In the supply chain we are, largely, at the front of the game when it comes to machine learning and artificial intelligence. Machine learning has been invaluable as we’ve begun to apply it to demand planning, and is helping with forecast accuracy.
However, supply chain leaders are now looking to use the same or similar technology, utilising machine learning and artificial intelligence, to improve the sphere of production planning. Yet this is a much harder challenge to take on compared with demand planning.
The ARC Supply Chain Planning (SCP) Market Research Study
This recent Supply Chain Planning Market Research Study by ARC looks at what is going on in the supply chain industry now, and what is on the horizon for the next five years. It specifically looks at technology. The central tenet of the study is that the supply chain planning market is “in the midst of a transition to SaaS”.
This is Software as a Service (SaaS) and represents quite a significant change in the way technology is managed. Instead of a large upfront cost for software which is immediately out of date, you pay an ongoing subscription which is provided as a service to your business over time.
From the supplier’s point of view, SaaS increases revenues and should make revenue more reliable. For the user, the initial costs are lower and systems are scalable according to need. SaaS makes sense with the increased availability and application of cloud-based solutions.
The report also gives particular attention to how production planning is central and crucial to supply chain planning and its solutions. Indeed, production planning accounts for 25% of the total market. Production plans are used in a variety of ways from daily schedules on the factory floor through to weekly and monthly capacity plans.
By adding machine learning and artificial intelligence into the equation, there could be continuous improvement in production planning. This is because, unlike a human analysing data, machine learning can take much greater quantities of data and analyse it efficiently, quickly, and in real-time.
This means production plans are continuously able to adapt to actual events and occurrences in the supply chain. It increases both accuracy and efficiency. Forecasting becomes more scientific and less educated guesswork.
Production planning is at the mercy of supply-side planning, and issues here greatly affect production planning. An example of this is lead times. If a long lead time is in place, or there can be multiple changes within the lead period, then a company is in the position of having to hold a greater amount of inventory at the ready.
However, these aren’t necessarily set in stone and therefore you can waste resources unnecessarily. If machine learning is in the picture then human-error and wastage can vastly be reduced.
This area for machine learning is something we need to focus on. The CEO of Adexa, Cyrus Hadavi, explains how there are too many variables for a human to manage the data and decision making. The result would draw on so many variables with incorrect or outdated information that the decision would be wrong.
He states: “with every iteration of planning, there are millions of variables which are constantly and dynamically changing”. Machine learning mitigates this. Adexa is advocating a machine that learns from mistakes and continuously updates its decisions according to changing parameters.
The Problem in Practice
However, creating this type of machine learning is considerably harder than when we’ve created applications in demand planning. In demand management, the system is taking continuous feeds of data from more clear-cut parameters and monitoring this on an ongoing basis to come up with accurate forecast data.
This data, in turn, is then invaluable for decision making. The outcome of whether or not machine learning has proven to be accurate provides the ‘machine learning’ or continuous feedback loop so that it becomes more and more accurate and worthwhile over time. In production planning, it doesn’t work like that.
In supply and production planning, the data is more ramshackle, to begin with. It’s likely to be coming from numerous sources and systems. Therefore, it’s harder to tame and to apply machine learning to – especially if those systems aren’t compatible with one another. Data gets missed, lost, or misunderstood.
The data are also hampered by the accuracy of the individuals who are responsible for updating their area – humans make mistakes and are fallible. In fact, often they cut out the data recording element altogether and the data simply isn’t available to apply artificial intelligence to. The incentive to keep on top of this isn’t the same as the reward gained from demand planning data recording and so it’s a difficult conundrum.
For machine learning to take place the data must be firstly available, and secondly in a good format. Without this, it cannot be analysed and used as a continuous feedback (and improvement) loop. The parameters need to be stringently set and adhered to. This may require the onus being put on the machine to remind the human to enter their data. However, it still has risks.
Can Machine Learning be Used in Production Planning?
Just because there are significant barriers to entry, and notable areas of concern, doesn’t mean that we can’t benefit from machine learning and artificial intelligence in production planning strategies.
There are systems available, and the technology is also there to create bespoke systems as they are needed which are also compatible with other systems. For example, there is the PI System from OSIsoft, which is used by the steel business ArcelorMittal amongst others.
Systems like this, when utilised in production planning, allow you to not only collect the right information but to also analyse it, visualise it, and then share it as needed. However, you do need to combine this technology with other elements and processes to get the most from it.