponentially smoothed curve of sales during weeks without promotion, and then to apply this to promoted and non-promoted store weeks to generate total base sales. Incremental sales can then be calculated as the difference between total sales and base sales. Since incremental sales are the sales attributed to promotional support in weeks with no promotion, base sales are equal to actual sales. An example of base and incremental sales is illustrated in Figure 12.

Figure 12 600 500 400 300 200 1001~ o

II II II II II II II II I'll II II "

-

Actual Sales

•

Promoted Weeks

•••

II I' IIII I' II II I' II

Base Sales

The key thing to note is that for a base sales calculation to be effective it needs to be conducted at both the bar-code and store level and then aggregated to the total market.

Causal conditions Experience suggests that the impact of different types of promotions varies significantly. In order to appreciate the impact of and to quantify the volume of promotions, different promotions are classified into different groups, although all would trigger the base line calculation.

263

Retail market measurement

Causal conditions to be collected and monitored will be dependent on those which are evident in the country or market which is the subject of the study. Example conditions may be: o o o

price reduction display multi-offer

o o

loyalty card points given with product print advertisements.

The purpose of classifying causal conditions separately is to allow promotional incidence and response to be monitored for individual promotional mechanics. As the causal data are collected in the store this ensures that what is reported is in fact what was on offer to the customer. Whilst it is possible to base-line using promotional programmes provided by a retailer, these are not always in reality implemented in the store.

Reporting As outlined, the two key measures of a promotional study are: o

depth and reach of promotion, increase in sales due to a promotion.

Depth and reach of promotions In order to establish the reach of a promotion it is normal to look at the percentage of non-promoted sales which were sold with the given promotion. This is illustrated in the equation and example below: , Base volume sales for promotion % of base volume sales for a promotion = -----------x 100% Total base volume sales Consider the volume sales for Product A that were due to print ad, no display: Week 1

Week 2

Week 3

Week 4

Quad Wk

Total base volume

33 g

36 g

25 g

26 g

120 g

Base volume sales with display

10 g

llg

8g

9g

38 g

38

% of base volume sales with display = -

120

x 100% = 31.7%

The use of this measure allows the client to understand the extent to which brands or products are promoted compared with competitor products within the marketplace.

Tim Bowles, James Brooks

264

Increase in sales due to promotion This measure identifies the percentage sales promotional condition being considered. A increase is that the calculation should only be as a base. This allows the true incremental identified. The calculation can be defined as:

increase which was due to the key factor in calculating the done using the promoted stores effect of the promotion to be

Incremental volume + week after incremental volume x 100%

% increase in volume sales = Base volume sales in stores promoting Week 342 18 120 4g 25 6g 36 26 3gg gvolume g Consider QuadWk the sales for335g Product A that were due to any deal: Week 1

18

% increase in volume sales with any promotion = -x 120

100% = 15%

In the above equation, the week-after effect has been included in the incremental sales. Where it is thought that the effect of a promotion may last longer than the time when it was in evidence e.g. print ads, the incremental sales of the following week of the promotion may be included. Figure 13 VOLUME OF PROMOTION VERSUS SALES UPLIFT % Volume increase 100

-,

----------------------l(ll-I-ta-Ii-a-n-------,

90

°Oriental

o

80

Stir Fry

o

70

Indian

.•

0

Traditional

60

o Tex-Mex

50

40

I 15

I 20

25

30

35

40

45

% Volume sold on promotion

50

Retail market measurement

The use of this measure allows the client to understand what is the most effective promotional mechanic for a brand or product and which products or brands respond best to promotions. This coupled with information on the level of promotion helps the client to understand which products it may be most appropriate to promote more or less in the future. This concept is illustrated in Figure 13. The arrows demonstrate products for which the client could either consider increasing or decreasing the level of promotional support, i.e. stir fry products generate good response when promoted but receive a relatively low level of promotional support. DISAGGREGATE USE OF SCANNING DATA The previous sections of this chapter have focused on how retail market measurement studies are constructed through aggregating data provided from individual retailers, i.e. sales are combined at an EAN level across all stores within a trade sector so that total sales are reported for an individual product line. The use of disaggregate level data focuses on using store level data at the EAN level to perform more sophisticated analyses. The purpose of using disaggregate level data is to increase the number of sales observations which can be used, thus increasing the ability to build a cause and effect relationship between sales achieved and the factors affecting them e.g. price, promotions and advertising. Taking a single product with two years of weekly data, there will be 104 sales observations (fifty-two observations per year). Using store level data can increase the number of sales observations to the tens of thousands and therefore means that the ability to conduct sophisticated analyses is greatly increased. As well as increasing the number of observations, using store level data ensures that averages are not used. Table 5 illustrates why the use of disaggregate analysis can be important. 41 493 £2.59 £2.9915 £2.88 £2.82 £2.24 £2.70 Price Sales Sales£2.99 46 112 17 £2.88 £2.82 £2.24 Week one

Week two 5 Table

265

Tim Bowles, James Brooks

266

U sing the total market data this table would suggest that as the price has increased from £2.59 to £2.70, sales have gone up from 41 to 46 units - an ideal outcome! The reality is, however, that the sales in stores charging the highest price have increased. So although no stores changed the selling price, the total market data would imply that there has been an increase in price. A further key issue is that because the market price is an average, it is not a price which was ever paid by a customer. This is why disaggregate level data are valuable in planning and evaluating the impact of different marketing initiatives. Multiple regression techniques can be used to isolate the impact of sales of different marketing conditions. Often the output of these models can be built into software packages which allow the end user to play out 'what if' scenarios. This provides ultimate utility for clients, allowing them to assess the impact of changing an element of the marketing mix without incurring the cost of doing so. In addition to regression analyses, it is also possible to group stores according to different characteristics to understand the impact different conditions have on sales performance. These analyses are often referred to as store group analyses and together with regression analyses will be discussed further in this section.

Regression analyses Sales of a brand are influenced by a variety of different factors - price, promotions, TV advertising, competitor activity, etc. The impact of each factor will differ in magnitude. Using store level data, which enable the measurement of the within store change in sales against the within store change in price/marketing condition, it is possible accurately to quantify the impact of each variable on brand sales via a custom multiple regression model. Sales = f (base price + promotional price + in store materials + print ads + displays + multi-buys + link saves + special packs + television advertising + consumer promotion + seasonality + etc.) Store level data are not only the most accurate sales data available, they also avoid having to use averages such as 'average price'. The issues associated with using 'average price' were outlined in the previous section. Most importantly, store level data have an abundant supply of sales observations. One brand will typically have well in excess of 30,000 robust sales observations for inclusion within any model, allowing the different marketing variables to be very accurately isolated and their impact on sales quantified.

Retail market measurement

Following construction of the model, simulations can be run to identify the impact on sales of a change in the marketing mix - a reduction in the presence of special pack (e.g. a pack with 10% extra free), for example. Dedicated studies to understand the contribution of everyday price, advertising or promotions, using the same techniques as above, are described below: Base price elasticity One of the most important issues in marketing is understanding how sales respond to changes in everyday (base) price. To understand this, an analysis of base price elasticity can be conducted assessing: o o

How sensitive are a product's base sales to a change in base price? How sensitive are a product's base sales to changes in the competitor's base price?

The answers to the above questions can help manufacturers better manage everyday price. This analysis is useful for strategic marketing decisions. Specifically, a manufacturer can use this analysis to evaluate historical sales response to changes in price and use this to measure what may result from future price changes. The base price elasticity (degree of sensitivity) of a product is determined through a store-level custom multiple regression model. The regression model examines within-store changes in base sales as a function of within-store changes in base price. The output of the regression model can be built into a software package, which will allow the client to play out 'what if' scenarios. Promotional price elasticity Consumers respond differently to everyday (long-term) price changes than they do to promotional (short-term) price changes. It is important to understand these two elements of pricing: everyday and promotional. The base price elasticity analysis addresses the long-term price elasticity of the product, while promotional analysis will explicitly address the short-term price elasticity of the product. Typically, most clients wish to have both elasticities quantified for their products when issues regarding price need to be addressed. Similarly to the base price elasticity, the promotional price elasticity (degree of sensitivity) of a product is determined through a custom multiple regression model built using store-level sales data. The regression model examines the increase in sales associated with a temporary reduction in price only, or a temporary price reduction accompanied by other trade conditions (e.g. in-store materials, display, etc.).

267

Tim Bowles, James Brooks

58

An example of this form of analysis is shown in Figure 14 which demonstrates the increase in sales which are achieved through different levels of price discount in conjunction with other promotional initiatives. Figure 14

Sales

0 200 400 1000 300

600 500

5

10

15

20

25

30

35

40

45

50

% Reduction in price TPR = Temporary price reduction

Effectiveness

of TV advertising

on sales

The premise behind reading the effects of television advertising is that sales over time are a function of marketing stimuli over time. In order to determine the element that television advertising contributes to sales, a store level custom multiple regression model is utilised. The analysis will, typically, enable the following questions to be answered: o o o o

What contribution to sales did television advertising make? What would sales have been, had the television advertising never taken place? Are our television advertising campaigns more effective this year than last year? Are there certain regions of the country which are more responsive to our television advertising than others?

Knowing the answers to the above will provide extremely valuable information. By precisely understanding the impact of television advertising, manufacturers are equipped with the knowledge to make more effective decisions with respect to future television advertising spending.

Store group analyses Store group analyses have many different applications. The underlying feature

Retail market measurement

269

of these analyses is the division of the sample into sub-sets, e.g.: those stores featuring a certain condition vs. those stores where the condition is absent. An example of a store group analysis is comparing the rate of sales for stores with different price points for a given product. An example of this is grouping the stores by the price charged for a given product. In Figure 15 the average sales per store week are shown for each of the price points, together with the frequency with which the price point was evidenced. This helps to identify critical price thresholds for a given product. In this example several price thresholds

can be seen, at which sales drop significantly when the price is increased. This helps the client in positioning the price of a given product. Figure 15 A verage base sales per store week

% Frequency

300 250

-

Product A

•

% Frequency

I 25 20

200 15

150 10

100

50 o

o

Base price

Optimal mix analysis The mix of individual lines that can be found on shelf is an important element in maximising the overall strength of the business for a brand. For example, a manufacturer may produce eleven variants of a given brand. However, not all retailers will carry the entire range. So understanding the mix of these items will assist the manufacturer in answering the following questions: o o

How do my sales respond to an increasing number of items carried? Does a common pattern exist, where the same 'core' items should always be carried?

270

Tim Bowles, James Brooks o

What is the optimal mix of products when seven items are carried?

In-store custom audits Observable data are collected, at bar-code level, by the fieldforce using handheld scanners (typically at the same time as collecting causal data outlined in the previous section). This can be matched to sales data for the same group of stores in order to provide insight into the impact of the in-store positioning of a brand on sales. The information can be used in the following ways: o o o

Determine the number of facings for a selection of chosen products. Ascertain the position on shelf of a product or range of products. Identify if a product is under or over-faced given the share of shelf space allocated versus brand share.

o

Assess if a product performs better when it is located above, below or adjacent to a complementary or competitor product. Determine how a product performs in stores where there is one facing versus in stores where there are a greater number of facings.

o

RETAIL MARKET MEASUREMENT: COMMERCIAL REALITY The previous sections of this chapter have demonstrated that the transition of retail market measurement from audit to scanning based technology has led to a 'sea change' in the services available. This in turn has dramatically changed the context in which these services are used. The primary use of these services is at the retailer and manufacturer interface. The retailer is the custodian of shelf space and hence the sales opportunity for manufacturers. As competition between brands increases the battle for shelf space becomes ever more critical. In addition these brands are also competing with the retailer's own products. This fierce competition, together with a much enhanced service base, has driven the concept of data-driven marketing. In an environment in which marketing decisions are made on fact-based data it is critical that retailers and manufacturers truly understand the factors which are driving sales. In reality retail market measurement has changed radically from an industry concerned with tracking and monitoring sales to one which is concerned with understanding the factors which drive sales. The examples of the services provided in previous manufacturer and retailer to manage the key areas of: o o o o

pncmg rangmg promotional planning new product introductions.

sections

allow the

Retail market measurement

The above have been identified as four of the key areas for focus within category management and efficient consumer response initiatives by both retailers and manufacturers alike. The advent of category management over recent years has put greater focus on using fact-based information to tackle some of the issues highlighted below: Pricing a a a a

What is the significance of different price points to consumers? What is the optimum price gap between brand X and brand Y? How will sales respond to a 10% increase in the base price? Will a drop in price of 10% generate sufficient sales to cover costs?

Ranging a a a a a a

Identify items which are performing well in the market but not stocked by a particular retailer. Identify high growth areas with the potential for own label development. Assess which categories over or under perform compared with the norm. Identify which categories offer the greatest potential for growth. Identify items with a high rate of sale with a low distribution and vice versa to determine which items could be substituted. Determine optimum product mix to generate the highest rate of sale for the total category.

Promotional planning a

Establish which products respond best to promotions.

a

Identify the most effective promotional mechanic for an individual product (e.g. is' a gondola end more effective than a 10% price reduction)? Assess the profitability of any given promotion. Forecast likely stock requirements for an up-coming promotion. Forecast impact of different levels of promotional price. Identify the impact at the category level of promoting an individual product or line.

a a a o

New product introduction o

Weekly sales data, by store, allow the impact of sales of new items to be monitored.

o

Identify high growth areas with the potential for own label development.

As can be seen, the answers to the issues raised represent a great opportunity both for manufacturers and retailers alike, in more efficiently managing their respective businesses. The reality is that rather than being a research

271

Tim Bowles, James Brooks

272

mechanism, continuous data have now become an integral part of business management. The analyses only really scratch the surface in the power of retail-based data. The next stage is sure to be an even greater exploitation of store-by-store data which allows decisions to be taken, not just at a total chain level but at an individual store level. This will allow both retailers and manufacturers fully to exploit the wealth of information by tailoring their marketing programmes to the individual store environment. There is, however, a sting in the tail. These types of services are only possible where retailers agree to participate. They have their greatest value the more transparent the data in the marketplace becomes. At its most transparent, a retailer will make its own sales available to a manufacturer in order that the manufacturer can invest in developing both its own and the retailer's business. This is an increasing trend within the data market. The thirst for information is not limited to retail-based data. Given the issues which data are now being used to address, integrating retail data with other data sources, e.g. internal data, geodemographics, profit information and advertising spend, is critical in order for clients to be able to take informed decisions. ACKNOWLEDGEMENTS The author appreciates the contributions of Mario Lesser, AC Nielsen; and from Peter Buckley, Mike Campbell, Bruce Dove, Mark Dye, of IRI InfoScan.