Competing brands sell to near identical user bases. Your customers are just like competitors customers, and vice versa. Brands compete for share of a single unsegmented market rather than each filling a specialist niche.

1. INTRODUCTION

A great variety of approaches to market segmentation have long been discussed and practiced, as reviewed for example by Dickson and Ginter (1987), Lury (1990), Dibb and Simkin (1996), Anschuetz (1997), Cahill (1997), Campbell (1998), Mitchell (1998), Gunnarson (1999), Kotler (1999), and others. Lilien and Rangaswamy (1992) typically said that “Segmentation, the process of dividing the market into consumer groups with similar needs, is essential for marketing success”.

Segments of apparently similar consumers may even be given names (e.g. “Worrier”, “Sensory”, “Sociable”, or “Independents”) with advertising and products targeted accordingly: Crest toothpaste for those who want to stop decay, Macleans for Whiteners, Colgate (stripe) for those interested in flavour, and so on (e.g. McDonald and Dunbar, 1998). Cornish (1990) has said that only by observing how interests vary between such groups can one work out the reasons why people behave the way they do.

Many organisations therefore invest heavily in segmentation studies. Some research agencies have become specialists, each mostly focusing on a new or proprietary segmentation analysis technique, as illustrated in the box below.¹ Our search of the literature has revealed no studies where demonstrable brand segments were shown to exist and to be stable over time, i.e. that brand A in fact appeals to a markedly different population of consumers than does the competitive brand B.

Collins (1971) however noted long ago that clustering or segmentation techniques typically ask what is the “best” grouping of the data that can be made, but ignore the more basic question whether any useful grouping of the data can actually be made at all.

Segmentation Literature mainly deals with Methods

In this paper we therefore ask whether the user-profiles of directly competing brands really differ. The results say “NO”, i.e. brand segmentation hardly or seldom exists. This conclusion is consistent with the fact that other papers in the voluminous segmentation literature do not actually report any coherent brand-segmentation results.

Functionally different variants of a brand may be for different needs, e.g. large and small pack-sizes; tamarind as well as tomato flavours; 2-, 3-, 4- and 5-door car models, or other “Stock-Keeping Units”/SKUs more generally (e.g. Singh et al 2000). And each of these variants may have its own more or less “segmented” following. But because of competition, all brands tend to have each of these SKUs (except for marginal ones). That is why profiles of brands as a whole  – in fact umbrella brands – need not differ (Ehrenberg et al 1997, Ehrenberg et al 2000).

We build in this paper on previous findings of a lack of brand segmentation which were based on consumer-panel data (Hammond et al 1996). We now greatly extend these earlier findings to more categories, far more potential segmentation variables, and a different type of data. The results agree with the Titford and Clouter (1998) view that for competition between brands, a buyer is a buyer is a buyer, whosoever he or she may be. The lack of brand segmentation explains why the real marketing issue is not “who buys”, but “how many buy”.

Exceptions with distinct sub-markets or functionally different groups of brands can show up. But they are rare in the extensive brand-data analysed here. We will also note that the practical sales implications of any brand segmentation may in any case be overblown.

2. THE DATA ANALYSED

Our analyses are based on extensive tabulations from BMRB’s Target Group Index (TGI) for 42 industries in the UK. The TGI is an on-going self-completion survey of brand buying and consumer attitudes across many categories including fmcgs, durables, and financial and leisure services, together with standard demographic and some media usage information. For each category, respondents state if they buy/use/serve the category, how often, and which specific brands. Potentially relevant attitudes, opinions and interests are captured through over 200 attitudinal, lifestyle, and need statements with which respondents rate their degree of agreement on a 5-point scale. Some examples are:

The standard TGI sample is 25,000 adults annually, but the actual user-base for a brand can be as low as 500 (e.g. see Table 1). Two batches of TGI data were analysed, with the same outcome. A first batch was for all brands in 13 industries with 100 attitudinal variables; the second batch of data, some two years later, compared the top 10 brands in 30 industries, with over 200 attitude variables. (Light bulbs were common to both batches and gave very similar findings, as did in effect everything else.)

As an example of the raw data, Table 1 shows the number of TGI respondents who used each of the top 10 credit card brands in the UK, broken down by Gender and Age. The full file is much larger, with 280 such variable-columns (i.e. the other demographics, media, and attitudes). The brands in the table are ordered by market share, to facilitate pattern recognition.

3. THE ANALYSIS: Deviations from the Average Brand Profile

Instead of using the complex but to us opaque segmentation techniques such as AID or CHAID that are commonly used in segmentation studies, we more simply sought to contrast the profiles of the different brands in each product category. We did so parsimoniously by comparing each brand’s profile against the profile of the average brand in that industry. This is a simple yet effective process that we now explain.

To bring out the make-up of each brand’s customers, percentages profiles were calculated from the raw data as in Table 1, together with the profile of the (unweighted) average brand (in effect a category profile). This is illustrated for Gender and Age again in Table 2. (Further aspects of data presentation that facilitate pattern recognition are briefly noted in the Appendix.)

The result showed few differences between the brand profiles. Thus in Table 2, on average 54% of card users were male. Eye-balling shows that this was much the same for each brand. Again for age, all the credit cards had very few customers under 24 (4% on average), and about 20% in each of the five larger 10-year or so age-groupings that the TGI used.

To quantify these brand similarities explicitly, the deviations of each brand’s profile from the average profile were calculated, as illustrated in Table 3. To summarise all these deviations, the average size of the deviations (ignoring their sign) was computed for each measure and for each brand, i.e. the traditional Mean Absolute Deviation or MAD, as shown.

All individual deviations over 5 percentage points were marked in bold in our computer output, to signal any larger differences. This cut-off point of 5 serves to highlight the few larger deviations in the tables – the largest in Table 3 for instance is 6 percentage points for Barclays MasterCard. But even such larger differences in profiles are generally far too small to merit any distinct marketing action. Thus Barclays MasterCard appealed to 60% males in Table 2, instead of the average of 54%.

For the credit card brands here there was virtually no difference in their users by gender (an average MAD of 2, even including the larger Barclays value), or by age (lower MADs of 1, since the absolute numbers were smaller with more classes). Table 4 illustrates this also for five typical attitude/lifestyle variables: most of the MAD’s still average round 1 or 2 points.

This form of analysis was repeated for eve-ry variable, for over 110,000 individual deviations in the 42 industries.

4. THE RESULTS

Overall, the individual brands’ percentage profiles deviated from each other by an average of 2 or 3 percentage points, which in our view is small – in effect zero. The difference between 15% of users of Card A saying “I try to keep up with technology” rather than Card B having only 11% technofreaks is not actionable.

Only around 8% of the individual deviations were more than 5 percentage points, and even these larger deviations averaged at only about 9. Just 2% of individual deviations were 10 points or more. Noticeable deviations were therefore exceptional, and were still small even when they did occur. Brands therefore rarely differed from the average brand in their category, and when they did so it was not by much, nor was it of practical importance.

Table 5 presents the global results, that is the average MADs for all of the demographics, all the media variables, and all attitudinal variables, for each of the 42 categories (ordered on their total unweighted MAD’s for visual convenience). Three of the average MADs are greater than 3 (namely, 6, 5, and 4 for Cigarettes, Tessas and Cat food) but are explained simply by their being based on relatively small samples.

The telling figure is the overall average MAD of 2 (or 2.1 more precisely) in the bottom right-hand corner. This pinpoints the general lack of effective deviations between the brand and category profiles.

The occasional larger deviations which occur sometimes cluster for several brands and relate to a submarket rather than to a specific brand (as it would have had to in order to characterise real brand segment). But such subpatterns are usually already well-known, or “as to be expected”.

Regionally for example, there is a Scottish sub-market for three locally-based Scottish banks (Bank of Scotland, Royal Bank of Scotland, and Clydesdale). And for RTS breakfast cereals, children somewhat prefer the pre-sweetened types (i.e. “children’s brands” – see already Hammond et al 1996, or ask Kellogg’s). This also seems to show up here amongst families with children and the types of confectionery they buy. Some clustering of responses across different categories are also noticeable for “Green”, for Diet, and for Exercise. Such systematic subpatterns in the data can be pursued further (best probably by someone with specialised knowledge of that market). But they are small and rare, especially among the ten or so larger brands in each category that we have mostly analysed here.

5. THREE TECHNICAL QUESTIONS

Three technicalities arise, for (i) the choice of variables, (ii) measurement biases, and (iii) sample sizes.

(i) Have the Wrong Segmentation Variables been used?

It seems highly unlikely that potentially powerful variables have been consistently omitted in a widely-used and long-running and virtually public measurement tool such as the BMRB’s Target Group Index. But it is easy to check any new candidate segmentation measure.

A particularly relevant potential segmentation variable for any given brand X is the other brands which buyers of X also buy – is there any clustering, e.g. buying of brand X going with buying of brand Y but not of brand Z? But consumer-panel data has already shown virtually no such brand segmentation (average MADs of 3 points, on somewhat smaller samples than here – Hammond et al 1996).

Such lack of segmentation by the usage of competitive items has also been the case with very extensive panel-based segmentation studies of television viewing, e.g. cross-analysing viewers of programme A by the other programmes they also viewed, whether of the same or a different genre (Ehrenberg 1986; Barwise and Ehrenberg 1988).

(ii) Possible Measurement Biases?

BMRB’s TGI covers many different products and potential segmentation variables in a lengthy self-completion questionnaire. This could be subject to measurement biases. Such biases would however matter little here since they would be much the same for the different brands that are being compared. Nor have we found order effects.

(iii) Sample Sizes?

Sample sizes of category users in the TGI are mostly large, averaging at 10,000 as noted above. (They could easily be increased by using TGI data over two or more years.) But for smaller brands, samples of their users are of course smaller – typically they are less than 1,000 for the five smaller brands in Table 2. This shows up in the slightly larger MADs for these smaller brands as in Table 4 (as was also noted by Hammond et al 1996), although it is exceptionally not so in Table 3. Typically also, the three larger MADs in Table 5 – greater than 3, for Cigarettes, Tessa accounts, and Cat food – were based on relatively small samples of less than 5,000 category users.

6. TYPES OF SEGMENTATION

Three broad types of possible segmentation need to be distinguished, by Brands, Category, and SKUs:

Brand Segmentation

This refers to possible differences in the make-up of the users of different brands, say Kodak, Agfa, and Fuji 35mm film. Such differences are rare, as we have seen here.

Category Segmentation (or sub-category segmentation)

This would relate to profile differences between users of functionally distinct products (e.g. users of Cameras versus users of Projectors) or users sub-types of products (e.g. Professional versus Disposable cameras).

Category segmentation could and does occur, though less widely than may be thought. This is suggested by the low MAD scores for each of the 30 markets in our largest (second) database, as shown in Table 6.

The similarity of the brand profiles across the 200+ TGI variables is repeated, demonstrating that these category markets also have very similar user profiles overall. This is not altogether surprising, since the industries covered in the TGI are typically for household products: the sorts of things most households buy, unlike say industrial goods and services (e.g. filing cabinets or combine harvesters).

In any case, category segmentation is probably more usefully thought of simply as “Knowing your market”. Any sub-markets that arise (e.g. Luxury cars or unscented cosmetics) are usually fairly small and already well-known. It is rare that “sophisticated” segmentation or clustering techniques are needed to discover or rediscover them. It is rare also that use of such techniques has successfully done so in the past. (As already noted, no lasting cases seem to be reported in the literature.)

SKU Segmentation

Different product-variants and even distinct Stock-Keeping Units may each have their distinct followings, i.e. on dimensions such as pack-size, flavour, car-models, etc. and also the myriads of bar-coded interactions of these dimensions (the 1lb size of unsalted Lurpak butter). Despite manufacturers’ extensive concern with product features and corresponding trade-off analyses, almost nothing systematic seems to be known about the market’s reactions to product variants, i.e. no generalisable results about SKU loyalty, switching, or segmentation (Fader and Hardie 1996 and Broniarczyk, Hoyer and McAlister 1998). This is one of the bigger agenda items for the Ehrenberg-Bass Institute in the coming 2 to 3 years.

These three forms of segmentation – brands, categories, and SKUs – are however seldom well distinguished in the segmentation literature (see earlier references). But we believe that they should be, since both for category and SKU segmentation, functional differences are of the essence and usually also almost totally self-evident (tea differs from coffee, and large packs from small ones). In brand segmentation on the other hand, product differences are almost totally absent, or at least covert.

7. THE ROLE OF COMPETITION

The key reason for the prevailing lack of brand segmentation is that products which compete directly and hence are substitutable generally do not differ much from each other overall, e.g. in taste, technical formulation, performance, or any functional feature of importance, including often even their appearance (subject to the legal limitations on “passing off”).

In practice, when a highly differentiated product appeal or positioning concept for a particular brand has been developed, no matter how intensely it may then seem to be preferred by one sector of the market, someone in the company usually soon sets about broadening that positioning so that the brand will appeal to more people and increase its potential market share (Moran 1990).

Competitive brands therefore deliberately aim to be “similar”, so as to be able better to compete. They copy each others’ sales-effective attributes and SKU dimensions, rather than seeking to differentiate themselves² (e.g. Young 1963, Ehrenberg 1974, Sampson 1993, Ehrenberg, Barnard and Scriven 1997, Perris 1999, Ehrenberg et al 2000). Pretty soon almost every brand is trying to appeal also to every other brand’s customers.

And consumers are mostly highly experienced and hence know that brands are at base the commodity with a name, i.e. “that brands are brands”. Hence when there is little functionally to distinguish one brand from another as is so often the case, any reputable brand “will do for the consumer” (Heath 1999). Nonetheless, the individual consumer does tend over time to identify with the brands he or she uses: “I use it, therefore I like it”.

8. IMPLICATIONS

The implications for brand positioning, targeting, and media planning are in our view simple and positive. Instead of being restricted to a small segment (and even perhaps enjoying the proverbial monopoly of a tiny niche), marketers operate in a large, unsegmented mass market, or at least in a large sub-market like luxury cars or dry cat food. However, not being limited to a small niche means also that one’s brand has more direct competitors: there is therefore more scope, and more need, for plain marketing (e.g. promotion, selling, logistics, quality control, advertising, and brand maintenance generally).

If your market is limited to cat owners, it is of course as well to know that. But there is probably no need officiously to strive to restrict your market if you cannot see it to be segmented with your naked eye from well-presented data (i.e. without CHAID or AID or Conjoint). Most often one can hardly avoid stumbling across the fact that it is cat owners who mostly buy cat food (and that some only buy dry cat food). Even a small Usage and Attitude survey would show how far the new product Nutrigrain is or is not eaten when people are on the move at breakfast time (“As Advertised”).

Furthermore, segmentation may not be of great sales importance even when it does occur. While the TGI data show that 6% more males shop at W.H. Smith than do so at other retail chains (Smiths being a newsagent), Smiths continue to have much the same high proportion of females shopping there as do at other chains. Smiths should not, we think, reposition itself towards males.

Similarly in the car market, where strong segmentation for small Luxury marques like Mercedes and BMW undoubtedly occurs, BMW actually gains more sales each year from the dissimilar Renaults in France or Fords in Britain than from the “closely clustered” Mercedes. That is simply because Renault and Ford, as local market leaders, are so much bigger than Mercedes (Ehrenberg and Bound 2000).

It is often suggested that advertising can help to create segmentation by differentiating your brand from functionally similar competitors and “adding values” (e.g. see Young 1926/1963; Porter 1985; Broadbent 1990; Dickson and Ginter 1987; Perriss 1999). But, especially the leading brands in a category – which typically are most heavily advertised – do not in our experience, attract different kinds of customers from each other. Hence advertising does not seem to work in the way that is intended, i.e. to add effectively differentiating values (e.g. see Ehrenberg et al 2000). In line with that, when advertisers risk millions of dollars in considering a change of agency, the major criterion is usually the agencies’ creative style rather than their motivational logic or strategically targeting and particular segment (Moran 1990).

The lack of a unique brand-user profile does not in any way mean that brand marketers (or top management) can give up on marketing, but the opposite. Marketers still need to publicise and sell their brand, make it memorable and look and sound interesting, refresh brand associations, sustain its quality and availability, deal well with complaints, and generally keep the brand salient with purchasers of the category. Advertising, and marketing activities more generally, can help a brand to stand out and maintain some sense of interest or even excitement, at least among the professionals who are actually marketing the brand.

APPENDIX

We have implied that there is little if any scope for the complex but popular multivariate analysis techniques that are quite often used in segmentation studies. In contrast, the tables in this paper were set out sufficiently well to let both the analyst and the reader use straight “eye-balling” to the patterns in the data, and any exceptions. It can even allow us to communicate the results and their logic to our clients.

To elaborate briefly on the relevant technicalities of data presentation, Table 2A shows the earlier data in a more traditional format. The patterns are much less apparent than before, and no clear exceptions stand out (except perhaps 3619 as the biggest number). The task of ignoring the decimals when reading down each column in Table 2A is for example visually quite onerous.

In contrast, the earlier Table 2 made it clear to the naked eye that the individual figures in each column differed little from their average. Hence we can first note and then remember that there are few differences in the profiles from brand to brand. This was easy to see, especially once one had been told what story-line in fact to look for. (The detailed variations in the numbers were brought out yet more explicitly in Tables 3 and 4, and summarised in Table 5.)

The process which can make such data more user-friendly is at times referred to as “Data Reduction” (e.g. Ehrenberg 1982 and 1994). This turns data into information and involves steps such as:

1. Rounding

The guideline is to round to just 2 effective digits. This helps one first t o perceive and then even to remember the numbers better. Here we have often used deliberate over-rounding to just one digit, since more precise quantities in Tables 2 to 5 would not matter (e.g. whether two profile percentages in Table 2 are 12.8% or 13.4%, or both just 13%. Or whether the deviations in Tables 3 and 4 are 1.2 and 4.3 or just the much simpler 1 and 4).

2. Ordering by size

Rearranging the rows of a table by some measure of size (such as the numbers of Users or market-shares) allows one to see visual correlations (i.e. high in one column tending to go with high in another column or not, plus isolated exceptions).

3. Averages

These provide both a summary and a visual focus. (One can readily see that the Male %s in Table 2 are all about the same, i.e. close to 54%, the average. That is easier than comparing all the individual percentages with each other – i.e. the first with the second, the first with the third, then the second with the third, the first with the last, and so on – and remembering the results).

4. Layout

This should be used to guide the eye, e.g. using white space, occasional rules, somewhat varying type-faces, and so on.

Footnotes:

¹ Data for segmenting specific markets need never be truly confidential – anybody can ask a sample of consumers some appropriate questions. In being able to avoid the cost of doing so – with BMRB supplying us with the TGI profiles – we have nonetheless maintained a degree of confidentiality by not citing the dates of data.

² A complication in quite a few cases is that a differentiated product-variant may in fact have a singular brand name (e.g. Kellogg’s All Bran, usually because the item was not large enough to attract a me-too or two, or because of patents with pharmaceuticals).