Many industries have huge expectations that optimization tools will improve their decision-making processes, aiming for diverse objectives, such as profit maximization or cost minimization, for example. However, sometimes the adoption of such tools takes time and faces some resistance.
Firstly, it is not always easy to demonstrate the gains from the models and algorithms before implementing them. Furthermore, there is always the "status quo" bias, where the decision-makers prefer to keep working as they always did. Finally, the main difficulty is to get the decision-makers confident that the proposed decisions are coherent and feasible in reality since they might be very different from the current practices.
The best way to deal with this resistance and obtain the users’ confidence is to understand the main reasons that justify the counterintuitive decisions that an optimization tool may suggest, and be vigilant!
Why are some decisions so counterintuitive?
Let's assume that the model is "error-free" and that the counterintuitive solutions do not originate from model incoherence. In that case, we need to understand how they can be applied and why they had never been tried before.
- The tool identifies gains obtained from different combinations of variables: after inserting equations representing the relationship between different variables, the model may explore these relationships to identify opportunities. An experienced professional might be able to recognize the impact of one variable on another, but the computing power of an optimization tool can outperform by looking at the combinations of multiple variables at once (even thousands or more).
- Decision-makers might be biased by their experiences: empiricism is the source used by many decision-makers, relying on their previous experiences for future decisions. The "availability heuristic" usually results in overestimating some decisions ("the best day of the operation had this configuration, therefore we need to aim for these values") and neglect others ("I have never operated like this, therefore it is not possible"). However, optimization tools do not have biases and only rely on the input data, equations, and solving methods to define the best decisions. This can lead to very different decisions that the decision-maker had never thought about before.
- People tend to simplify complex problems to find a solution: sometimes some logical decisions that seem simple can lead to inappropriate outputs. As humans, when we face complex problems, we create rules and mechanisms to simplify them. A classic example can be observed in large companies that integrate a long production chain aiming to reduce their costs: as a simplification, it is common that this goal is uniformly broken down to the departments. However, the most appropriate decision here would be to identify the best opportunities as a whole, seeking a global reduction, even if sometimes the cost of a specific department increases.
- Lack of knowledge of some important concepts and theories: optimization is the basis of many classical theories, such as micro-economy (marginalism and equilibrium, for example). Therefore, the results of some models can be justified by these concepts. For example, it is a common decision to produce the most because it is necessary to dilute the fixed cost, or because the market demand is high. However, by using an optimization model the proposed decision might be different, with the definition of an optimal production level that will maximize profits. This is the kind of solution that might be counterintuitive but is explained because the marginal revenue of an extra production will be smaller than the marginal cost, and the profit will be reduced.
- Values that were not explored might represent opportunities: one characteristic of an optimization tool is to identify the combination of variables that contribute the most to the objective and explore it until it reaches the limits. In this process, it might explore some values that had never been tried, and discover potential gains! In these cases, the decision-makers need to evaluate if there are risks in applying such decisions.
Can my optimization tool be wrong?
In all the aforementioned cases, the optimizations models find opportunities that are not seen by the decision-makers, improving the decisions and results. However, sometimes it is possible that the tool proposes incoherent solutions that are impractical in reality, resulting from flaws in the development process and model creation.
This might happen for several reasons. Therefore, it is important to identify and avoid them:
- The model does not include important parts of the decision scope: during the development of an optimization tool, the developers and users must agree on its scope. If there is miscommunication and some important variables and parameters are left out, it might result in decisions that do not represent reality! To prevent this, a good practice is to define the scope in detail in a document that must be checked and agreed upon by all involved parties.
- Some constraints are not considered: sometimes the tool may suggest a solution that cannot be applied. After investigating why one cannot apply it, some constraints that had not been considered are identified. The calibration phase is essential to prevent that: the developers and users should go through all the variables noting if there are limits that haven't been added.
- The relationship between variables is not well represented: in an optimization tool, especially mathematical models, it is very important that the equations defining the relationship between different variables are well designed. These equations might come from known relations (such as the equation defining the profit as revenue minus cost), from literature studies, or from the development of regressions to predict the impact of some variables on another. To ensure these relationships are well represented and will not generate wrong solutions, it is important to validate them with the process specialists, and to execute tests in the model with different situations, observing how the model will behave.
Leveraging your decisions with the support of optimization tools
As we can observe, the reasons why an optimization tool might propose such different solutions from those proposed by humans' intuition are varied. Its main purpose is usually to assist decision-making, and, therefore, users should treat it as an advisor and always be judicious and open-minded: allow the discovery of different possibilities and assess and understand their impacts and risks, but do not blindly rely on the proposed solution!
After all, it is usually good that the decisions proposed are different from intuition. Otherwise, no gain opportunities would be identified and the tool would be useless!
Author: Cassiano Lima - Senior Consultant at Cassotis Consulting
Co-author: Fabio Silva - Senior Manager at Cassotis Consulting