You are on page 1of 12

MANAGEICENT SCIENCE

Vol. 18, No. 6, Februaiy, mS


Prmlti in V.SJI.
TOWARD A NORMATIVE MODEL OF PROMOTIONAL
DECISION MAKING*
DAVID A. AAKER
University of California, Berkeley
This paper presents a normative model of promotional decision making and re-
views current literature in the model context. The model emphasizes the long-run
impact of promotions and draws upon stochastic buyer-behavior model technology.
In particular, a stochastic model is used to predict the level of brand acceptance ob-
tained from a group of new triers attracted by a promotionconsumers with no recent
use experience with the brand. This brand acceptance ia made a function of the com-
position of the new-trier group. Finally, attention is focused upon the probability
distribution of those attracted by the promotion, conditional on the nature of the
promotion. This distribution is used to develop expressions for the expected long-
term worth of a new-trier group attracted by a specific promotion.
This paper presents a normative model of promotional decision making and reviews
current literature in the context of this model. The objective is to help provide a more
systematic basis for promotional strategy development. What kinds of promotions
should be used? To what consumer types should they be directed? How long should
they last? What about timing? It is to these kinds of decisions that the model construc-
tion is addressed. The benefit of a normative model is usually not the machinery it
makes available but, rather, the vocabulary it creates, the structure it provides, and
its suggestions for future research. This model is discussed in this spirit.
It was estimated that $4 billion was spent on sales promotion in 1966, about one-
fourth as much as was spent on advertising [17, p. vii]. Coupon promotion alone was
estimated to be $1 billion in 1960 [15]. Despite the importance of this component of
the marketing mix, little attention has been paid to it by management scientists.
Promotion is here defined as those short-run marketing activities other than ad-
vertising and personal selling of manufacturers and retailers that are designed to
stimulate consumer purchases. An important class of promotions are those that offer
a direct extra value to the consumer as a purchase inducement. In this category are
coupons, contests, premiums, and price-off campaigns that provide incentives for a
brand purchase. They may be offered by the manufacturer directly to the consumer
or indirectly through the retailer by means of a trade deal. Also in this category are
special gifts and double-stamp days that encourage a consumer to visit a store and are
offered by a retailer. Although this article is written from a manufacturer's viewpoint,
the analysis would be identical if store choice instead of brand choice were the focus.
Another class of promotions are those directed at the retailer that provide in-store
stimulants to purchase. This would include inducements to use special displays.
An important characteristic of promotions, as here defined, is that they are tem-
porary in nature and therefore tend to stimulate immediate consumer purchases.
Thus, relatively enduring purchase incentives such as trading stamps offered by re-
tailers are excluded. A discussion of the various decision parameters will further serve
to clarify the general problem. This will be followed by the development of the model's
objective function.
* Beceived Match 1971; revised November 1971, June 1972.
693
594 DAVID A. AAEER
Decision Parameters
Let 6 represent a decision vector with respect to a particular promotion. Ultimately,
we want to determine an admissible 6 that will maximize an objective function. The
number and nature of the components of 6 will depend upon the specific context of the
decision. We shall specify here only a few of the more important and most representa-
tive components.
One promotion decision involves the selection of the promotion type. Several ex-
amples, such as couponing, have already been mentioned, but the number of different
kinds is actually limited only by the inventiveness of creative staffs. Let p denote the
promotion vehicle or type. After the type is selected, its unit value still must be de-
termined. Should the unit value (to the consumer) of the premium or cents-off coupon
be ten cents or twenty cents? Let s denote the unit value or size of the promotion.
The determination of the target audiences for the promotion is a third decision.
Clearly, the effectiveness of the promotion will depend upon the segnientation strategy
accompanying it. Let x denote a vector of relevant descriptors of those upon which the
promotion is to be focused. For example, x could specify target levels in terms of age,
income, and product-class usage of those to whom the promotion is addressed.
A fourth decision is the timing of the promotion. When is the optimal time to under-
take such an event? Let t be the number of months since the firm's last promotion.
Such a variable will certainly represent a relevant timing consideration. Promotions
that are too frequent will jeopardize the temporary image that is usually important to
their success. There are many other ways to specify the timing, of course. It may be
worthwhile, for example, to consider the history of competitive promotions. The degree
of detail included in this dimension will depend upon the information available to the
decision maker and on the nature of certain functions to be introduced.
There are many other decision dimensions that could be considered. One of these is
the duration of the promotion. Another is the selection of the method used to distri-
bute the promotion if this is needed (direct mail versus magazine advertising). How-
ever, we shall not formally introduce any additional decision dimensions. Once the
model is formulated, the 6 vector can easily be expanded or modified as required. The
decision vector 6 can now be written in terms of the decision parameters introduced as
e = (p, s, X, <, ).
The Objective Function
The objective function, to be denoted by F(e), represents the value to the firm that
the promotion is expected to generate. Value is interpreted to mean the net present
value of the incremental sales stream generated by the promotion. It is really a predic-
tion of the promotion's value, conditioned upon 6, the decision parameter vector. The
problem can be formally stated as:
,.. select 0 to max y(6),
^ ' subject to: 7(6) ^ 0, Budget Constraints,
where:
0 = (p, s, X, t, ), the decision parameter vector,
F(9) = the net present value to a firm expected from a promotion described by 6.
In the balance of the paper the objective function wiU be expanded by considering
why promotions are used and the nature of the buying process involved. As additional
terms are introduced, the estimation problem will be addressed, drawing upon existing
research.
TOWARD A NORMATIVE MODEL OF PROMOTIONAL DECISION MAKING 595
A promotion can increase sales by enticing present users to increase their purchases
and by obtaining sales from highly price-sensitive buyers. However, the dominant
purpose of most promotions involves a longer time horizon. The hope is to entice new
customers to the brand, who will ultimately develop some level of loyalty toward it,
and to increase the loyalty of existing customers. Thus, the promotion is largely an
investment expected to return future sales that would not materialize without it.
To formalize these remarks, let
(2) y( e) = .s(e) - A - C(e) + L(e) -\
where:
5(6) = total gross margin contributed by the brand during and just after the promo-
tion,
A = total gross margin that would have been generated in the absence of the promo-
tion during a comparable time period,
C(6) = direct promotion cost, excluding the unit cost,
L(6) = the value of any increase in loyalty among existing customers as a result of
the promotion,
W(6) = the long-run value of a group of new triers of a brand attracted during
promotion.
The Short-Term Value of the Promotion
The first three terms in equation (2) reflect the immediate financial impact of the
promotion on the firm. The sales pattern of the product and unit cost data will de-
termine the total gross margin, *S(0), obtained during and just after the promotion.
The relevant time period must extend beyond the promotion when sales are often
depressed because of loyal buyers and others who use the promotion to "stock up"
on the brand. The tendency for a temporary sales decline after a promotion has been
frequently observed [16, p. 86], but, in at least one case [8], it did not materialize. The
gross margin that would be generated in the absence of the promotion. A, and the
promotion cost, C(6), must be subtracted to obtain the net short-run contribution of
the promotion. The promotion cost will include the costs of distributing the promotion,
administrative costs, an appropriate allocation of the supporting advertising program,
and costs of misredemption. The latter term can be quite significantas high as 80
percent [16, p. 47]. The determination of C(6) is usually fairly tractable; the prediction
of the (>S(6) A) term is normally more difficult.
There have been several approaches proposed that provide a start toward determin-
ing (iS(6) A). The dependent variable in a study by Hinkle [13] was the increase in
market share enjoyed by a brand during a quarter in which a promotion was operating.
The base from which the increase was measured was either the previous quarter's
market share or a projection based upon a flow-type Markov model. The study, in-
volving coffee, tissues, and frozen dinners, generated the following findings. A promo-
tion was more effective if it represented a large value (s), was unique (p), and included
acceptable packaging innovations (p). The impact of a promotion was reduced if the
brand engaged in dealing activities too frequently (t). Ehrenberg and Goodhardt [8]
also used a stochastic model projection as a reference for determining the immediate
sales induced by a deal.
Massy and Frank [20] explored short-term price and promotion effects using weekly
market share as the dependent variable. Current and lagged indexes of relative promo-
tion magnitude were among the independent variables. They applied their regression
596 DAVID A. AAKER
model to different segments, showing that those with high promotion elasticities
tended to be nonbrand loyal, users of small packages, and patronizers of chain grocery
stores (x). Webster [28] found that deal-proneness was positively related to housewife
age and negatively related to brand loyalty and the size of the individual purchase (x).
Montgomery [22] found dealing activity to be related to venturesomeness, media
exposure, and gregariousness as well as to nonbrand loyalty (x).
Kuehn and Rohloff [16, p. 119] made an effort to relate promotion-stimulated sales
to promotion size, s. Sales data were used on a liquid household cleaner that had price-
off deals of 8ff, lOff, 12^, 13{i, and 15f(. The dependent variable was a measure of the
rate at which the promotion is sold through at the retail level. From the results, they
postulated an s-shaped response curve which indicated that there exists a threshold
deal size. Sales seem much more sensitive to changes in s above the threshold than
below it.
Determining L(6)
We now turn to the inherently less tractable concept of the long-term value of a
promotion. Central to these discussions is the distinction between two types of buyers
attracted during the promotion period. The first are those who have previously pur-
chased and used the brand and are thus familiar with it. The second are those who are
unfamiliar with the brand"new triers." They have never used it or the use experi-
ence occurred sufficiently long ago that it has been effectively forgotten. One researcher
[3] found that when fifteen months had elapsed since the previous purchase of a brand
in one product class, the buyer's subsequent purchasing patterns were similar to those
of a new trier who had never before purchased the brand.
The long-term value of the first type of consumer is denoted by L(0) in equation (2).
It reflects the HuUian learning theory concept that a brand purchase and use experi-
ence will increase habit strength and therefore brand loyalty. An approach to deter-
mine 1/(0) has been developed by Kuehn and Rohloff [16]. They start by suggesting
that the state of a family with respect to a brand can be represented by the family's
brand purchase probability. The measure of a promotion's long-term impact is the
change in this "state" (conversion) caused by the promotion. The pre-promotion state
is obtained by applying the linear learning model (the parameters of which are previ-
ously estimated from aggregate data) to an individual family's purchase sequence. The
family's initial probability is estimated separately from prior data. The post-promotion
"state" is determined either by applying the model to subsequent purchases or from
subsequent family purchasing statistics. A measure of the long-term impact of a spe-
cific promotion type is then an average of these "conversions" weighted appropriately
by the family purchase volume. Using this measure, they compare a wide variety of
promotion types.
This creative approach of Kuehn and Rohloff, though appropriate for the measure-
ment of L{(t), cannot be applied to W{(i) for two reasons. First, for new triers, the pre-
deal state should clearly be zero. Yet the linear learning model constrains the proba-
bilities away from zero. This constraint can be significant (e.g. above 0.3), particularly
for light users. Second, the linear learning model has been demonstrated [2 (for one
product class)] to be an inadequate model of the purchasing process following the
"first" purchase of a new trier.
The determination of all functions becomes particularly difficult when competitors
frequently run promotions and where there is substantial competitive interaction.
Thus, a promotion may be retaliatory and defensive in nature. An attempt to intro-
TOWARD A NORMATIVE MODEL OF PROMOTIONAL DECISION MAKING 597
duce competitive factors would involve, as a minimum, focusing market models upon
more than just one brand. In one such effort, MacLachlan [18] developed a nonsta-
tionary Markov model in which the state space included the brand of interest, all other
national brands, and all private label brands. He then explored the influence that
variables such as the relative dealing activities of the major competitive aggregates
had upon the transition probabilities. He found, for example, that loyalty to the brand
of interest had some tendency to increase as it increased its dealing activities relative
to other national brands, but had a definite tendency to decrease when it did more
dealing than the private label brands. Dealing activity had little effect on the brand's
ability to increase transition rates from the competitive aggregates. Loyalty to private
label brands was much more sensitive to dealing than was loyalty to national brands.
It is possible to argue that L(e) is often zero or insignificantly positive. The act of
purchasing familiar brands is usually habitual in nature. The purchases merely reflect
previous decisions made with respect to the product class. Therefore, the buyer's
"cognitive map" tends to be unaffected by the purchase and subsequent use experi-
ence. The process is similar to that denoted by Howard and Sheth [14, p. 258] as rou-
tinized response behavior, where the buyer exercises his predisposition toward an
evoked set of a few brands. What positive effect the purchase might have would proba-
bly be offset by the image-tarnishing effects of the promotion.
Empirical support for this position has been provided for one nondurable consumer
product class [1]. The following specific hypothesis was tested:
With respect to a given brand, aggregate long-run purchase patterns of families familiar
with this brand (non-new triers) are not affected by interim promotion purchases of the
same brand.
A Bernoulli model was assumed to be operative for this type of consumer. The spe-
cific hypothesis inquired as to whether the Bernoulli parameter changed as a result of
a promotion purchase of that brand. For each sample member, an estimate of his
Bernoulli parameter was obtained, both before and after the promotion purchase. The
interest was in the difference between these estimates. The estimates were made from
the two five-purchase sequences preceding and following the promotion purchase. Only
those who had purchased the brand in question during the past fifteen months prior
to the promotion purchase were included in the sample. Since the average interpur-
chase time was about six weeks, there were many in the sample whose previous pur-
chase of the brand was separated from the promotion purchase by more than five pur-
chases. A total effective sample size of more than 4,000 was used. The data source was
the MRCA National Consumer Panel. Overall, there was no observed tendency for a
(non-new trier) buyer to alter his brand loyalty on the basis of an interim promotion
purchase. In fact, for four of the seven brands considered, there was an average de-
crease in the Bernoulli parameter estimate.
Measuring Tr(6)
We now turn to the problem of measuring W^d), the worth of a group of new triers
attracted to a brand during a promotion. The worth of a new-trier group will certainly
depend upon their buying potential, the brand's profitability, their acceptance of the
brand, and some discount factor. More specifically, the group's value can be repre-
sented by the following expression:
(3) W{B) = iNim)g-P - TFo = {Nvm){P - W,)/{1 - g) = Wy-Wa,
598 DAVID A. AAKEB
where:
= the long-run value of a group of new triers attracted during a promotion,
n = index of a time period of analysis (i.e. one year),
N = the number of new triers in the group,
V = average per capita product-class purchasing volume per time period of the
new-trier group,
m = gross margin of the brand,
g = discount factor representing the cost of capital and the risks of the market,
P = factor representing the brand's long-run share of the group's purchasing volume
operationally, the asymptotic market-share prediction of a brand choice stochastic
model.
Wo = long-run worth of new triers who would have been attracted in the absence
of the deal,
TFi = iNvm){l/l - g)P.
The first term, N, indicates the number of new triers attracted during a promotion.
The expression, W(fi), must directly refiect the promotion's success in gaining new
triers. The second term, v, represents the buying potential of the new-trier groupthe
product class purchasing volume expected in the future. It reflects the fact that heavy
users are potentially more worthwhile than light users. The third term, m, represents
the profit contribution of the brand. Clearly, the brand's profitability should be ex-
plicitly considered in an evaluation of promotions aimed at attracting new customers.
An implicit assumption in equation (3) is that both v and m are constant over time.
If information in a brand's long-range plan would support an alternative assumption,
it could easily be introduced with a simple modification of equation (3). Perhaps, for
example, the product class is expected to decline and face price pressure because of
external technical developments.
The discount factor, g, should normally be fairly largeperhaps 25 percent per year
since it is absorbing the uncertainties of the market. It might be reasonable to re-
quire it to increase over time, though this complication is not considered.
The term, P, is a summary measure of the brand's ability to gain acceptance and
its "staying power." It is the asymptotic market-share prediction of a brand-choice
stochastic model that is applied to the group of new triers attracted during the deal-
ing period. There are many future market forces that will influence market share, of
course, possibly causing it to move away from P, its predicted level. This risk will in-
crease as one moves out in time measured in years. Still, P is, in a Bayesian sense, a
best expectation of the future. In the absence of other information, it should be an
appropriate indicator of the brand's ability to gain future sales among the new-
trier group.
The transient market-share prediction is omitted in equation (3) to avoid compli-
cating the discussion. It can be added easily when the circumstances warrant. The
transient in one nondurable product class was found to span three or four purchases
for new triers of existing brands and six or seven purchases for new triers of a new
brand [3].
There are several brand choice models that seem capable of representing the deci-
sion process following a first purchsae of a brand. Two [3], [24] were designed spe-
cifically for this task. Another [21] has demonstrated success in a similar situation.
Purchase-incidence models, designed to model interpurchase times and predict sales
levels, may also be considered, especially where the product class is not well defined
[9], [19]. These models would provide NvP instead of just P. An assumption is that
TOWARD A NORMATIVE MODEL OF PROMOTIONAL DECISION MAKING 599
these models can be applied to new triers of existing brands as well as new triers of
new brands. The buyer who has not bought a brand for an extended time period is
thus assumed to have a capacity for learning about the brand purchased that is
comparable to that of a new trier of a new brand. There is empirical support for this
assumption [3], [24].
It should be noted that the group of new triers obtained during a promotion will
contain members whose first purchase was not associated with the promotion, since a
promotion rarely achieves 100 percent coverage due to distribution complications.
Including these new triers in the data base used for the stochastic model parameter
estimation will, of course, provide a welcome increase in sample size. There is some
evidence that it will not severely inhibit model performance [3].
The term Wa is the corresponding value of those new triers who would be attracted
in the absence of a deal. It would be a distortion to credit a deal \\'ith those new
triers who would have appeared in its absence. Ehrenberg and Goodhardt present a
rare eflort to measure Wo using the Negative Binomial Distribution Model [8].
It should be noted that two problems have been by-passed in our analysis. First,
we have corrected for those who would have been new triers during the dealing period
even if the deal had not existed. We have not, however, considered those who would
have tried soon after this time period. Obviously, it is of little value to alter a date of
purchase by only a few months. Second, there are going to be users who are attracted
to the brand by the recommendation or more indirect influence of the new triers at-
tracted by the deal. This secondary effect will be particularly relevant when the brand
is new. It will be less so for existing brands in relatively low-risk, established product
classes in which word-of-mouth activity would tend to be minimal. Parfitt and Collins
[24] handle both problems rather elegantly by fitting a penetration curve for a brand
and observing the change in its asymptote that can be associated with a promotion.
Predicting
Equation (3) provides an expression for the net present value of an income stream
generated by a new-trier group enticed during a promotion. To shift the focus from
measurement to prediction it is useful to consider N and P as random variables and
to introduce two additional expressions. The first is the joint probability distribution:
f{N, y I e), where y, also a random variable, is a vector of descriptors (age, usage
level, etc.) of the new-trier group. In a probabilistic sense, f{N, y \ 6) indicates the size
and composition of the new-trier group that -will be attracted by a promotion de-
scribed by e. It might be useful also to condition fiN, y | 6) on the market share and
maturity of the brand. It is reasonable to assume that the quantity of potential new
triers might be negatively related to each of these dimensions. A dominant brand has
probably received more trials than a small brand. Similarly, as the brand matures, the
number of potential new triers will decline. However, the temptation to add variables
at this point will be resisted. Appropriate generalizations should be evident.
There has been some effort to relate the intensity and nature of promotions to N,
the number of new triers the promotion attracts. Parfitt and Collins [24] provided
several examples involving new and existing brands where a promotion dramatically
increased the number of new triers the brand was projected to receive. One was a
rare example of a trade promotion designed to temporarily reduce the price. In that
instance, the projected penetration level went from 25 to 36 percent.
Claycamp and Liddy [6] developed a regression model to predict early trial of new
consumer packaged goods. The model explained more than 70 percent of the variance
600 DAVID A. AAKER
when it was applied to thirty-five new product introductions. The independent vari-
able with the highest beta weight was the coverage of consumer promotions adjusted
for type and value of offer. Among the other variables included were product aware-
ness, whether a family brand was used, and packaging distinctiveness. Hamm et al.
[11] showed that a free sample increased purchase intentions. Chien [5], using experi-
mentation, determined the effect of various promotions on obtaining new physician
prescriptions. He found that the quantity of new prescriptions was significantly af-
fected by samples, and also by literature mailings. Pessemier et al. [25] cross-tabulated
new triers (early and late) and non-new triers of new heavy-duty detergent against
fifty-seven descriptive variables. Among the more significant descriptor variables were
willingness to try new brands, certainty about current brands, information transmis-
sion habits, and whether the housewife received a free sample.
There has been little effort to relate promotion characteristics, 6, to the type of new
trier attracted, y. There is, of course, a rich and relevant literature concerned with
describing those who tend to try new productsthe early adopters [26], [27]. The
study by Pessemier et al. was in this tradition. Another was the Montgomery and
Armstrong study [23], although they focused on new triers of an existing product,
Crest, just after the American Medical Association's endorsement. They found that
those trying Crest tended to have children and to be nonbrand loyal, gregarious,
venturesome, exposed to mass communication, and somewhat deal-prone.
The second expression used to shift the focus of equation (3) to prediction is the
following:
(4) P = P{y, e).
This equation states that P is likely to depend upon promotion descriptors and upon
the type of new trier attracted during the promotion. Let us illustrate equation (4)
with an example taken from a recent study [4]. In that example, equation (4) took
the form of a linear regression model with four independent variables:
(5) P* = a -\- a^v -\- aJ, + a^d -f a^s,
where :
V = index of purchasing volume,
I = index of brand loyalty,
d = dummy variablecoded as 1 if the first purchase was associated with a promo-
tion; otherwise as 0,
s = size of the promotion,
P* = the asymptotic market-share of a brand choice stochastic model conditional
upon a given purchase sequence.
Thus, in this case:
y = {v, I) and 6 = (d, s).
In equation (3), P was defined as the asymptotic market share prediction of a brand
choice stochastic model applied to a new-trier group. In this example, the definition
of P was modified (yielding the P* of equation (4)) to expand the number of data
points and to exploit the available information. The modification consisted of condi-
tioning P upon the purchase sequence. Since a binary choice was postulated and since
five purchase sequences were considered, thirty-two different subgroups of new triers
were created, with potentially thirty-two distinct values of the dependent variable.
The exact form of P* in general will depend upon the model specified. In most models.
TOWARD A NORMATIVE MODEL OF PROMOTIONAL DECISION MAKING 601
however, those new triers with a purchase sequence following a trial purchase of
(1,1,1,1,1) would have a higher P* than those with (1,0,0,0,0).
Equation (5) was explored empirically with four brands with some interesting re-
sults. As equation (3) indicates, the value of a new trier will increase with his usage.
However, these results suggest that the higher-volume user is more difficult to "win
over," once he is brought to trial, than the more average user. Evidence relevant to
the f{N, y \ 6) distribution indicated that a person who has exhibited a high level of
brand loyalty may be difficult to attract. In our example, however, there was a clear
indication that those buyers with a tendency toward brand loyalty are more likely
than others to accept a brand once they are brought to trial. The effect of d and s on
brand acceptance was negative, as expected, reflecting the image-tarnishing effects of
a promotion, but not as markedly as other research had found [24], [8].
Haines [10] also attempted to relate brand acceptance among new triers to promo-
tions. He applied a learning model to the introduction of a new product in thirty-
three different geographic areas. The projected brand share and the rate at which the
market moved toward this projection were obtained for each area. This rate was af-
fected by prior product availability and advertising expenditures. The steady state
brand share level was primarily infiuenced by per capita promotional expenditures.
With f{N, y I 6) and P{y, 6) introduced, it is now possible to develop an expression
for E[W{B)]. In equation (3), Wi is presented as a function of five terms:
(6) W, = W,{N, V, m, g, P).
Substituting equation (4) into equation (6), we have:
(7) Wr = WriN, V, VI, g, y, 6).
In equation (7), the term, v, is now redundant. As equation (5) showed, the term, y,
logically can contain v as one descriptor of the new-trier group. Dropping v leaves:
(8) TTi = W,{N, m, g, y, 6).
The terms m and g are determined internally by the firm. The term 6 is a vector of
firm decision parameters. The remaining two terms, A^ and y, are random variables
determined by the market in accordance with the probability distribution/(A?^, y | 6).
Taking the expectation of Wi, with respect to f{N, y | 6), yields Ef(Wi | 6). It pro-
vides a precise expression for the long-term value of a group of new triers that may be
attracted during a promotion. It recognizes that the size and composition of the new-
trier group will depend upon the characteristics of the promotion. Further, thr ac-
ceptance of the brand by the new-trier group will depend upon the nature of that group
and also upon the type of promotion that attracted them to try the brand.
It can now be observed that the expectation of the second term. Wo, in equation
(3) is simply Ef[Wi | s = 0]. It is the expected long-term value of the new-trier group
that would be obtained, on the average, in the absence of a promotion. The expression,
s = 0, here symbolizes that no promotion is offered.
Some Implications
As noted at the outset, the benefit of a normative model is usually not the machinery
it makes available but, rather, the vocabulary it creates, the structure it provides, and
its suggestions for future research. The model reported in this paper is no exception.
It suggests that research on promotions should be directed toward the long-run impact
of promotion instead of toward short-run measures such as change in market share.
602 DAVID A. AAKER
In particular, research on the f(N, y, 6) and P(y, e) functions is needed. How will the
size and composition of the new-trier group be affected by promotion characteristics
and competitive interactions? And, how will the promotion and new-trier characteris-
tics affect brand acceptance? Data bases and stochastic model technology now permit
management scientists to make real progress toward the development of an opera-
tional model of promotional decision making.
Hardin and Johnson recently surveyed users of consumer panels and concluded:
National consumer panel data appear to be underutilized, , , , The dynamics of brand
switching, new trial and repeat buying, items likely to be purchased in combination, and
the extent to which new business is created by promotional activities rank relatively
far down the list of purposes to which national panel data are put. The full utilization
of these data may depend to a large extent on the development of better marketing
decision models [12, p, 367],
Hopefully, the model suggested in this article will, in some measure, contribute to this
development.
References
1, AAKEH, DAVID A,, "The Long-Term Value of Temporary Price Reductions," Unpublished
Ph.D. dissertation. Graduate School of Business, Stanford University, 1969,
2, , "A New Method to Evaluate Stochastic Models of Brand Choice," Journal of Marketing
Research (August 1970), pp, 300-306.
3, , "The New-Trier Stochastic Model of Brand Choice," Management Science, Vol, 17, No,
5 (April 1971), pp. B-435-450,
4, , "A Measure of Brand Acceptance," Journal of Marketing Research, Vol, 9 (May 1972),
pp, 160-167,
5, CHIEN, RO BERT I,, "Testing Prescription Drug Promotions," Journal of Advertising Research,
Vol, 4 (September 1964), pp, 9-11,
6, CLATCAMP, HENRY J , AND LIDDY, LUCIEN E, , "Prediction of New Product Performance: An
Analytical Approach," Journal of Marketing Research, Vol, 6 (November 1969), pp, 414i20,
7, DooB, A,, CAKLSMITH, J,, FRBEDMAN, J,, LANDACHER, T, AND TO M, S,, "Effect of Initial Sell-
ing Price on Subsequent Sales," Journal of Personality and Social Psychology, Vol. XI
(1969), pp. 345-350.
8, EHRENBERG, A. S, C, AND GOODHARDT, G, J., "Evaluating a Consumer Deal," Working Paper,
Aske Research, Ltd,
9, FouRT, L, A, AND WooDLOCK, J, W,, "Early Prediction of Market Success for a New Grocery
Product," Journal of Marketing, Vol. XXV (October 1960), pp, 31-38,
10, HAINES, GEO RGE, "A Theory of Market Behavior After Innovation," Management Science,
Vol, 10 (July 1964), pp, 634r^58,
11, HAMU, B, CURTIS, PEBRY, MICHAEL AND WYNN, HUGH F , , "The Effect of a Free Sample on
Image and Attitude, " Journal of Advertising Research, Vol, 9 (December 1969), pp, 35-37,
12, HARDIN, DAVID K, AND JO HNSO N, RICHARD M, , "Consumer Purchase PanelsConcerning the
State of the Art, " Journal of Marketing Research, Vol, 8 (August 1971), pp, 364-367.
13, HiNKLE, CHARLES L, , Temporary Price Reductions as an Element of Marketing Strat-
egy, Marketing Science Institute, Philadelphia, Pa., 1965.
14, HO WARD, JO HN AND SHETH, JAGDIBH, A Theory of Buyer Behavior, John Wiley & Sons, I nc,
New York, 1969.
15, KENYO N AND EICKHART, A Study of Couponing, New York, 1962.
16, KuEHN, ALFRED A. AND RO HLO FF, ALBEBT C , "Consumer Response to Promotions," Pro-
motional Decisions Using Mathematical Models, edited by Patrick J, Robinson, Allyn &
Bacon, I nc , Boston, 1967,
17, LuicK, JO HN F . AND Z IEGLER, WILLIAM LEE, Sales Promotion and Modem Merchandising,
McGraw-Hill Book Company, New York, 1968.
18, MACLACHLAN, DO UGLAS L., "Market Responsive Demand Models: A Comparison of Variable
Markov and Distributed-Lag Approaches," Unpublished Ph.D. dissertation. School of
Business, University of California, Berkeley.
TOWARD A NORMATIVE MODEL OF PROMOTIONAL DECISION MAKING 603
19. MASST, WILLIAM F. , "Stochastic Models for Monitoring New-Product Introductions, " Appli-
cations of the Sciences in Marketing Management, edited by Frank M. Bass, Charles W. King,
and Edgar A. Pessemier, John Wiley and Sons, Inc. , New York, 1968.
20. AND FRANK, RONALD E. , "Short-Term Price and Dealing Effects in Selected Market
Segments, " Journal of Marketing Research, Vol. II (May 1965), pp. 171-185.
21. MONTGOMERY, DAVID B. , "A Stochastic Response Model with Application to Brand Choice, "
Management Science, Vol. 15, No. 7 (March 1969), pp. 323-337.
22. , "Consumer Characteristics Associated with Dealing: An Empirical Example, " Journal
of Marketing Research, Vol. 8 (February 1971), pp. 118-120.
23. AND ARMSTRONG, J. SCOTT, "Brand Trial after a Credibility Change, " Journal of Adver-
tising Research, Vol. 10 (October 1970), pp. 26-32.
24. PARFITT, J . H. AND COLLINS, B. J. K. , "The Use of Consumer Panels for Brand-Share Predic-
tion, " Journal of Marketing Research, Vol. 5 (May 1968), pp. 131-145.
25. PESSEMIEH, EDGAR A., BURGER, PHILIP C. AND TIGEHT, DOUGLAS J. , "Can New Product
Buyers Be Identified?, " Journal of Marketing Research, Vol. 4 (November 1967), pp. 349-354.
26. ROBERTSON, THOMAS S. , Innovative Behavior and Communication, Holt, Rinehart & Winston,
New York, 1971.
27. RoDGERS, EVERETT M. , Diffusion of Innovation, The Free Press, New York, 1962.
28. WEBSTER, FREDERICK E. , JR. , "The 'Deal Prone' Consumer, " Journal of Marketing Research,
Vol. II (May 1965), pp. 186-189.

You might also like