The need for trade-offs

One of the reasons that planning problems in supply chain management are complex, is because there are multiple objectives that potentially conflict with each other, like delivery reliability and efficiency. For example, when long campaigns are created on machines to reduce setups, orders might be produced too early or too late – to fill the campaign. However, planners are obliged to deal with this reality. They will need to make trade-offs, weighing one goal against the other. What makes it even more difficult, is that different objectives are measured with completely different units of measure. For example, delivery reliability can be measured in the percentage of orders delivered before or on the delivery date, and efficiency can be measured by the sum of the setup time on a resource. In a specific case, a planner can ask himself, what is more important: reducing the lateness of this specific order by 5 days, or reducing the setup time on this bottleneck resource by 6 hours?

Planners will often be inconsistent in such decision making, and potentially influenced by pieces of information that are not relevant at all for this decision. For example, the planner might have had a recent conversation with the plant manager, complaining about the level of setups, which might make the planner bend more towards reducing setups – until a sales manager reminds the planner about late orders.

Objective function

When optimizing supply chain planning problems, the trade-off between conflicting objectives has to be quantified, in order to ‘tell’ the algorithm what to do. This normally means that an objective function is designed, which gives the sum of the weighed individual objectives. This way, the algorithm will optimize something which is a balanced total of the individual objectives. We can now also establish when an algorithm is performing well, as it ’simply’ needs to beat a human planner in reaching a certain level on the summed weighed objectives. Normally, it takes several iterations of setting objective weights, generating a plan, judging its quality and re-setting the weights.

The pitfall of focusing on one objective

For practitioners, it can be difficult to grasp this concept. For example, a company might state that the delivery reliability should be 100%, no matter what. Or, a sales manager might state that delivery reliability ‘always comes first’.  This may sound reasonable but is a very wrong way to formulate the objective of a plan. First, a plan is a complex thing, and just pushing for one objective will harm several other objectives. For example, a planning system might choose to plan much overtime or use very expensive material, just to reach the last 2% of delivery reliability. Is that really worth it? Second, in many environments there will always be something late – and it will physically be impossible to deliver in time, when there is a capacity shortage, or when the manufacturing leadtime is longer than the time to the delivery date. Third, the question of what is achievable depends on situational characteristics – for example, when there is overcapacity, it is much easier to reach a high delivery reliability.

The need for objective functions

These considerations for optimization go beyond the domain of supply chain management. Optimization experts who work in complex domains will recognize the challenges related to creating good objective functions. It can be done, when the users understand that objectives do interact with each other and that an ultimate ’referee’- a total summed objective – is needed to establish the quality of the optimization output. I believe this approach can also be applied to the challenges we are facing with the COVID pandemic. Currently, many governments base their policy of restrictions on the capacity of the healthcare system. Other objectives, like lost life quality for the complete population, or the costs of rescue packages, seem less important – or at least, not part of an explicit trade-off. However, such objectives can and should be considered together, as also here, a balanced decision should be made.

Our models are incomplete

The difference between supply chain planning problems and COVID policies, is that the precise effect of a specific restriction (like closing primary schools) is not known. However, estimates of this exist and are already being used to determine whether restrictions can be eased or not. What is missing in such models, is the impact of restrictions on other objectives, like quality of life for the affected group, unemployment rates, debts. It is like a supply chain planning algorithm, only optimizing delivery reliability. As I stated before, the optimizer will do just this, whatever the cost to other objectives. Instead, governments should quantify the impact of measures to all relevant objectives. They could even create different scenario’s, based on different sets of weights for the objectives, and ask the people’s representatives to choose one of the alternatives.

It is hard – but must be done

Is this difficult? Yes, but we have to make the effort, as the stakes are too high to rely on one-dimensional decision making, and complex problems deserve the right type of models to generate solutions. When we are able to develop models to determine the effect of political party programs on socio-economical factors, we should also be able to quantify all objectives when selecting Covid restrictions. And, as with all models, we should not follow the outcomes blindly, but use common sense and vision during the interpretation.