PAUL D. BERGER* This article discusses how parameters of a cooperative advertising arrangement are currently decided upon and presents decision models which demonstrate that quantitative analysis can lead to a better decision when profits are to be maximized. A real-world application of the methodology is presented. Vertical Cooperative Advertising Ventures Many articles have described the state of cooperative advertising [2, 3, 4, 5, 8]. However, the application of mathematical methodology to decisions on levels of cooperative advertising and the resulting dollar split has not been investigated. This article will show how quantitative analysis can be used to determine the optimal parameters for vertical cooperative advertising ventures. COOPERATIVE ADVERTISING IN to maximize: X)S(A by choice of X, assuming that the retailer, once told -Y, will choose A to maximize: c -\- X)S{A) - (2) A. Using calculus of variations, one can solve for the optimal values of X and A: PRACTICE More than two billion dollars is expended annually for vertical cooperative advertising, and 27% of all manufacturers use this type of advertising [4]. In practice, cooperative advertising ventures frequently are specified on a 50-50 basis, with each participant paying half of the expense. Historically this 50 50 basis seems to be derived from a sense of fairness rather than systematic profit maximization. This sense of fairness is misguided because other arrangements based on profit maximization could benefit both parties. A DECISION - r - (1) (3J (k - r)[S'{A)f - [S'{A)Y + S(A)S"(A) ^ 0 and once determining A from {?'), X is such that: (4) X = \/S'{A) - ( k - c). Optimal conditions (3) and (4) arc derived in Appendix A. However, if (3) yields a negative value for A, the optimal A = 0, and if (4) yields a negative value for X, the optimal X ^ 0. The above conditions hold when SiA) is concave, otherwise these conditions are necessary but not sufficient. Here one must evaluate PR for each A satisfying (3) and pick the optimal A as that yielding the largest P,i . It has been proven that the optimal A must either be zero or a value on a concave portion of the S{A) curve [1]; thus any convex portion of the curve can he ignored as a possible source for the optimal A. Appendix A also deals with solving (3) for A and demonstrates that there may be a remarkably simple analytical solution. In general, (3) can be solved via many standard computer search routines. Now the assumption of a deterministic S(A) curve is eliminated. In the more realistic situation, S{A) is a random variable with expected sales as a function of ^ , ii(A). If one redefines the criterion as maximization of expected profit, the optimal value of A is found in (3) by replacing the terms S{A), S'lA], and S"{A), by fi(A}, ix'{A) and M"(^), respectively. In the present situation, P,M and Pti represent the profits now earned by the manufacturer and the re- MODEL A manufacturer sells a product to a retailer at a price fixed by competition, and an advertising allowance is given to the retailer as a fixed discount per item. Suppose that the variable cost to the manufacturer is Sl'/unit, the selling price from manufacturer to retailer is Sc/unit, and the gross margin of the retailer is S/:/unit. Suppose that the manufacturer will give SA" unit discount to the retailer as an advertising allowance. The retailer chooses A, his advertising level. The manufacturer wishes to determine the best level of X. One first assumes that S(A), the sales resulting as a function of ^, is a deterministic function of ^. When the sales function is deterministic, the best level of X is that which maximizes profit. Thus the manufacturer wishes * Paul D. Berger is Assistant Professor of Management, Boston University. 309 Journal of Murki'iing Research, Vol. IX (August I972t, 309-12 310 JOURNAL OP MARKETING RESEARCH, AUGUST 1972 Figure 1 BENEFIT FROM COOPERATION variance of sales also a function of A, a'^iA). By looking at past data and using standard techniques of assessment of subjective judgment, management's assessments of tj.{A) and a"{A) were fitted to the functions: = IO" (4 - 2e-•*'«') standard cases' = 100 This assessment was the weak link in the procedure. Because variance of sales increases as expected sales increase, it was easier to assess variability related to expected sales, as opposed to directly measuring variability of sales. Management's uncertainty about ti{A) is reflected in a large assessed variance. One can graph n{A) as in Figure 2. Not only did the variance of sales vary with advertising level, but also the reasonable range of sales. Further assessment led to a range of sales as a function of advertising, and the following values were examples: RETAILER'S PROFIT tailer, who each choose one parameter (the manufacturer choosing X, the retailer, A). If the two parties cooperate (move away from their individual optimal values), -dwd jointly set X and A, they can attain a total profit in the system, T, higher than the sum of PM and Pff . As shown in Figure 1, parties must now choose a point along the total profit line. Clearly, the agreedupon point must be between points a and b for any point between these represents a situation in which both parties are better off by cooperating. At point d, the retailer will refuse to cooperate, essentially forcing the situation back to point c. At point e, the manufacturer will refuse to cooperate, and thus the result will be point c. Details of this analysis are in Appendix B. APPLICATION To apply these basic ideas, a soft drink manufacturer's optimal strategy will be determined and compared with current practice. This manufacturer deals with the only independent bottler in a given area distributing the manufacturer's product to retail outlets. The bottler signs a contract specifying the price at which the concentrate is supplied and the fraction of the advertising expense that the manufacturer will pay (currently, 50%). The bottler is then free to choose the advertising level; the amount o^ advertising done by the manufacturer is assumed to be already pledged and thus unaffected by this contract. The manufacturer controls only the price and fraction of the bottler's advertising expenses that he pays. The mathematical model has sales as a random variable, with expected sales a function of A, p.{A), and Expected .sales One standard deviation 2 X 10" 3 X 10" 4 X lOi^ .53 X 10'-' .72 X 10'' .90 X 10" Range of sales .7 X 10^ -^ 3.,30 X 1.24 X 10^ -> 4.,76 X 1.81 A iU -^ 6,,19 X Next, k was specified at .25 per standard case and V at .03 per standard case, where k and r were the gross margin of the bottler and the variable cost of the manufacturer. Further, the manufacturer was assumed to be linear in utility (an expected value decision maker), while the utility function of the bottler was assessed. Using standard techniques for quantification of preference [7], the representative utility function for the bottler was fitted to the curve U{X) = —l/X for positive dollar amounts X. The bottler is averse to risk and would pay a risk premium of I % of assets to avoid a gamble of 50% chance of gaining \0V<, of its assets and 50% chance of losing 10% of its assets. The structure of the actual situation suggested two constraints: 1. Negotiating constraint: the bottler's expected utility must be at least some value U{SP). 2. Operating constraint: if the bottler's actual profit is below some value of SL, the manufacturer must make up the difference. It was decided initially to have P = S200,000 and L = S50,000. The optimal solution was found to be: c = S.178, / ^ .75, and A = S160,000. That is, the manufacturer charges S.178 per standard case, pays 75% of the bottler's advertising expense, and with these input parameters, the bottler will choose an advertising level of S160,000. At this solution, the manufacturer has an expected profit of S412,000 and the bottler has a cash equivalent of $200,000. For a decision maker averse to ' A standard case is equivalent to 24 X 8 = 192 oz. of soft drink. VERTICAL COOPERATIVE ADVERTISING VENTURES 311 risk, the cash equivalent is less than the expected value of the gamble. Details of the analysis are in [1]. Figure 2 EXPECTED SALES AS A FUNCTION OF ADVERTISING A COMPARISON OF SOLUTIONS The arbitrary practice of setting/at .50 can be quite costly to the manufacturer. For example, using no quantitative analysis, l e t / = .50 and c = $A4- (halfway between r and k). The manufacturer would receive an expected profit of S316,000. If partial quantitative analysis is used, let / = .50, but c be set to maximize expected profit. This would yield c = S.166 and an expected profit to the manufacturer of S398,OOO. Thus the manufacturer loses an expected profit of at least 814,000 and perhaps up to vS96,000. The higher profit could be realized at no cost to the bottler. With complete cooperation and an overall system maximization at A — SI50,000, the manufacturer would receive an expected profit of S433,OOO if the bottler remains as well off as before. (There is no actual c or / here, as the total dollars in the system are simply divided appropriately after the fact.) In practice, however, complete cooperation may not be operational, and partial cooperation may be more fruitful. This takes the form of increasing L up from ^S50,000. That is, the manufacturer guarantees the bottler a minimum actual profit of L. As L increases from S50,000 to a maximum of S200,000, the latter essentially being full cooperation, the following results are obtained. For all cases but manufacturer's expected profit of 8412,000, the bottler would probably receive some of the profit for cooperation: S[A) 4 X 10' 2x 10 parameters (c, / , A) are singularly or simultaneously varied from optimality. APPENDIX A To maximize: (5) p^ = {c - r ~ X)S(A) given that once told X, the retailer will choose A to maximize: (6) p^ ^ (k ^ c -\- X)S{A) - A. Then from (6), the retailer will choose A such that: (7) d P ^ / d A ^ (k ~ c + X ) S ' { A ) - 1 = 0 , or: f $ 50,000 100,000 125,000 150,000 175,000 198,000 2a). 000 Manufacturer's expected profit $412,000 412,000 416,000 420,000 425,000 432,000 433,000 Bottler's cash equivalent $200,000 2a), 000 2{X),000 200,000 200,000 200,000 200,000 In summary, this solution yields an expected profit to the manufacturer significantly higher than current practice. Increasing the degree of cooperation will yield even a higher expected profit, while complete cooperation yields the highest expected profit. S'iA) ^ l/(k - (8) Forming the Lagrangian function from (5) and the constraint (7): (9) L= {c~r The major implication of this study is that quantitative analysis may yield significantly better solutions to cooperative advertising decisions than those solutions currently in practice. Other types of cooperative arrangements, for example one in which the advertising subsidy is a fixed dollar amount per item purchased, may be analyzed by similar methods. The sensitivity of the solution to the assessment of the n{A), (T^(A) functions can be examined in detail [1]. Also, the profit figures can be examined as various X)S{A) + \((k ~ c -h X)S'(A) - I). Taking OL/dA and dL/dX, and setting them equal to zero: 00) (c - r - X)S'{A) -\-\{k - c + X)S''(A) = 0 and (11) CONCLUSIONS ~ -S{A) + \S'(A) - 0. Solving for X in (11) and plugging it into (10), and solving together with (7) yields: (12) (k - T)[S\A)f - [S'{A)]' + S(A)S"(A) ^ 0, as the equation that the optimal value of A must satisfy. Then X is found from (8). Solving (U) may be analytically formidable, but under certain circumstances the solution may be remarkably simple. Suppose S{A) is a concave exponential function: (13) S(A) = Ni~ Ni> N,> 0, m > 0 JOURNAL OF MARKETING RESEARCH, AUGUST 1972 312 ^. = ((c - r - pictured below: X)S{A)) (16) S(A) which equals: + ((/c - c + X)S{A) - A), (17) Taking the first and second derivative of S(A), let R = f-'"^, and one can solve (11) analytically and find that R is the positive root of the quadratic equation {(k - r)Nrm + N-,)R' - N^R - A'I = 0, and from the definition of R, A ^ -\og R/m. If i? > 1, the optima! ^ - 0. Then find X = e'^^/mNi - {k - c) or 0, whichever is larger. Now suppose that S{A) is a random variable with expected sales as a function of advertising, t^i^)- To maximize expected profit, (5) is replaced by: = E((c - r (14) = {c-T X)S{A)) - and (6) is replaced by: E(P,) = E{{k - c + X)SiA) - A) (15) ^ {k- c + X)ii{A) - A. Here p.{A) has replaced S{A). It follows that the optimal conditions (12) and (8) for determining A and X are the same here except n{A), fi'{A), and fi"{A) replace S{A), S'{A), and S"{A), respectively. Expected sales as a function of advertising form the basic input. APPENDIX B The choice of A can maximize the sum of the profits and thus total dollars in the system: {k - Y)S{A) - that A. S'{A) = \J{k — The above is maximized at A such r ) , a value different from that chosen when total profit is not of concern; but each party essentially performs a suboptimization. If the parties cooperate and set A at the amount that maximizes total dollars, extra revenue is generated. How this extra revenue is shared must be worked out between the two parties involved. With sales as a random variable, the total expected profit is E{PM + PR) which equals (fc - V)\i.{A) - A, and total expected profit is larger with A such that \i{A) ^ l/(fc — r ) than the sum of the previously expected profits. Similar results are obtained by examining E(PM) and E(PR); the extra expected profit is shared (after negotiation) by jointly choosing an X such that each party has higher expected profit than without cooperation. REFERENCES 1. Berger, I'aul D. A Decision Theoretic Approach to Cooperative Adverlising," unpublished doctoral dissertation, Massachusetts Instituie of Technology, 1969. 2. Burton, Philip Ward. Retail Advertising for the Small Store. Englewood Clilfs, NJ.: Prentice-Hall, 1951. 3. Hutciiins. Musher. Cooperative Advertising The Way to Make it Pay. New York: Ronald Press, 1953. 4. Lockley, Lawrence C. Vertical Cooperative Advertising. New York: McGraw-Hill, 1957. 5. Lyon, Leveretl. Advertising Allowances. Washington, D.C: The Brookings Institute, 1932. 6. Luce, Robert and Howard Raiffa. Games and Decisions. New York: John Wiley & Sons, 1957. 7. PraU, John, Howard Raiffa, and Robert Schlaifer. Jntroduciion to Slalislieal Deeision Theory. New York: McGraw-Hill, 1965. 8. Salisbury, Philip. Cooperative Adverlising—An Effeetive Merehanciisiitg Tool or a Disguised Racket? Sales T'raining File Number 4, S-52G, Newspaper Advertising Executives Association, 1952.