There is a Pareto Law - but not as you know it | Website for Corporate Sponsors

Abstract

If 80% of your buyers contributed barely 20% of your sales then it might make sense to ignore these people. But if that same 80% contributed almost half your sales…?

Pareto share is a measure of the concentration in purchase weights. It is usually expressed as the proportion of sales that come from the brand’s top 20% heaviest buyers. When buyers all buy at similar rates then concentration is low (reaching a minimum of 20% of buyers accounting for 20% of sales or 20:20). When purchase frequency varies between a few heavier buyers and many very light buyers then concentration is high (reaching a maximum of 99:20).

In marketing, the Pareto law (or 80:20 rule) is widely interpreted as 80% of sales and/or profits coming from the top 20% of a brand’s buyers.

We give important guidelines for measuring Pareto share and address two key questions:

1) Is there really a Pareto law? The answer is yes.
2) Is it correct to represent it as 80:20? The answer is no.

We report brands’ Pareto shares across 17 categories. We find that the most regular 20% of customers account for around 60% of purchase occasions, not the 80+% widely assumed. Category buying, follows the same pattern, i.e. the top 20% of buyers account for about 60% of category sales.

The law-like nature of this Pareto pattern means the concentration of the customer base is not a consequence of marketing strategy. It’s not something you can, or should, seek to change.

The fact that the Pareto share is less extreme than usually portrayed raises serious concerns about marketing strategies that are highly skewed towards heavy buyers.

MARKETING’S PARETO LAW

Pareto share analysis is based on the work of an Italian-French economist – named Vilfredo Pareto (1848-1923), who studied concentration in wealth distribution. He is commonly quoted as finding that just 20% of the population owned 80% of the land (though he himself never referred to “80:20” *1. Just how this idea got into the marketing canon is unknown. In marketing, Pareto share typically refers to the concentration of sales, that is, the proportion coming from the heaviest 20% of a brand’s buyers. The Pareto law that 80% of sales come from 20% of a brand’s customer base is well known. Textbooks generally teach this principle without any accompanying empirical data (Baker 2000 is an exception), but rather as a proven and widely known law. It is commonly (mis) used to justify highly targeted marketing strategies (e.g. Koch 1999).

*1. Nor did he suggest that it was a law-like pattern generalising beyond wealth distribution.

Quite incredibly, given its status in marketing lore, it is not known if the Pareto pattern is law-like. It’s not known how far it extends, and where it does not apply. Nor is it known if 80:20 is a fair representation of the pattern, if it does exist.

The few papers that do report empirical analysis have found that a category’s Pareto Share is more typically around 50% – the top 20% of consumers account for around 50% of category sales, with the other half of sales accounted for by the remaining 80% of consumers (Schmittlein et al. 1993; Twedt 1964) *2. Even less is known about brands’ Pareto shares. That the typical Pareto Share might be less extreme than usually portrayed has major implications for how customer-marketing costs should be distributed across a brand’s customer base.

*2. Such “heavy half” patterns led to Gerald Goodhardt’s 20:30:50 rule which states that the 20% heaviest buyers account for 50% of purchases, the 50% lightest buyers account for 20% of purchases, and so the middle 30% of buyers account for 30% of purchases. i.e. 20:30:50 buyers accounting for 50:30:20 purchases.

Hence the importance of our investigation. There is both scientific and managerial benefit in knowing if there is indeed a law-like pattern, and under what conditions Pareto share varies. Managers need benchmarks so they know how much variation from the 80:20 rule is truly unusual, and if the strategic implications that flow from a Pareto share in the vicinity of 80% should be applied. Academics need to know whether or not they should really be teaching the Pareto law as a law, at least insofar as it applies to sales.

In our analyses of the Pareto share metric for brand sales we found a stark, consistent pattern (empirical generalisation); there is indeed a Pareto law in marketing. But it is wrong to say it is as extreme as 80:20. We also find that there are known conditions under which it varies due simply to measurement/ analysis issues – and you need to know of these if you are calculating or comparing Pareto shares.

CALCULATING THE PARETO SHARE METRIC

To date, Schmittlein, Cooper and Morrison (1993) have made the greatest contribution to scientific knowledge concerning Pareto metrics in marketing. They show that meaningful interpretation depends on understanding the effect of time and how the analyst considers non-buyers in calculation *3. They show that the latent, or ‘true’ Pareto can vary a great deal from the observed because of analytical decisions.

*3. This is actually a simplification (even if it isn’t that simple). Technically oriented readers are urged to read Schmittlein, David C., Lee G. Cooper, and Donald G. Morrison (1993), “Truth in Concentration in the Land of (80/20) Laws,” Marketing Science, 12 (2).. Others can be reassured that in practice the other technicalities don’t appear to matter too much in determining a brand’s ‘true’ Pareto share.

They point out that including people who don’t buy the category at all (and therefore contribute nothing to sales volume) increases the concentration. It tends to make the ‘heavy buyers’ look like they contribute a great deal in comparison to these ultra light (i.e. non) buyers.

So to validly compare the concentration of different customer bases only those who purchased in the time frame should be included.

Similarly when comparing brands, if all category buyers are used, not current buyers of the brand, then brands with less than 100% penetration during the period will show more extreme Pareto shares. The few buyers of the brand look very heavy indeed compared to the majority who do not buy it. Indeed a brand with less than 20% penetration must have a (nonsensical) Pareto concentration of 100:20 *4 if non-buyers are included. It is perhaps this sort of Pareto calculation that has previously led analysts astray.

*4. It is a common mistake to think that Pareto share expressions must sum to 100%, e.g. 30:70, 20:80, 10:90.

When calculating Pareto shares for brand buyers there is another complication. This is the influence of the timeframe, and its corollary, purchase frequency rate and the chance that light buyers will make their (typically) one purchase. The more time/purchase occasions, the more light buyers will be included in the sales figures. This then increases Pareto share as it enhances the difference between the greater number of light buyers and the greater purchases of heavy buyers.

To illustrate this, consider that in a single week all the gasoline buyers who have bought are likely to have done so at a very similar rate, i.e. almost every buyer has bought once, maybe a few have bought twice. So the top 20% of buyers can account for little more than 20% of sales volume (i.e. 20:20). But over the longer time period of three months, heavy buyers will have bought many times (more than 24 times) whereas many light buyers who drive infrequently will have only bought far less often. After one week of sales the Pareto share was 20:20; but after three months of sales this may change to 40:20 (with the top 20% accounting for 40% of sales). This means that the observed Pareto Share is not a static metric, but varies, predictably, according to time.

Our analysis and others (Rungie et al. 2002; Schmittlein et al. 1993) suggests that for many packaged goods categories an analysis period of one year approximates the long-run Pareto share.

Another analyst decision that could impact on Pareto metrics is whether to measure sales in terms of purchase occasions, units sold, or volume sold. In this report we stick with purchase occasions in line with the previous empirical work on Pareto share. It’s not obvious how the different measures might affect Pareto metrics. More regular buyers might also buy more or larger packs thereby increasing polarisation between them and lighter buyers. But the reality of different pack sizes might mean that some households elect to buy more often and some less often, so purchase frequencies show more variation. To investigate we conducted several additional analyses using volume and units, these showed ever so slightly higher Pareto shares. So we reached the same conclusion as Schmittlein et al (1993) that the different measures show the same picture.

A search of a wide variety of popular marketing textbooks reveals that very few actually cover the Pareto Share calculation and its implications for interpretation, i.e. they ignore these complexities of measurement and analysis. Therefore, it is perhaps not surprising that the Pareto law is often over simplified and misrepresented even by those with access to the data.

METHOD & RESULTS

We used extensive data from the TNS Superpanel and calculated the percentage of volume accounted for by the heaviest 20% of buyers (in terms of purchase occasions) across 17 different product categories. This was first done at category level, where all buyers were considered together regardless of the brands bought. Then it was conducted for each brand individually. For all analyses non-buyers (of either brand or category) were excluded.

Time frames of three months and 12 months are shown here to illustrate the effect of time extension on the observed metrics.

Our results are purely empirical, unlike some previous work on Pareto shares which calculate them from modeled data (e.g. Habel et al. 2003; Schmittlein et al. 1993). This is important as we are trying to establish if there is an empirical law, but it also means our results, particularly for smaller share brands, are subject to a degree of random sampling variation (i.e. some of the reported difference between brands will be illusory).

Results: Pareto Share for Categories

First we report the Pareto share metric for entire categories. This includes anyone who bought any brand.

If a category has a more extreme Pareto share, then marketers with a brand in the category can more easily justify targeted acquisition efforts, such as target marketing towards heavier households, e.g. credit accounts, volume discounts, targeted media etc.

We found that the Pareto share across the 17 categories averaged 51% in three months and slightly more concentrated at 58% after 12 months (Table 1). This is much less concentrated than the proverbial 80:20.

Table 1: Pareto Share Across 17 Categories

Category Averages	Avg. Purchase Frequency pa	% Volume Accounted for by Heaviest 20% of Buyers
		3 months	12 months
Dog food	41	55	60
Carbonated Soft Drinks	37	53	56
Yoghurt	30	52	54
Chocolate	29	52	54
RTE Cereals	21	48	50
Fruit Carbonates	17	55	61
Fresh & Wet Soup	13	53	56
Ice Cream	12	52	55
Washing Powders	8	44	48
Fruit Drinks	8	56	67
Spirits	7	59	70
Body Sprays & Deod.	6	47	53
Carbonated Water	6	54	69
Sauces & Ketchup	6	44	50
Still Water	6	59	70
Analgesics	5	48	58
Hair Conditioner	4	43	53
Average		51	58

Category Pareto Share is about 60%

The Pareto share metric was relatively similar across categories, despite wide variation in product types and category purchase rates. The difference between three and 12 month Pareto share is greater for categories with low purchase frequency. This is because of the greater opportunity for light buyers to buy and therefore be included as buyers in the time period. However overall the change in Pareto share is quite small given the large extension in time.

Results: Pareto Share for Brands

We now report the same Pareto share analysis within the respective buyer bases of each of the 20 highest share brands in each of the categories reported in Table 1.

An extreme Pareto share metric for a brand means that it has a small group of buyers who represent a lot of sales (and a large group who deliver few sales). This justifies the use of costly marketing efforts to retain the brand’s more valuable consumers. While the more even the distribution of buying frequencies (lower Pareto share) is, the more the challenge becomes to reach out to the wider customer base cost effectively.

We examined how brands vary within a market, as well as over time. Table 2 provides an example of average Pareto share metrics for brands within the category of body sprays & deodorants. Again note that the observed Pareto shares consistently increase as the time period extends.

Table 2: Pareto Share for Brands in the Body Spray and Deodorant Category

Brand	Market Share (%)	% Volume Accounted for by Heaviest 20% of Buyers
		3 months	12 months
Sure	16	42	53
Lynx	14	41	53
All Others	14	40	51
Impulse	8	45	55
Soft & Gentle	7	39	52
Rightguard	7	39	51
Dove	6	36	48
Tesco	4	43	53
Vaseline	3	39	51
Asda	3	42	54
Adidas	3	35	45
Gillette Series	3	39	50
Other Brands	3	39	49
Physio Sport	2	37	50
Sanex	2	35	45
Nivea	2	36	46
Wilkinsons	1	41	54
Arrid	1	37	45
Natural Plus	1	35	47
Mum	1	35	46
Average		39	50

Brands within a category have a similar Pareto share

These metrics are subject to random sampling variation, especially for the smaller brands (see Rungie et al. 2002). In spite of this we find that brand Pareto shares tend to be similar across all brands within a market, with Mean Absolute Deviations (MADs) ranging from 2 to 7 percentage points. Average brand Pareto shares are reported by category in Table 3.

Table 3: Average Brand Pareto Shares by Category

Brands in the Category of:	Avg. Purchase Frequency	Average 3 Month Brand Pareto Share	Average 12 Month Brand Pareto Share
Dog food	41	55	68
Carbonated Soft Drinks	37	54	67
Yoghurt	30	55	66
Chocolate	29	49	60
RTE Cereals	21	51	61
Fruit Carbonates	17	50	64
Fresh & Wet Soup	13	48	58
Ice Cream	12	45	55
Washing Powders	8	44	59
Fruit Drinks	8	43	56
Spirits	7	50	62
Body Sprays & Deod.	6	39	50
Carbonated Water	6	47	64
Sauces & Ketchup	6	60	68
Still Water	6	38	51
Analgesics	5	42	54
Hair Conditioner	4	34	44
Average		47	59

Brands have a long run Pareto share around 60:20

Table 4: Average Pareto Share of Brand Types

Category	Private Labels	National Brands
Still Water	69	71
Dog Food	68	67
Carbonates with Lemonade	67	66
Yoghurt	69	66
Fruit Carbonates	64	64
Carbonated Water	64	63
Spirits	66	57
RTE Cereals	62	58
Chocolate Confectionery	62	60
Fruit Drinks	55	61
Fresh & Wet Soup	58	60
Automatic Washing Powders	55	57
Ice Cream	58	52
Analgesics	54	54
Sauces & Ketchup	54	47
Body Sprays & Deodorants	54	49
Hair Conditioner	46	44
Average	60	59

Different types of brands show similar Pareto shares

Results: Pareto Share for Private Label Brands

Private label brands have started to dominate many categories, and continue to grow. While national brands differ from each other in terms of marketing strategy, private label brands are a much greater contrast. These private label brands differ from national brands in that they have restricted distribution and very little product category level advertising. But they tend to be much more strongly present within a store as they typically span many different categories. We were interested to see if these markedly different strategies result in differing Pareto shares. This is a test for boundary conditions of the law.

We looked at private labels separately to see if they differed from national brands. The results show no difference (see Table 4). The Pareto share of private labels is the same as for national brands in the same category.

Results: Pareto Share for Durables and Services

The 60:20 law describes grocery product categories rather well. But what of durables, services, and B2B categories? Many of these are characterised by longer inter-purchase intervals and greater irregularity of purchase. In which case the true level of concentration can be difficult to see.

We do note that it is common in services and B2B categories to assume extreme levels of sales concentration. This may come from analysts mixing quite heterogeneous categories together (e.g. small business and consumer banking). Or inaccurate databases where defectors are still listed as current customers and therefore are included in analysis.

It’s possible that these categories have different Pareto shares but we should not immediately assume they are more extreme, they may be less, for example over our lifetimes we probably all buy similar numbers of refrigerators. More research is needed.

We have conducted one analysis for cars using five years of Australian purchasing data. The category Pareto share was only 32%. In this very infrequently purchased category even in 5 years the heavy buyers aren’t much heavier than other buyers. About 20% of buyers had bought 4 or 5 times during the period, while 70% of buyers had bought 2 or 3 times.

We have also conducted one analysis for business telecommunications which recorded an extreme Pareto share of 90% due largely to a single customer being responsible for almost 40% of revenue. In such situations where customers are very different (i.e. from the Federal Government to a small restaurant) Pareto share analysis seems of little value. It simply reveals what is already known, while the normal strategy implications of an extreme Pareto do not apply, i.e. the mass of small customers are still essential to create the phone network.

More research is needed in the area of services, not just to determine what the Pareto shares can be, but also to sensibly interpret them.

MANAGERIAL IMPLICATIONS

To summarise our findings, a correct representation of the Pareto law for sales of consumer goods is 60:20. While the observed (i.e. not necessarily ‘true’) Pareto share will depend somewhat on the time frame chosen for analysis and the purchase rate of the category.

This pattern is law-like and covers both categories and brands. This is not to say that there aren’t real differences in ‘true’ Pareto shares but they are not large.

We now discuss the implications for the allocation of marketing resources, the ability to change Pareto share, how brands grow, and using the Pareto share as a metric.

Allocation of Resources

The size of the Pareto share provides insight into how marketing resources should be distributed.

Both category level and individual brands’ Pareto shares turn out to be about 60% of sales, not the 80+% that is widely assumed. For brand’s managing their customer base it means that a focus on the top 20% will result in a large amount of a brand’s sales potential being neglected. Lighter, less frequent buyers still make an important contribution to sales, and being less frequent buyers they arguably need more reinforcement as the brand is only a small part of their lives. Therefore media and distribution that reaches the wider market cost effectively is vitally important.

For brand’s seeking to grow revenue and the size of their customer base, the category’s Pareto law means that, while some potential customers are definitely worth more than others, efforts to target the very top customers need to be tempered by the knowledge that small groups of customers, even of the of heaviest customers, still offer limited overall sales potential. There is also the problem that these heaviest customers spread their purchases over more brands. Therefore they will be more costly and difficult to recruit. See Anschuetz (2002) for a discussion of the profit dangers of focusing on recruiting heavy buyers alone.

Further, there are some very real practical problems in targeting heavier buyers.

Firstly, they are harder to identify than is usually realised (and than consultants promise). Secondly, buyers change.

Even if no real change in behaviour occurs amongst buyers there is random variation, which makes it hard to identify who the true heavy buyers are. Someone who buys three times a year, i.e. once every four months, may one year buy four times and next year twice. And, of course there is also some real change, which makes it hard to identify who the future heavy buyers are. If a very short time period is used to categorise buyers then the random fluctuations in individuals’ buying means that people will be incorrectly classified. While if a very long time period is used then real changes will mean that some people will be classed as heavy even though they recently changed to light (e.g. switched or downgraded the brand), and some will be classed as light, even though they have recently, or are about to, buy more.

The consequence of these two facts is that the future sales potential of individuals looks different from their past observed buying. Non-buyers and light buyers turn out to be heavier than you thought they were, and heavy buyers lighter than they first seemed. This is neatly illustrated in a two year analysis of a leading brand of ketchup in the US using IRI and AC Nielsen panel data (Anschuetz, 2002). Although the brand was stationary, not growing or losing sales volume, 14% of its sales came from households that did not buy it at all in the year before, i.e. people who the brand’s marketing team would have considered were non buyers of the brand. While the small group (9%) of heavy buying households delivered 34% of volume, down from the 43% they delivered in year 1. See table 5.

Table 5: Sales Volume from Different Buyer Groups One Year Later

Buyer Group	% of Sample	Buyer Frequency in Year 1	Representing Brands Sales Volumes in
			Year 1 (%)	Year 2 (%)
"non-buyers"	45	0	0	14
"light"	22	1	14	15
"moderate"	25	2-4	43	36
"heavies"	9	5+	43	34
Total	100		100	100

Over time the heavier buyers get lighter, and the non / lights get heavier. Source: (Anschuetz 2002) IRI panel data USA.

That the lightest buyers became heavier and the heavy buyers became lighter is a “regression to the mean” phenomenon that happens for all brands *5.

*5. We call this law-like pattern “the law of buyer moderation” and the degree to which it is due simply to random variation in individual’s purchase frequency is predicted by the NBD-Dirichlet theory. These Dirichlet benchmarks can then be used to detect real change – this will be discussed fully in a forthcoming Institute report.

Consequently any future targeting of heavy buyers based on past buying behaviour is likely to be somewhat off-target.

Changing Your Pareto Share

Although the Pareto rule is usually presented as an empirical law it is also regularly portrayed as being caused by marketing strategy, i.e. being within managerial control. And managers do not feel that the condition is immutable (Dubinsky and Hansen 1982) but that it can be changed to the firm’s advantage *6. The pattern is said to occur due to a lack of targeting on behalf of the company, something that can be corrected by being more selective in choosing customers (Quester et al. 2004 p. 339). Along these lines some authors advocate dropping (“firing”) very light buyers (e.g. Hise and Kratchman 1987; Kotler and Keller 2006). Other strategies include finding more heavy buyers or working to improve the value of light buyers.

*6. It is implied that having a more extreme Pareto share is better, though it is not entirely clear why this should be so.

However, our results cast serious doubt on the viability of such strategies by showing that the Pareto rule is indeed an empirical law. This makes it something that management would be better to work with rather than seek to change *7. If the Pareto share really was due to management strategies, or brand equity, or some other idiosyncratic aspect of the brand, then we would expect to see much more variation between brands. Yet we see little – even between brands with substantially different market shares or even across different categories.

*7. Engineers, rocket scientists, builders and plumbers would be laughed at if they said they were seeking to change the laws of physics rather than working within them. Though if it were a builder or plumber working on your house you probably would not think it was humorous.

Brand Growth

The difference between larger and smaller brands provides insight into how a small brand will (or will not) change, should it achieve any share growth objectives. This research includes some of the world’s largest and most successful brands in Coca Cola, Kraft, Masterfoods and Heinz, and these brands had the same concentration as brands one-tenth their size. Therefore, brands can vary 10 fold in the number of customers they have, yet this has very little bearing on the distribution of the sales within each brand’s customer base. This is despite differing levels of brand performance, and the myriad of strategies that would have been used to achieve this performance.

Brand growth then will come from increasing the size of the customer base and not the make-up. This involves focusing on all types of buyers: light, medium and heavy. See Institute Report 31 “How Brands Grow”.

Pareto Share as a Metric

To measure Pareto share and use it as a marketing metric, it is important to note that Pareto share is not static over time, but will gradually increase as purchase occasions increase. Generally it reaches a plateau for frequently purchased categories at around 12 months. We reiterate Schmittlein, Cooper and Morrison’s (1993) warning that it is easy for the uninformed analyst to calculate Pareto share in a misleading way. Therefore when comparing Pareto shares, pay attention to the time frame/purchase incidence, the unit of analysis and how non-buyers are treated.

However, with careful measurement the observed Pareto share can be compared between brands, and against the Pareto law’s 60:20 benchmark. This could be useful in detecting unusual brands. For example, we suspect that excessive use of volume-based price discounts (e.g. discounts for multi-pack buys) may distort a brand’s Pareto share. In which case the Pareto metric might be a useful indicator of problems in marketing strategy. More research on this application is needed.

FOR THE TECHNICALLY MINDED

As discussed, Pareto share reflects the concentration in purchase frequency rates of brand buyers. This is like the K parameter of the NBD distribution, a distribution that quite accurately models frequencies of buying rates. Theoretically the Pareto share varies with differing K values like this (Habel et al. 2003):

This variation in Pareto share is interesting in theory, but an analysis of K values over more than 50 categories shows that, in real world repertoire markets, brand K values are all less than 2 (Driesener 2005). So looking to the above chart we should predict all real-world Pareto shares to be about 55%, i.e. not far off what we observe.

ACKNOWLEDGEMENTS

In particular we would like to thank Andy Roberts and TNS for access to the Superpanel data, and Maria Clemente & Jia Lu for calculating the Pareto shares from the database. We would also like to thank Andrew Ehrenberg for stimulating this investigation and Melanie Ceber, John Scriven, Gerald Goodhardt and Malcolm Wright… for discussion and suggestions.

Ehrenberg-BassSponsor Website

There is a Pareto Law – but not as you know it