use of consumer panels for brand-share prediction of consumer panels for brand-share prediction 133...

J. H. PARFITT and B. J. K. COLLINS*

This article describes a mefhod of predicting the market share for newlylaunched brands and the future equilibrium share of established brands aftermajor promotional activity. The method rs dependent on the continuous purchasingdata obtainable from consumer panels and is developed from numerous casehistories from the Attwood Consumer Panel in Great Britain to illustrate the use of

the technique and its refinements.

Use of Consumer Panels for Brand-SharePrediction

After the launch or introduction of a new brand orspecial promotion of an existing brand, much can begained from knowing the nltimate outcome long be-fore it becomes apparent from observing historic dataon sales figures or brand-share trends. When the out-come is apparent in the usual way, it is often too latefor action that might have been effective earlier. Thereis great marketing value in an early warning predictionsystem. One such system is brand-share prediction us-ing continuous data from consumer panels.

The objects of this article are:1. to describe the method used to obtain a prediction

of brand share from panel data.2. to illustrate how much these predictions are likely

to be valid in practice by drawing on case historiesfrom several analyses.

3. to demonstrate some of the ways the predictiontechniques have been improved by experience inthe use of analyses.

4. to comment on some of the marketing lessonsemerging from a study of these analyses.

The data used here are from analyses of the AttwoodConsumer Panel results in Great Britain after 1960.

BASIC METHOD OF PREDICTING BRANDSHARE FROM PANEL DATA

The basic method was first described in some detailby Baum and Dennis [2]. Although several refinements

*J. H. Parfltt is research and technical director of theAttwood group of companies in Europe. B. J. K. Collins ismanager of the marketing interpretation department of Att-wood Statistics, United Kingdom.

have been made to improve the accuracy and speed theprediction date, the basic method is the same. In 1961,however, it was considered an analysis to be used onlyin studying newly launched brands. It is now realizedthat the method can be used equally effectively to studyexisting brands.

The raw data of the analysis are the continuous pur-chasing records of individual households. From thesedata, three basic components are selected:

1. The cumulative growth in the number of newbuyers of the brand or product being studied.

2. How often these new buyers buy the brand orproduct again after their first-recorded purchase.These repeat purchases of the studied brand areexpressed as a proportion of the total purchasesin the product field by these buyers; this is calledthe repeat-purchasing rate.

3. The rate of total product field purchasing of theseparticular buyers compared with the average ofall buyers in the product field, i.e., are they usu-ally heavy, light, or average buyers (by volume)in the product field. This is expressed as a buyinglevel index, with the average buying level for thefield being 1.0.

Cumulative Penetration

To illustrate, assume a brand-share prediction analy-sis covers newly launched Brand T in an establishedand comparatively static product field (for example,toilet soap). The first requirement is to persuade house-holds to try Brand T. If the advertiser cannot do this ona reasonable scale, there will be no future to predict.However, assuming reasonable retail distribution and arealistic advertising and promotion budget, gettinghousewives to try the brand is comparatively easy. The

131

Journal of Markel'mg Research,Vol. V (May 1968), 131-45

132 JOURNAL OF MARKETING RESEARCH, MAY 1968

Figure 1CUMUUTIVE PENETRATION OF BRAND

40r fstJmatedultimate level

34%

lO -

O 4 e IZ ib 20 24 2B 32 36

Weeks after launch o( 8rand T

"TOIS I buvers of toilel soap = tOO%.

analysis, here, begins with the brand's launch. As eachnew Brand T buyer is picked up from panel records,she is recorded on a cumulative penetration curve simi-lar to Figure 1. It is not necessary to wait until thetotal cumulative penetration has been completely ob-served before predictions can be made. Once the shapeof the curve is determined and a declining rate of in-crease is observed, it is possible to make a reasonableestimate of the ultimate likely penetration.

Repeat-Purchasing Rate

The ultimate success of any brand depends on thewillingness of consumers, once having tried it, to con-tinue purchasing it—normally this means to the exclu-

sion (partial or whole) of the brands they were previ-ously using. In oversimplified terms, persuading theconsumer to try a brand is a function of distribution,advertising, and promotion, but getting her to keepusing it is also a function of her acceptance of the prod-uct. The first is much more difficult to improve than thesecond. The new brand should be more than techni-cally comparable with competitive brands already onthe market. It must, for a significant number of con-sumers, be sufficiently more acceptable than the brandsalready used if they are to change their purchasinghabits. The success or failure of the product's acceptance(against the background of advertising and promotion)is expressed by the repeat-purchasing rate.

As each new Brand T buyer appears on the panelrecords, her purchasing behavior in the product fieldis isolated as a continuous record over time. This isshown in a hypothetical buying pattern, Table 1, inwhich Brands R and S represent the existing competi-tive brands.

The number of first-time buyers of Brand T accumu-lates to seven in the first three weeks of the study. Therepeat-purchasing rate is calculated from the periodafter the first purchase of Brand T, here in two-weekintervals. Tt is not an expression of calendar time becausecalculation begins for each buyer from his date of entryinto the market, i.e.. Buyer 2's first opportunity for re-peat purchasing is in Weeks 2 and 3, but Buyer 7's firstopportunity is in Weeks 4 and 5. In each of the two-weekperiods these seven buyers made ten purchases in theproduct field of which successively 60 percent, 50 per-cent, and 40 percent were repeat purchases of Brand T.This pattern of a declining repeat-purchasing rate is nor-mal because second, third, fourth, etc., purchases of anew brand still tend to be exploratory and are also often

Table 1ILLUSTRATION OF REPEAT-PURCHASING RATE CALCULATION FROM A HYPOTHETICAL EXAMPLE

Weeks 8 10 11

Buver:1

2

3

4

5

6

7

Repeat-purchasing rate

T

T

' T R

1 SL _ _ _ -.

T 1 T

T

T

T

4 7

1

I

1L

n1111

T

— — -1

^ 1

T '

T

S

610 ~

R

S

T

T

R

60%

~r111

1111

R

1

^ I

R 'L

T

S

510 ^

R

S

T

T

R

50%

1

1

1

—!

1

1

1

R

T

T

R

410

T

S

1 ^tu — _

T

S

= 4 0 %

I

' R1

1

T

R

—1

1

' R11 S

410

1

1

1

=

R

S

T

T

R

40%

USE OF CONSUMER PANELS FOR BRAND-SHARE PREDICTION 133

still a benefit from promotional activity of the launch.The critical point for the prediction of the ultimate brandshare comes when the repeat-purchasing rate begins tolevel off, as in Figure 2. At this point, considering an es-timate of the ultimate likely penetration, it is possible tocalculate what the brand share will eventually be.

Buying Level Index and Prediction Calculation

It is estimated in Figure 1 that Brand T is likely toreach approximately 34 percent of potential buyers (apotential buyer here is a buyer of toilet soaps). In Figure2 the repeat-purchasing rate for these buyers has leveledoff near the 25-percent level. With the assumption thathouseholds having tried Brand T purchase toilet soap onan average at the same rate as the average of total buyersin the market, then the predicted share for Brand T is:

estimated penetration x repeat-purchasing rate Xbuying rate index. . . . (i)

in this example this is 34% x 25% X 1.00= 8.5%.

The buying level of buyers of Brand T could be an im-portant element in the calculation because if the house-holds who tried Brand T were particularly heavy buyersof toilet soaps (maybe an index of 1.2 compared withthe average of all toilet soap buyers), or were particularlylight buyers (maybe an index of 0.8) then this would con-siderably influence the calculation of the ultimate shareof Brand T. This is illustrated as follows:

Heavy buyers of toilet soap34% X 25% X 1.20 = 10.2%

Average buyers of toilet soap34% X 25% X 1.00 ^ 8.5%

Light buyers of toilet soap34% X 25% X 0.80 = 6.8%.

Assumptions Underlying the Prediction

Two basic assumptions are implicit in the predictioncalculation:

1. The retail distribution of the new brand is uni-formly high in the area under study or, failing that,it is not substantially worse now than it is likely tobe in the foreseeable future.

2. Besides the advertising and promotional activityaccompanying and during the brand's launch (in-cluding competitors' retaliatory measures), the cir-cumstanees of the market will remain much thesame in the future as they have been during theprediction measurement.

This, of course, is asking a lot of any market in reality.Actually, observations of the relation between predictedshares and actually shares achieved over periods of upto two years suggest that the prediction will be a reliableguide to likely ultimate share in almost all cases in mar-kets where no major changes take place. If a majorchange occurs in the market after the prediction hasbeen made—such as the launch of another new brand

Figure 2

REPEAT-PURCHASING RATE FOR BRAND T

4O

3O

2O

\O

25%

1to4

5to8

9to12

13to2O

Weeks after first purchase ot Brand T :

that is successful—then the prediction will no longer bevalid because the new circumstances of the market willnot have been considered in the calculation.

HOW VALID IS BRAND-SHARE PREDICTION?

Two questions inevitably arise from the use of brand-share prediction analyses;

1. Are the predictions correct, i.e., does the brandreally settle down around the share predicted forit?

2. Even if the predictions are correct, can the predic-tion calculation be made before it is clear fromother data what will happen to the brand anyway,i.e., can the prediction be made in suflicient time tosuggest a course of action that might not be other-wise obvious from other available data such assales figures and current brand-share measure-ments?

These two questions are studied later with actual casehistories.

How does brand-share prediction work in practice?Look at an actual prediction calculation made by Baumand Dennis [2] to see how the brand progressed after theprediction had been made. The penetration and repeat-purchase patterns of the brand (which is Signal tooth-paste) are shown in Figures 3 and 4. The predicted sharefor Signal was 40 percent of 37 percent = 14.8 percent,assuming the total field-buying rate for Signal buyerswas about average. Unfortunately, since the buying levelindex refinement was introduced sometime after thisanalysis, the buying level index for this case is not known.

Now look at these: (a) at what stage in the market de-velopment for Signal the prediction was made, and (b)at what level the Signal share found equilibrium, and forhow long it remained there.


The first point is best studied by locking at the four-week shares for Signal during the launch period (Figure5). Here the position is slightly complicated since thelaunch was phased across the country over a 12-weekperiod which somewhat delays the point of final predic-tion.

Thus, the prediction was made as a first estimate asthe brand reached its peak, and as a final estimate as itbegan its inevitable decline from the peak, i.e., between16 and 20 weeks after launch.^ The predicted level wasreached between 36 and 40 weeks after launch; the prod-uct stayed at this general level (between 14 and 15 per-cent) for two years following the prediction. The launchof the lluoride toothpastes in the middle of Year 3 upsetthe market's equilibrium sufficiently to destroy the basison which the Signal share had been predicted. Thus theprediction of the Signal share was made just after the

Figure 3

CUMULATIVE PENETRATION OF SIGNAL"

Figure 4REPEAT-PURCHASING RATE OF SIGNAL BUYERS

6O

£ 5O

° 5 JO

l O

t EstimatedultimBle

I levelII« 4 0 %

1 I I J L I ] I 1 I I

O S lO IS

Weeks after tirsi purchase of SignalSource: Aiiwood Consumer Panel. Great Britain

30

2O

- lO

Estimated ultimatepenelralion

— -37%

First estimatemade here

1

Finai estimatemade here

O 5 lO 15 20 25 3O IS ^OWeeks after launch of Signal

"Total dentifrice buyers in 26 weeks -- 100%Source: Attwood Consurner Panel. Great Britain.

peak of the launch but long before it could be determinedwhere the share level would stabilize. The prediction re-mained valid for two years in a highly competitivemarket. To what extent is this a reliable guide to the useof these prediction analyses, or was it accidental?

Before study of the evidence, it should be recognizedthat a substantial proportion of brands do not reach satis-factory share levels and are withdrawn. In these easesthe brand-share prediction analysis indicates the lowultimate share, but the accuracy of the prediction can-not be proved because the brand does not stay on themarket. So, prediction analyses first indicate the likeli-hood of success or failure, and only in adjudged suc-cesses does the opportunity usually occur to assess theaccuracy of the prediction. Failure generally takes the

^The length of time required, after the start of the analysis,for a prediction to be made depends on two factors: (a) theaverage frequency of purchase in the product field, and (b) theconcentration of advertising and promotion used to obtainrapid market penetration by the brand.

form of exceptionally low repeat-purchasing rates, i.e.,the brand makes a reasonable penetration into the mar-ket but very few people continue to purchase it. Figures6 and 7 give actual examples from a toiletry product.

Thus, a penetration of 17 percent of the market,

Figure 5

SIGNAL'S SHARE OF TOTAL DENTIFRICE MARKET INTHE FIRST 11 FOUR-WEEK PERIODS AFTER LAUNCH

Firstestimate

Signalshare

FinalPrediction

18

15 14

12 3fc 4O16 ZO 24 28 32

Weeks after launch

Source. Attwood Consumer Panel. Great Bricain


Figure 6

CUMULATIVE PENETRATION OF BRAND Y"

Estimatedultimate level

17%

O 4 8 12 Ib 2O 24 2&

Weeks after Brand Y launch

" Toial buveis in the product field 100%.

Source: Attwood Consumer Panel. Great Britain.

though not all that healthy, could nonetheless form thebasis for a viable brand share. If, for instance, the re-peat-purchasing rate had been 35 percent (less than thatfor Signal) on a penetration of 17 percent, this wouldproduce a predicted share of around 6 percent. In thisparticular field, already well-fragmented with competi-tive brands, this could be regarded as a reasonably suc-cessful launch. In fact, the repeat-purchasing rate sta-bilized at the very low position of 6 percent, and 6percent of 17 percent produces a predicted share of 1percent—far too low to be viable for a nationally pro-moted brand in this field (and too low for need to applythe buying level index). Clearly the fault lies with therepeat-purchasing rate; buyers having tried the branddid not feel disposed to use it again. In other words, theproduct did not gain acceptance.

So, prediction of failure has to be taken at its facevalue, but prediction of success can be studied furtheron accuracy—by studying what actually happened tothe brand share after the launch. The results of some24 successful ones where share predictions were madeand the field was measured on the Attwood panel forthe following 18 months are shown in Figure 8.

The critical period, representing that for which theshare prediction was to apply, is the six-month periodbetween 12 and 18 months after the launch. Thus thepredicted brand share (single-valued) is compared withthe range of actual brand shares achieved (based on 6 X4 weekly periods). The diagonal line on the chart passesthrough all points where predicted and actual brandshares coincide. The vertical bars show the range ofactual four-week brand shares achieved between 12 and18 months after the launch. Thus the extreme top right-hand example on the chart had a predicted share of 24.0percent (immediately afterward) and achieved actualshares in the period 12-18 months afterwards ranging

from a maximum of 25.2 percent to a minimum of 22.0percent. The fields represented in this summary of pre-diction analyses cover products like dishwashing liquid,toilet tissue, butter, instant coffee, and floor polish, andinclude predictions in static total markets and in verydynamic total markets.

Of course, actual brand shares are seldom completelystatic—particularly in product fields with considerablepromotional activity and particularly on a measurementas frequent as a four-week one (shares that can be quitestatic on an annual average basis will often show con-siderable fiuctuations on a four-week basis). The shareprediction, therefore, although it is expressed as a singievalue, really means the brand share will eventuallyassume this level as an average. All estimates of ultimatelevel of brand share are subject to the necessary condi-tion that the market in which the new product is launchedremains relatively stable in the period after the product'sintroduction. Figure 8 shows these predictions can beextremely accurate, particularly in relatively stable mar-kets.

SOME REFINEMENTS OF THE TECHNIQUE

Earlier Prediction of Estimated Ultimate Level of Pene-tration

It is possible to use a marketing model of penetrationthat allows an early prediction of the expected ultimatelevel. Details of the model's statistical basis and an in-dication of the method of fitting the model to the actualdata and the accuracy obtained, are given in the ap-pendix.

The model's application enables an estimate of ulti-mate penetration to be made before the curve shows anymarked tendency to level off. This, in itself, is valuable

Figure 7

REPEAT-PURCHASING RATE OF BRAND Y BUYERS

20r-

•s 10

1lo

4

Source

5 9to to

8 12

Weeks after first purchase

13lo

20

Aitwood Consumer Panel. Great Britain


Figure 8COMPARISON BETWEEN PREDICTED AND

ACTUAL BRAND SHARES"

Line of roinddence bslweenpredicted and actual shares

25 r*-

Perceritaqe of O'Srtrt^ted brand share

"Based on six four week reports one yeat later.

Source Allwood Consumer Panel. Greal Britain.

in planning marketing strategy. It is also helpful in re-ducing the length of the minimum study period necessaryto produce a first estimate of expected brand share. Theselected minimum time period is thus dependent on thevalidity of the secondary marketing variable, the repeat-purchasing rate. This measure is dependent on the fre-quency of purchase of the field containing the launchedbrand and may thus be determined before the studybegins, assuming a reasonably smooth launch for thenew brand.

In many instances the penetration curve has alreadybegun to level off at the end of the selected minimumtime period even though it has then been found that thededuced ultimate level based on the fitted mode! may beat variance to that produced by free-hand extrapolation.Some examples of actual penetration curves are givenin the appendix to illustrate this point. However, thegreatest advantage in using the penetration model is thatthe projected estimates provide an expected growth ofnew buyers with which it is possible to compare the sub-sequent actual penetration. The differences may then beattributed to such marketing disturbances (extensiveprice-cutting, free sampling, couponing, increased ad-vertising) as are evident, in the form of a quantitativemeasure of additional penetrations and, after furtheranalyses, additional brand share.

Allow for the Launch of a New Product/Field

Occasionally a new product cannot be considered aconstituent part of an established field. Then, it is ob-

viously not possible to apply the brand-share technique.However, the repeat-purchasing rate can be redefinedin terms of quantity or expenditure per four-week periodfor a buying household and thus estimate an ultimatesales level. This calculation is, of course, possible in allprediction analyses, but it serves no advantage over es-timations by brand share and, therefore, is only usedwhen the product under study has no clearly definedmarket on which to base share.

Repeat-Purchasing Rates at Different Stages of Entry

A prolonged study of repeat-purchasing rate behaviorhas led to the conclusion that on average the sooner abuyer enters the market for a particular brand, thehigher will be that buyer's repeat-purchasing rate. Thatis, the early entrants to a new brand's market tend tohave disproportionate importance in contributing to theultimate share that brand will achieve. For example,Brand B, an edible fat brand, is launehed in one com-mercial TV area. It was already available in the areabefore its official launch because the distribution toother areas overlapped into it. Six months after launch,its estimated penetration in the market was 30 percent,the repeat-purchasing rate had leveled at 15.5 percent,and the buying rate index of Brand B buyers was 1.05,producing an estimated brand share of 4.9 percent.When this analysis is done separately by the buyers' dateof entry into the market, the ultimate share predictionhas sharply contrasting elements (see Figure 9 and Table2). This conclusion has important marketing connota-tions to be considered later.

Thus, the longer after the launch date a buyer enteredthe market for Brand B for the first time, the lower theaverage repeat-purchasing rate and the smaller therelative contribution to the ultimate brand share. This

Figure 9

CUMUUTIVE PENETRATION OF BRAND B ANALYZED BY

BUYERS' DATE OF ENTRY INTO MARKET FOR BRAND B"

J O r

tn 2O -

9 l O -

Cumulative oenotratian of Brand B

13 to £4 weeks later

125 6%1 /

Estimaied ultimate penstration 30%

Existing buvers before iniroducdon 11 9%)

0 2 A b 8 10 17 M It) IS ZO 21 24 26 2& 3O

Weeks after launch"Total buvefs in one Btoduct field = 100%.

Source: Attwood Consumer Panel, Greai amain.


T a b l e 2

REPEAT-PURCHASING RATE FOR BRAND B ANALYZED BY DATE OF FIRST ENTRY INTO MARKET

Buying group by dale of entry

1, Existing buyers before Brand B wasolTicially launched

2. First-time buyers in the first 6 weeks3. First-time buyers in the second 6

weeks4. First-time buyers 13-24 week period5. Estimated first-time buyers after 24

Total {or average)

Contribution to cumula-tive penetration of

Brand B by each groupof buyers

1.9%

13.25.0

5.5(4.4)

30.0%

Repeat-pu rchasingrate of eachbuying group

40%

1715.5

«.5(8.5)

15.5%

Buying

1

11

1(1

1

rale inde.\

.21

.00

.12

.08

.00)

.05

Contribution of eachgroup to estimated

brand share

0.9%

2.20.9

0.50.4

4.9%

Source: Attwood Consumer Panel, Great Britain.

pattern is consistent in all these analyses, and analyzingthe repeat-purchasing rate by the point of entry on thecumulative penetration curve is now standard practice.It has three considerable advantages:

1. Greater accuracy is obtained in the ultimatebrand-share prediction calculation because therepeat-purchasing rate of new buyers estimatedstill to enter the market can be determined at thelatest marginal rate rather than the average rate(that would tend to overestimate).

2. Knowledge of the repeat-purchasing rate of themost recent entrants to the market can indicatewhether there is any point in trying to make specialefforts to increase the brand's cumulative pene-tration.

3. It contributes much to the understanding of factorsthat determine the brand's share and to marketingactivities aimed at changing that share.

Total Market Cumulative Buying Levels to DetermineShare Predictions in Changing Markets

An assumption not yet mentioned in relation tobrand-share prediction calculations is that the totalmarket level is not affected by the launch of the newbrand. For a wide range of product fields that assump-tion is substantially true, but for some it is not, and inthese cases, it could seriously affect the accuracy of aprediction calculation if it is not considered. Highlyseasonal markets are special and require the sameanalytical techniques for correction.

To allow for these possibilities, the accumulation ofboth total buyers of the product field and buyers of thestudied brand is produced. It is only when the rate ofincrease in accumulation of buyers for the brand ex-ceeds the corresponding rate for the total market thatthe brand's marginal penetration is contributing any-thing to its ultimate level of penetration. In Figure 10,although the penetration of the brand appears to surgesuddenly upwards, it is only the result of seasonal fac-tors. The true penetration does not alter at all.

Cumulation of Case History Data Enables QuickerAssessment

One result of many brand-prediction analyses in thelast six years is that something approaching a set ofrules begins to emerge from the data. From these it ispossible to determine relatively early whether the pat-tern of the brand being measured shows elements ofsuccess or failure. They are not rigid rules but haveconsiderable value in two ways:

1. A relatively small cumulative penetration ac-companied by a relatively high repeat-purchasingrate might suggest that special promotional ac-

Figure 10ISOLATION OF SEASONAL FACTORS IN CUMUUTIVE

PENETRATION OF A BRAND (BRAND J)"

£ 30 -

£ 2O -

10 -

-

Total fi

\

Stan of high

eld ^ ^ " ' ^

y

Brand J

{40%) (42%)

season

(42%|

\

(42%)

\

(42%) (41%)

/

r —

(41%)

16 2O 24 28

Weeks after launch

"All households = 100%.

Source Attwood Consumer Panel. Great Britain.


Figure 11RELATIONSHIP OF REPEAT-PURCHASING RATES

TO PENETRATION LEVELS IN 31 CASE STUDIES

\ . The line ofcoincidencebetweenpeneirationand repeat-puichasing

. arale.

O IO 2O 3O 40 5O bOEstimaied uKimate repeat purchasing rate"

"Multiplied by ifie buying rate lacior.Source: Attwood Consumer Panel. Great Sriiam

tivity or improved distribution could make a sub-stantial improvement in ultimate brand share.

2. The exact relationship between a relatively highaccumulation of buyers and a low repeat-purchas-ing rate can help explain whether a product de-ficiency or a more flexible factor is at fault.

One thing is certain—-there is no rule about the levelof repeat purchasing to be expected at ditferent levels ofpenetration. There is no natural equilibrium level ofrepeat purchasing associated with particular levels ofpenetration, shown in Figure 11 where several examplesof the relationship of repeat-purchasing rates to pene-tration levels are plotted, with eontour lines of constantbrand share.

Applying Brand-Share Prediction Analysis to Estab-lished Brands

It was assumed, when these analyses were first de-vised., that they would apply only to measuring progressof a new brand from its launch; for a long time theanalyses were used only for this purpose, probably fortwo reasons:

1. Conditions existing when the brand was launchedfor new buyer accumulation and repeat purchasingare somewhat different from those applying to anestablished brand.

2. Even if the above is not true, established brandsdo not have the dramatic changes accompanyingthe launch of a new brand, and there would,therefore, be nothing for a share-prediction analy-sis to do.

It has since been observed that movements belowthe surface of an established brand differ only in degreerather than kind from those present at the launch of anew brand. Share-prediction analyses have been of con-

siderable use, for example, accompanying major pro-motions of existing brands to determine how muchultimate effect they are going to have on brand share.Some examples are shown later. They are now becomingimportant in the measurement of loyalty patterns forestablished brands—since, of course, the repeat-pur-chasing rate is a way of expressing loyalty. The onlytechnical difference in analysis for an established brand,compared with a new brand, is that it begins at a fairlyarbitrary point in time (although its selection is usuallyrelated to the start of a promotion) and the accumulationof buyers is based only on buyers picked up after thattime. From then it is exactly like an analysis for a newbrand, even to the point when the repeat-purchasingrate is highest for buyers who come in first (here, ofcourse, buyers who come in first will include a highproportion of regular and loyal buyers).

MARKETING FACTORS

The purpose of test marketing is to try to determinethe likely success or failure of a product before it iscommitted to a national launch. Using a brand-shareprediction analysis in a test marketing situation willprovide certain advantages:

1. It will determine within basic preset limits whetherthe brand is likely to succeed or fail.

2. If success is predicted, it will determine, quite ac-curately, what degree of success expressed inbrand-share terms is likely to be achieved. Thishas obvious advantages in production, advertising,and promotion planning.

3. The prediction will be made long before the sameconclusions could be drawn from sales or standardresearch data. Since timing is a considerable fac-

Figure 12INFLUENCE OF AN OFFER TO RETAILERS

ON CUMULATIVE PENETRATION OF FOOD BRAND C"

EstimatedLiKimate levels

36%

5 2O

10

Price cutting slarlsIntroduction ofnew flavor.

25%

20%

0 I 2 3 4 5 i> 7 S 9 IO I I 12 I ] 14 IS Ib 17

Four-weak periods after tbe launch

•^Toral buyers 'n the product nelO 100°<i.

Source Aitwood Consumer Panel Great Bmain.


Figure 13INFLUENCE OF A 50 PERCENT PRICE CUT O N

CUMUUTIVE PENETRATION OF DETERGENT BRAND X"

3O

2O

IO

Eslimatedultimale levels

• 3 1 %

Introduclion of50% off offe'

/ 20%

Projections

- 2 - I O I 2 3

Four-week periods from ^tarl of offer"Toial buyets in the product field =- 100%.

Source Aitwood Consumer Panel, Great Britain.

tor in the value of a test marketing operation, earlyprediction of results is clearly important.

However, certain marketing conclusions emergingfrom a study of these analyses go beyond the basic con-clusions of the analysis itself. That is, the later a buyerenters the market for the studied brand, the lower herrepeat-purchasing rate is likely to be. The importanceof this conclusion to marketing is considerable, and the

range of its application must be exploited. To illustratethis point, let us make an unreasonable statement—atleast the statement is unreasonable in its simple form—although it is often an implicit assumption in certainmarketing decisions made by marketing men who wouldrecognize its unreasonableness in the simple form:

"My brand has achieved a 20 percent penetra-tion of the market, with a 30 percent repeat-purchasing rate. This will give me an ultimateshare of 6 percent of the market. If I coulddouble my penetration of the market, I wouldpush my brand share up to 12 percent."

The implicit assumption, is that an increase in pene-tration can be achieved without loss in repeat-purchas-ing rate. The only possible exception follows. If therepeat-purchasing rate were kept artificially low by ablockage in distribution, and if the distribution werethen substantially improved, then the repeat-purchasingrate might remain constant when the penetrationdoubled. In fact, assuming distribution remains fairlyconstant, the effect of doubling the penetration in theexample could be anything from hardly any increase inultimate share (maybe from 6 percent to 6.5 percent) toa substantial increase (unlikely to exceed a movementfrom 6 percent to 9 percent). Which result it is seemsto depend considerably on the circumstances of thelaunch to date and the methods used to improve thecumulative penetration.

A more sensible statement is "it is comparativelyeasy, within limits, to influence cumulative penetration,i.e., the number of first-time buyers, but it is extremely

Table 3ESTIMATES OF ULTIMATE MARKET SHARES LIKELY TO BE ACHIEVED BY BRANDS C AND X, BOTH

WITH AND WITHOUT PRICE CUTTING

Cafegory

Berore introduction of price cuttingbefore new flavorafter new flavorexpected further buyers

TotalAfter introduction of price cutting

before offerafter offerexpected further buyers

Total

Before offer's introductionAfter offer's introduction

buyers without offerbuyers wilh offer

Ultimate penelralion X

15%55

25%

20%133

36%

20%

2011

Repeat purchasing =

Brand C

20%1818

19%

19.5%1010

15.3%Brand X̂*

25%

256

Ultimate brand share

3.0%0,90.9

4.8%

3.9%1.30,3

5.5%

5.15%

5.150.655.8%

Buying rate index

1.03

1.031.01

" Apparent effect of offer on ultimate brand share = 0.7% (or 15% increase).'• Apparent effect of offer on ultimate brand share = 0.65% (or 12i% increase).<= Share calculation allowing for buying rate index.Source: Attwood Consumer Panel, Great Britain,

uo JOURNAL OF MARKETING RESEARCH, MAY 1968

Figure 14INFLUENCE OF PROMOTION

ON PENETRATION OF BRAND G "

40

3O

20

Estimateduliimate levels

Clientreceivedfifsi

_ projections

Start ofcampaign

- - 2 7

O I 2 3 4 b 7 a IO I I 12 13 14 15 16

Four week periods after launch"Toial buyers in the product held 100%.

Source: Attwood Consumer Panel. Great Britain.

difficult to create or infiuence repeat purchasing for anylength of time." The time for influencing repeat purchas-ing is at the product development stage. By the time itreaches the market, there is only very limited room tomaneuvrc. particularly if the product has comparativelylow acceptance among new consumers compared withother competitive products. Some examples follow.Deep Price Culling as a Means to Improve Penetration:

(a) a new brand. Brand C, using attractive retailer-trade terms to reduce the consumer price, at a timewhen the penetration curve was leveling oft, and(b) an established brand. Brand X, using a 50 percentprice-cut offer, also when the penetration curve wasleveling otf.

In each case there is a sharp increase in penetration as-sociated with price cutting, running against the normaldirection of a penetration eurve (shown in Figures 12and 13). The net increases in penetration associatedwith the otfer were:

Brand C—11% (43% increase),Brand X—11% (55% increase).

The repeat-purchasing rates of buyers brought inbefore, and after, the price cuts, respectively, appear inTable 3. In both cases the price cutting considerably in-creased the number of first-time buyers but added pro-portionately much less to the ultimate brand share. Thereason is that new buyers gained thus showed relativelylittle inclination to repeat purchase. In effect, it appearsthat they were only prepared to be in the market at thesubstantially reduced price. When the price movedback towards normal, they dropped out. Particularlywith Brand X this is a clear, if extreme, example of theease with which penetration can be influenced but notrepeat purchasing.

Other Promotions to Improve Penetration: (a) a newbrand, Brand G, and (b) an established brand, BrandH.

In both, special promotions were used when the pene-tration curve for the brand was leveling ofl\ and in both,the promotions produced a substantial and unseasonalincrease in penetration (see Figures 14 and 15). Thenet increases in penetration associated with the promo-tions were:

Brando—12% (44% increase),Brand H—35% (117% increase).

The repeat-purchasing rates of buyers brought in beforeand after the promotions, respectively, are shown inTable 4. In these examples, the promotions had beensuccessful in adding new buyers but, to a greater extentthan the price-cutting examples; they had also beensuccessful in achieving a healthy repeat-purchasing rateamong these new buyers, and therefore, made significantincreases in the ultimate brand share.

The purpose of these two sets of examples is not todemonstrate that substantial price cutting will do verylittle to ultimate brand share, and other promotionalactivity will be more successful. This may be true, par-ticularly in product fields with a high elasticity of de-mand related to price, but there are insufficient ex-amples here to prove it. What the examples do show isthat it is much easier to increase penetration than toimprove ultimate brand shares and that the extent towhich a brand-share improvement follows from an in-crease in penetration appears to depend on the methodsused to increase penetration. The examples also demon-strate that brand-share prediction can be very eflectivein analyzing purchasing patterns for established brandsas well as new brands.

Figure 15INFLUENCE OF PROMOTIONAL ACTIVITY O N

PENETRATION OF ESTABLISHED BRAND H"

Estimaiedultimate lavets

O I 1 I 4 5 6 7 B q IO II 12 13 14 15 16 17 IB iq ZO 21 22 23 24

Four-week periods alter start of study"Total buyers in product field = 100%.

Source: Attwood Consumer Panel, Gieal Britain.


ESTIMATES OF LIKELY ULTIMATE MARKET

Tab le 4

BY BRANDS G AND H, WITH AND WITHOUT PROMOTIONAL ACTIVITY

Ultimate penetration X Repeat purchasing = Ultimate brand share

Before promotionbuyers already inexpected further buyers

TotalAfter promotion

buyers already inbuyers coming in after the promotion

am327%

24%15

Brand G"

34%25

33%

34%17

8.2%0.75

8.95%

8.2%2.6

10.8%

Ultimate penetration X Repeat purchasing-Buying rate factor = Ultimate brand share

Before the promotionfirst 12 weeksnext 24 weeks and expected further

buyersTotalAfter the promotion

first 12 weeksnext 24 weekslast 44 weeks and expected further

buyersTotal

22.1%7.9

30.0%

22.1%7.5

35.4

65.0%

50%29

46%

50%2920

34%

1.290.85

1.18

1.290.850.90

1.03

14.3%2.0

16.3%

14.3%1.86.4

22.5%

" The promotions apparent effect on the ultimate brand share = 1.85% (or 23% increase).** The promotions apparent effect on the ultimate brand share = 6.2% (or 38% increase).


POSTSCRIPT

Psychological Grouping Analyses for Brand-Share Pre-diction

The Attwood Consumer Panel housewives are classi-fied by several psychological or attitudinal categoriesbased on attitude statements rated by the housewiveson a five-point scale. From her reactions to the state-ments in a particular attitude area eaeh housewife wasgiven a position on that scale. For this analysis purpose,the scale "willingness to experiment in shopping" hasbeen used. The penetration of a new product and thesubsequent repeat-purchasing rate have been analyzedseparately for each of four groups on the scale, rangingfrom those classified as "most willing to experiment"to those "least willing to experiment." The proportionof total housewives falling into each group follows:

ping," was prepared to try the new product than any ofthe other groups. The proportion diminishes steadilytowards Groups 4 and 5, "least willing to experiment inshopping," in which the proportion trying the newproduct is only a little over one-half of that in Group 1.Since this is a new product, with an ill-defined totalmarket, the repeat-purchasing rate is expressed in

Figure 16

CUMULATIVE PENETRATION OF A NEW PRODUCT/FIELD

BY "WILLING TO EXPERIMENT" ATTITUDE GROUPS"

3Or Eslimatedultimate levels

Group Rating Percent

1 Most willing to experiment in 18%shopping

2 303 374 & 5 Least willing to experiment in 15

shopping

The penetration of the new product into each group onthis scale is shown in Figure 16. Thus, a higher propor-tion of Group 1, "most willing to experiment in shop-

O 4 8 12 16 2O 24 2B 32

Weeks after launch af new product"Total househoWs in specified group =- 100%.

Source: Attwood Consumer Panel Great Briiain.


Table 5RELATIVE CONTRIBUTIONS MADE TO ULTIMATE BUYING LEVEL OF NEW PRODUCT BY HOUSEWIVES CLASSIFIED BY THEIR

WILLINGNESS/UNWILLINGNESS TO EXPERIMENT IN SHOPPING

Group Total housewives Penetration X Repeat purchasing = Ultimate buying level'

I. "Most willing to experiment"2.3.4 <fc 5. "Least willing to experiment"

18%303715

(27%)(22)(20)(15)

4.9%6.57.42.3

28352610

1.42.31.90.2

24%39334

100% (21.1%) 21.1% 27.5 5.8 100%

= Expressed in packages per four-week period per 100 households.Source: Attwood Consumer Panel, Great Britain.

absolute quantity terms, but this makes no differenceto the deduced conclusions. For each of the four groups,the estimated ultimate rate of repeat purchasing fol-lows:

Group RatingRepeat-

purchasingrate"

1

234 & 5

Most willing to experiment inshopping

Least willing to experiment inshopping

28

352610

" Given in terms of packages per four-week period per 100buying households.

Thus, although Group 1 members were more pre-pared to try the new product the first time, they wererelatively less willitig to continue buying the productthan Group 2 housewives. Groups 4 and 5 were notonly the least willing to try the new product the firsttime, they were also the least willing to continue buyingit. Groups 1 and 2 made the biggest relative contribu-tions in terms of the ultimate level of product purchases(see Table 5). Although Groups 1 and 2 make a similarrelative contribution to the ultimate level of purchases,i.e., they contribute some 30 percent more to the ulti-mate purchasing level than their proportion in the popu-lation, they do so by different paths. Group 1 house-wives show a greater tendency to try the product in thefirst place but less tendency to continue buying it com-pared with Group 2 housewives who arc more cautiousabout trying it, but once having done so are more in-clined to continue using it.

It is possible that a heavy concentration of advertisingand promotion on housewives in Groups 1 and 2 (thehalf of the population with above-average willingness toexperiment in shopping) would increase the ultimatebuying level of the product. However, many moreanalyses of this kind are needed before these abstractconcepts can be used with any confidence. At thisstage, they serve only as directioti into an area of iso-lating new brand triers who continue to use the product.

CONCLUSIONS

This article attempted to explain five main points:1. Brand-share prediction, derived from continuous

consumer panel data, works. Many analyses con-ducted from Attwood panel data confirm a closerelationship between predicted and observedshares.

2. Prediction, itself, can be made long before thestabilization of share and, hence, is an excep-tionally valuable marketing tool.

3. Refinements of the prediction technique intro-duced since the first studies in 1960 have improvedthe accuracy of prediction, advanced the predic-tion date and increased the understanding of theunderlying characteristics of consumer purchasingbehavior.

4. Analyses over a wide range of product fields sug-gest the existence of several rules of consumer be-havior important in aiding marketing decisions.

5. There is still much to learn from the ramificationsof this basically simple technique, particularly inattitude groupings, loyalty studies, and the rela-tionship of advertising and promotion to ultimatebrand share.

APPENDIX

Objective

This appendix is to:1. formally define terms and visually explain the

underlying brand-share prediction model.2. briefly describe the statistical method used for

forecasting the penetration. (The descriptionwill be brief since details of this and other workare included in a Ph.D. thesis on statistics tobe submitted to the University of London soon.)

3. indicate the degree of predictive accuracyachieved.

Brand-Share Prediction Model

Defining / as a discrete time variable, measured afterthe launch of the new product (Brand a) within a de-


Figure 17CONCEPT OF BRAND-SHARE PREDICTION

in productfield

Non-buyersof Brand T

Buyersof Brand T

Totalvolumein productfield

66%

\ ^34%

59%

"41%

< ^

s W

Buying rate index

59% volurne66% Buyer

= 0.90

34% buvers

- 1.20

Totali, field

level= 1.00

Brand shares bvvolume/expe nd iture

Brand T All others

CZI

75%

59%

Total volumein product field

100%

75%

"25%'

S9.8'!i

Penetration =̂ 34?;, Buying Rate index = 1.20 Repeat-putc ha sing rate = 25% Estimated brand share= 34% X 1.20 X 25%

= 10.2%

fined product field (Field A), then:N(t) is the number of new buyers of Brand a intro-

duced at time /,/"(/) is the number of new buyers of Field A intro-

duced at time /,P{t) is the penetration of Brand a within Field A at

time /,

X 100,

r - O

(i.e., as a percentage).P is the ultimate penetration of Brand a within Field

A - limit P{t).

M{t,r,s) is the amount of Brand a purchased inperiod ,v beginning at time /• after the firstpurchase of Brand a, aggregated over allbuyers of Brand a, based on data avail-able at time t.-

E(t,r,s) is the amount of Product field A purchasedin period s beginning at time r after thefirst purchase of Brand a, aggregatedover all buyers of Brand a, based on dataavailable at time (,

R{t,r,s) is the repeat purchasing rate for Brand awithin Field A at time /, for purchasing

°In all cases amount is weight, expenditure, volume, etc;s is the selected length of the repeat-purchasing interval de-pendent on the frequency of purchase of buyers of Productfield A.

interval s beginning at time r after thefirst purchase of Brand a.

X 100,

(i.e., as a percentage).R{t,s) is the ultimate repeat purchasing rate for

Brand a within Product Field A at time/ (based on a Purchasing interval s).

^ limit R{(,r,s).r-»oo

fV(a,t) is the amount of Field A purchased in periods beginning at time / — .?, by all buyersof Brand a.

f^{A,t) is the amount of Field A purchased, in period5 beginning at time r — s, by al! buyers ofProduct field A.

B(a,A,t) is the buying rate factor for Brand a withinField A at time t.

W(a,t)W{A,t) '

(i.e., as a proportion).Thus, given the three parameters P, r{t,s), and

B{LX,A.,I), the brand-share prediction model is the simplemultiplicative model:

% brand-share prediction

% X % X B{a,A,t)100

Figure 17, a hypothetical example, illustrates the


concept of brand-share prediction. In this example:

P = 34%,

Ris,t) = 25%, and

B{a,A,t) = 1.20.

Although 34 percent of all buyers of Product field Ahave made at least one purchase of Brand a, they repre-sent 41 percent of the total volume purchased of Prod-uct field A, buying at a rate 20 percent higher than theaverage Product field A buying rate. Thus, resultingonly from the penetration of Brand a, the effectivebrand-share ceiling is 41 percent. This share may onlybe attained if all households making a first purchase ofBrand a continue to buy only Brand a within Productfield A. However, it is estimated that, having made afirst purchase of Brand a, 25 percent of all subsequentpurchases made in Product field A will be (ultimately)Brand a. Therefore the estimated ultimate brand shareis 25 percent of 41 percent, or 10.2 percent.

It is advantageous to produce repeat-purchasing rateanalyses by segmented buying groups defined by theperiod of their first purchase of Brand a. The model isthus modified to become:

% Brand share prediction

X XJ={ 100

where Pi%, Ri(t,s)%, and Bi{a,A,t) represent theultimate penetration, repeat purchasing rate and buyingrate factor for the /th buying group, ot which there are h.

Figure 18

CUMULATIVE PENETRATION MODEL"

Raw data

Projections

PfOttuct fields

Esti matedultrmate Dish-mashinglevels liquid

30?; (3 0%r Flour

27% 12 O W Toothpaste

. 2 2 % (1.5%)" Floor and

furniture polish

.7%l' CakBs

A S 12 lt> 2O 24

Wesks after launcn of new product

0100% = All housB>i(ilri5 buying in siiecihecl field.

''Percantage standard deviation ol fitlHd

model actual data far 28 week neriod


Penetration Forecasts

When trying to formulate a penetration model, afamily of growth models were considered and, basedon several case histories, the modified exponential formproved the best choice.

Let

AP{t) =1) - Pit - 1)

K is the ultimate penetration,a is the rate of growth parameter,

ande(/) is the random error associated with the measure-

ments at time /.Therefore, AP(t) = a{K - P{t)) + *(/).

This model is intuitively reasonable since it infersthat the rate of increase ot penetration at time / is pro-portional to the expected number of new buyers. Thedeterministic part of this stochastic model may be re-duced to:

P{t) = K{\ - e--'),

where e is the exponential function.It was suggested by Fourt and Woodlock [3], andAnscombe [I] described a method of obtaining theparameter's maximum likelihood estimates. However,the computations involved are arduous, and thereforea method has been devised applying the method of dis-counted least squares detailed in 14], to provide rela-tively quick estimates of a and K. As mentioned earlierthe results o'( this work wiil be published soon.

Predictive Accuracy

However, it is obviously necessary to indicate thedegree of accuracy with which the penetration model fitsthe data. In Figure 18 examples are given in which,based on the first 12 weeks' data, forecasts were pro-duced for the next 16 weeks. Later, after receipt of sub-sequent observed data, it was possible to compare thepredicted and actual data.

The critical measurement of predictive accuracy wasdefined as the percentage weighted standard deviation(p.w.s.d.). For each example given, the 28 week p.w.s.d.is given in absolute percentage terms, in brackets be-side each specified product field. The p.w.s.d. at time tis defined as:

Pit - 0where

P(t — /) is the (t — i)th (actual measurement of^ penetration),P(t — i) is the {t — i)th (forecasted measurement of

penetration),W' is the weighting function (iv taken as 0.6),


Figure 19

THE ACCURACY OF THE FIHED MODEL

6 r

IO 15 2O 25 3O 35 AO 45 5O

Percentaye of estimated ultiniaie peneiraiion

and

s(t) is a measure of the percentage deviation of thefitted model from the actual data giving more weightto the diOerences associated with the latest observa-tions, i.e., those most critically affecting the estimate ofK (see Figure 19).

Based on these results, it is possible to discriminatebetween relatively stable penetrations and thoseaffected by major marketing disturbances. By compar-ing the projected estimates of penetration, from theinitial relatively stable state, with the actual penetrationin the disturbance period, it is possible to qLiantify theeffective increase or decrease in ultimate penetrationand, with further analyses {repeat-purchasing rates,etc.) in eventual brand share.

REFERENCES

1. F. J. Anscombe, "Estimating a Mixed-Exponential ResponseLaw," Journal of the American Statistical Association, 56(September 1961), 493-502.

2. J. Biiiim iind K. E. R. Dennis, "TTie Estimation of the Ex-pected Brand Share of a New Product," ESOMAR Congress,1961.

3. Louis A. Fourt and Joseph W. Woodlock, "Early Predictionof Market Success for New Grocery Products," Journal ofMarketing, 25 (October 1960), 31-8.

4. W. G. Gilchiist, "Methods of Estimation Involving Dis-counting." Journal of the Royal Statistical Society, Series B,29. No. 2, (1967). 355-69.

use of consumer panels for brand-share prediction of consumer panels for brand-share prediction 133...

Documents