There seems to be a trend to promote supply chain planning that excludes human involvement. Perhaps this is the result of a renewed faith in algorithms, big data and artificial intelligence (AI). As with many hypes, expectations about its impact are not underpinned with facts and sound reasoning. There are several reasons why we need human involvement in supply chain planning activities. The most important one is about modelling. Every supply chain planning ‘algorithm’ (let’s talk about the meaning of that word later) uses some model of the supply chain to work. The algorithm might even put conditions on how the model is allowed to be represented. In any implementation project, many choices are made about what to include and exclude in a model. Trying to be 100% complete is a recipe for a never-ending project, and delivering a model which is incredibly sensitive to changes in the environment. It is simply too expensive – in most cases – to create a completely fitting model, and hence, the results of any algorithm using that model, must be interpreted and corrected by a human planner.

What do we call an algorithm? An algorithm is a series of steps that are programmed into a decision support system to automate parts of planning or scheduling decisions. Whether the approach is from AI or some other academic field of study, does not really matter. In all cases in practice, algorithms consist of multiple techniques, where one uses the output from the other. In other words, nearly all algorithms in practice are of a hybrid nature. It is again a question of modelling how the pieces will work together.

Some professionals, even supply chain experts, assume that when you are able to formally define the problem (which is hard in itself, as we stated earlier), then it should be possible to feed the problem into some algorithm or ‘optimizer’, which will give the optimal solution in a reasonable time. This idea is very wrong – our current state of technology and computer hardware simply does not allow for this. Algorithms in practice are not ‘context-free’ – they are designed to solve a specific category of problems. Most algorithms for realistic supply chain planning problems contain heuristic elements, to make the search for a good solution more efficient. This means that when the nature of the problem changes, the algorithm needs to be changed too.

The extend to which algorithms are suitable to solve supply chain planning problems also depends on the planning level. Modelling and optimizing an S&OP puzzle is different from optimizing a job shop schedule. In most cases, it cannot be determined what the optimal solution is – even in hindsight. Perhaps forecasting is one exception to this, as the realization actually represents the perfect plan (although you could even dispute whether some real-life events could have been forecasted at all, even with a perfect process).

To conclude, human planners will for a long time remain to be an essential part of any supply chain planning process. The challenge is to design the right division of tasks between human planners and decision support systems, which contain algorithms. Furthermore, organizations should carefully select the right persons to be planners, and offer them the right amount of training. Evaluating performance in planning is hard, which means that it is challenging to determine whether a planner is good or not. Decisions that are taken here and now, might have an adverse effect tomorrow, or might jeopardize performance in an adjacent department. A good way to determine the quality of a planner, is to evaluate the planning process which he or she has used – peer reviewing. This is also the only way to actually learn, as learning by doing will only work when there is explicit reflection.