Artificial Intelligence is everywhere. It’s in the small things you have been using frequently for the many years: finding the fastest route to work in the morning, autocorrecting a typo or showing relevant posts on your LinkedIn feed. You have most-likely even used it multiple times before reading this today.
We are used to these applications of AI for many years now. Lately, however, AI has been gaining a lot of attention. And rightfully so! Artificial Intelligence is used to achieve impressive results: writing complete essays within seconds, holding realistic conversations and generating realistic images based on short text prompts.
With these rapid advances, the amount of AI-related content exploded. You can now find an endless supply of blogs, social media posts, and even courses on how to use AI tools like ChatGPT. From health care to financial markets, AI is (being) implemented in nearly every sector. Therefore, we at Gordian are wondering where AI could best be implemented in our field of work. In this blog, I want to focus on the role of AI in spare part management by dissecting it in two parts.
1) Is AI ready for Spare Part Management?
Is AI ready to be used by companies operating in the spares-related supply chain? The answer to this question heavily depends on the intended way the model is to be used. Within inventory management, AI is often associated with forecasting. However, training an AI model requires data, lots of it. For slow-moving spare parts, the amount of available demand data is very limited. Even for relatively simple calculations like reorder point calculations, slow-moving spare parts often have too little demand data for using statistics reliably. For AI models, this (lack of) information issue is an obstacle that might be impossible to overcome.
Maybe we should therefore not focus on whether or not AI models can match or beat conventional methods. We should look for new opportunities, where we focus on the strengths of this technology and the weaknesses of conventional methods. AI models are great in finding patterns in large amounts of data. This could help planners in for example exception management. Tactical planners check updated parameters for only a selection of SKUs. These exceptions are determined by following fixed business rules. Using AI, it might be possible to narrow down this selection even more, allowing the planner to focus on what is necessary.
For a different possible application of AI, we move from a tactical focus (determining inventory parameters) to an operational level. On a day-to-day basis, warehouse employees take several actions to handle solve and prevent machine downtime. Examples of these operational interventions include expediting repair jobs, lateral transhipments and cannibalization. On this level, there are endless different scenarios that could require different actions. For example, when considering the current inventory level, the expected remaining repair time of an asset and the expected future demand, an AI model could help by notifying when some interventions should be taken.
In short, whether AI is ready for SPM depends on how you want to use it. Especially in areas where conventional methods struggle (e.g. where the number of parameters is really high), AI might prove itself to be a suitable solution method. And with the current trajectory of AI-related use-cases and research, spare part management might soon become more intelligent.
2) Is Spare Part Management ready for AI?
The second question might be a bit more difficult to answer: are we ready to adapt? We have been using and relying on conventional calculations for several years. Many companies rely on these formulas to minimize downtime, prevent backorders, and assure safe use of equipment. Some conventional calculations might be very complex, but it has always been possible to trace back how any input parameter changes the output.
This is different for AI systems. Thousands, or millions of parameters (i.e. weights and biases for every node in a neural network) are tweaked to match inputs to outputs. This practice is often referred to as a black box as it is extremely difficult to understand how a trained model “thinks”. Are we ready to trust that the model did not make an incorrect assumption that will result in a bad decision when it is given a situation it has not seen before (i.e. has not been trained on)?
We often see that planners are cautious when moving to calculated, statistics-based inventory parameters, as these do not capture the business knowledge that they have. For most planners, and especially the maintenance workers relying on the availability of spares, it will be too soon to move to AI-trained inventory parameters. The same might hold for the previously mentioned exception management. Using rules, it is clear when an exception is triggered, but how can a machine decide which parts are relevant to check? Also, this requires trust, trust that will be difficult to create.
Although sometimes it might look like it, AI is not a magic, one-size-fits-all solution. Before companies start considering implementing an AI method to, for example, determine stocking parameters, the model will have to be researched, implemented, and evaluated thoroughly. And even then, we should not forget one essential aspect: do the inventory planners, spare part managers and maintenance workers trust the model enough to rely on its decisions?