JUDGMENT AND DECISION-MAKING PROBLEMS IN MEETING THE DEMAND

Demand planning activities can be divided into two steps: statistical analysis of historical data and managerial interpretation of market information. In the first stage, quantitative information, especially historical sales data, is analyzed to identify the quantifiable components of the series, that is, the level of demand, long-term trend, cyclical oscillations and seasonal effects. However, some information relevant to meeting demand is not directly quantifiable – promotions, new product launches, competitor actions, etc. – and, therefore, must be interpreted for the correct allocation of resources to meet the demand.

Typically, a sales, marketing, or operations specialist is responsible for interpreting and incorporating this information into the sales forecast. The success of this second stage of the demand planning process is based on the basic premises that this specialist has extensive knowledge of the market and that he will use all his rationality or “common sense” to carry out this interpretation.

However, budget and time constraints in companies compromise, in part, this process, as they do not allow all the necessary information to be collected and analyzed to improve the decision-making process. To get around these restrictions, planners unconsciously use some rules that simplify the process of interpreting market information. These rules are known as decision heuristics.

Despite their practical importance, streamlining the decision-making process, heuristics lead to systematic, predictable errors that are difficult to eliminate. The more complex the environment, the more difficult the information interpretation process is and the more prone the planner will be to unconsciously use these heuristics. The three main heuristics of probabilistic judgment will be seen below – Representativeness, Availability and Anchor and Adjustment – ​​and their impacts on demand planning.

Representativeness

The representativeness heuristic is characterized by the search for peculiar aspects of a probabilistic event that correspond to a stereotype. For example, planners forecast the demand for a new product based on the similarity (representativeness) of this product with previously launched ones, according to the success of these launches. In some cases, the use of this heuristic can be a very useful instrument for a first approximation. In others, however, it can lead to pre-trial errors.

To try to clarify how the representativeness heuristic works in the human brain, consider the following sequences of heads (K) and tails (C) of an unbiased coin:

CCCCKK
KCKCCK

Groups of students from MBA and Improvement programs in the Operations and Logistics area were asked which of the two sequences is the most likely to occur and the majority (80%) answered that it is the second. The correct answer in this case is that the probability of either sequence occurring is equal. To understand why most executives responded that it would be second, it is necessary to analyze the process of forming a sequence of heads and tails.

The first formation rule on an unbiased coin is to expect approximately 50% heads and 50% tails. When executives unconsciously analyzed this rule, they found that both strings met this requirement. However, the second formation rule is that the order of heads and tails is random, which is not recognized by the brain in the first sequence and, therefore, the executives indicated the second as the most likely to occur.

This probabilistic judgment problem also occurs in lottery games. Although any sequence of six numbers drawn has the same probability of occurring, few people bet money on six consecutive numbers, as the human brain has difficulty recognizing this type of sequence as random. For these examples, it seems trivial to identify the failure to identify the random component. However, in a series of sales it is not always easy to identify this component among the level of demand, long-term trend, seasonal effects and cyclical fluctuations. These elements are often confused and make it difficult to interpret demand behavior.

Failures in the representativeness heuristic occur mainly due to problems with the intuitive notion of randomness and insensitivity to sample size. In demand planning, these biases can cause failures in the extrapolation and monitoring of demand, as well as making the planner “see” trends in the sales series.

To exemplify these failures, an exercise was proposed for a group of executives from the operations area. In it, a sequence with six sales figures was presented, without information about the product or the market, referring to the months of May 2002 to October 2002, as shown in Table 1.

Then, the executives were asked to project the sales volume for the following month (November/2002) and say what percentage of confidence they had about that forecast (0% no certainty and 100% absolute certainty). It was not surprising that all executives put a number greater than the October value for November sales, because with the available data one could imagine a sequence of increasing values. However, the fact that approximately 80% of the executives placed more than 50% as a confidence percentage for a forecast made without knowledge of the product or the market, and with only six values, indicates an overconfidence in a very small sample. , where the random effect outweighs the effects of a possible long-term trend.

Graph 1 shows that the next value in the presented sequence is a “valley” of diesel oil demand in Paraná, which was the sequence used in the exercise. The failure in the executives' evaluation stems from the difficulty in recognizing that a series with six increasing numbers can be a random sequence. It is fundamental that the planner is aware that, in the short term, the random effect is preponderant in the evaluation of sales variation and that trend effects should only be considered in the long term.

Availability

The availability heuristic is characterized by evaluating the probability of a certain event occurring by the frequency with which examples or occurrences of this event are available in memory. An emotionally charged, vivid, tangible and/or specific event will be more readily available in memory than a subjective, bland and/or non-emotional event.

For example, studies with people who have suffered some type of violence, such as kidnapping or assault, indicate that, in the post-traumatic period, these people start to develop exaggerated protection mechanisms, such as checking several times if they are not being followed, distrusting everyone people around them and, in some cases, not wanting to leave the house. This phase lasts about a month (this type of behavior for longer periods can characterize the development of post-traumatic stress disorder). However, the fact of having been kidnapped or robbed has no influence on the possibility of a new event of this nature occurring. The fact that this type of event is intense and emotional causes the brain to overestimate the likelihood of a recurrence.

Thus, it can be seen that the biases caused by the use of the availability heuristic are related to problems of vivid and recent information, since a fact that makes an event more “available” in memory does not necessarily make it more likely. It is not always easy to distinguish what really impacts the probability of an event occurring. Understanding the availability heuristic is essential in exceptional situations in demand planning, such as promotions and sporadic events.

The realization of a sales promotion changes, in general, the pattern of demand in the period in which it is carried out. As it is a specific event, the promotion has an intense effect on demand planning decisions for the following periods and, mainly, on future promotions. A successful promotion, i.e., when sales increase significantly and stocks are eliminated (sometimes leading to product shortages), results in increased stock levels in subsequent periods, as shown in Graph 2.

It is observed that the shortage of stock constitutes a vivid event for the planner, who increases the stock level in subsequent periods to protect himself against possible shortages of products. On the other hand, when the promotion is not successful, that is, sales do not reach the planned levels, inventories tend to decrease in the following periods, as shown in Graph 3.

Thus, information referring to a non-regular period of sales has a significant impact on decisions to meet demand in subsequent regular periods, as well as future promotions. In the first case, the lack of products caused the planner to overestimate the probability of a stockout. On the other hand, the expectation of unconfirmed sales caused resupply decisions to be reassessed. It is of great importance that the planner is aware that the intensity of an event changes his perception of the real probability of a new occurrence of the event.

anchor and adjustment

In the anchor and adjustment heuristic, the planner evaluates an event from an initial value, adjusting it with the available information until reaching a final decision. The initial value, known as the “anchor”, can be obtained from historical data, from the way the problem is presented or from random information. The problem with this heuristic is that, in dubious situations, information that is not very relevant can have a great effect on the final decision, if it is used as an anchor for later adjustments.

In negotiations, it is very common for the experienced negotiator to try to “plant an anchor”, that is, an initial value from which the details are discussed (“adjustments”) to arrive at the final value. This happens because these negotiators know that, once the anchor is set, value adjustments are unlikely to be enough. This technique is often used in complex business negotiations, but it also works well in the purchase/sale of real estate and automobiles.

An example of the use of this heuristic for probabilistic judgment was presented by Bazerman (2004), based on studies carried out by Harvard University with tax auditors from the main auditing firms in the United States, who were divided into two groups.

The first group was asked: “Do you believe that more than 200 companies among the thousand largest American companies fraud their balance sheets?” And then, "What's your best estimate of the number of companies out of the top 1 US companies that cheat on their balance sheets?" For the second group, the questions were: “Do you believe that more than ten companies, among the thousand largest American companies, fraud their balance sheets?” And “What is your best estimate of the number of companies, out of the top XNUMX US companies, that cheat on their balance sheets?” It can be observed that the questions are practically identical, changing only the number of fraudster companies (anchor) in the first question. The results obtained for the second question are shown in Table XNUMX.

The answer to the first question was irrelevant to the study, as the objective was just to “plant the anchor” for the second question, in which the auditors had to estimate a value for the number of fraudulent companies. Despite the assumptions that the assessment capacity and knowledge of the market should not be very different between the two groups of auditors, the average of the answers to the second question was completely different in each group. This result shows the influence of the anchor in the process of evaluating the probability of an event occurring.

The biases caused by the use of the anchor and adjustment heuristic come from the inappropriate use of the anchor and/or insufficient adjustment of this value. When they do not have the information recorded or are unaware of the appropriate methods to carry out the sales forecast, some demand planners use the artifice of deciding the future value of sales by adjusting, based on tacit knowledge of the market, the value of sales in the same period of the previous year or month (anchor). This type of approach can cause significant errors in demand planning, since the past value may have little representation for the current sales situation, as, for example, in series with a marked trend.

Other biases

In addition to the analysis biases caused by the three main heuristics presented above, there are two other biases that cause significant impacts on the market information interpretation stage in the demand planning process: the “confirmation trap” and the “retrospective forecast”.

The “confirmation trap” is related to the asymmetrical weights assigned to available evidence for the “hypothesis tests” that are conducted in the planning process. Evidence that confirms the planner's beliefs is generally given greater weights than information that contradicts initial expectations. This makes opinions once formed extremely resistant to change. In the demand planning process, it is essential, therefore, to seek contradictory evidence, rather than confirmatory evidence, which will bring more benefits to the analysis process.

In turn, the “hindsight forecast”, also known as hindsight, refers to the famous phrase: “I already knew”. After the occurrence of an event, the planner tends to overestimate his predictive capacity, that is, after knowing the actual sales for a period, the planner starts to see a direct relationship between the different events that led to that sales volume and mistakenly estimates the assertiveness the prediction he had made. This bias reduces the planner's ability to learn from the error, a determining factor for improving the sales forecast. In addition, in collaborative planning processes, where the opinion of all those involved is essential for the effectiveness of the process, this bias can generate criticism of the predictive capacity of some of those involved.

How to avoid biases?

Complete elimination of biases in demand planning is very difficult to achieve, but its effects can be greatly reduced. The first step is to raise awareness of the existence of judgment heuristics and recognize their impact on the planning process. Bazerman (2004) also suggests a series of strategies to improve the decision-making process, such as acquiring experience and technical knowledge of sales forecasting and using linear models based on expert judgment.

In addition, a normative approach can be used, initially carrying out statistical analyses, incorporating available market information and finally combining these two pieces of information according to the degree of precision that can be associated with each type of information. It is very common to construct a confidence interval for quantitative analyses, but one is rarely used for the interpretations made by planners. The proposal is precisely to provoke a new reflection in the interpretation of market information.

CONCLUSION

Despite being a fundamental part of managerial functions, judgment and decision-making activities are, in general, little explored in the training of managers. Specifically in demand planning, in which the interpretation of market information is a determining factor for the success of the process, knowledge of the heuristics of probabilistic judgment allows the planner to reflect on the assumptions for his decision and, with this, minimize the forecast error of demand.

Thus, we sought to present the main heuristics and their impacts on demand planning, highlighting that, to a greater or lesser extent, they are routinely used to simplify decision-making in increasingly complex environments and with budgetary and time constraints. bigger times.

BIBLIOGRAPHY

Bazerman, Max H.. Decision Making: For Business and Economics Courses. Elsevier Editora, Rio de Janeiro/2004, 5th ed., 348 pages.

Bukszar, Ed. Does overconfidence lead to poor decisions? A Comparison of Decision Making and Judgment Under Uncertainty. Journal of Business and Management, Winter 2003, 9(1), pp.33-43.

Gomes, Luiz Flavio AM; Gomes, Carlos Francisco S. and Almeida, Adiel T.. Managerial Decision Making: Multicriteria Approach. Atlas Publisher, São Paulo/2002, 348 pages.

Goodwin, Paul and Fildes, Robert. Judgmental Forecasts of Time Series Affected by Special Events: Does Providing a Statistical Forecast Improve Accuracy? Journal of Behavioral Decision Making, Mar. 1999, 12(1), pp.37-53.

Heller, Robert. Decision Makers. Makron, McGraw Hill, São Paulo/1991, 348 pages.

Jones, Harry and Twiss, Brian C. Technology Foresight for Planning Decisions. Zahar Editores, Rio de Janeiro/1986, 1st ed., 348 pages.