With all the constant changes in organisational priorities, product mix and manufacturing processes, supply chain planning systems need to continuously adapt and improve autonomously. Just like an autonomous vehicle, they need to plan, sense and respond in real time with little user intervention.
Systems increase the velocity of doing business by having the ability to optimise millions of variables in balancing demand and supply. This requires an accurate representation of the supply chain, namely a digital twin. More importantly, systems need to have the ability to learn and improve themselves. This is accomplished by having self-correcting models, self-improving processes and self-optimising algorithms.
Supply chains are constantly changing. For example, supplier lead times can change over time or equipment efficiencies may change depending on the season. A self-correcting system detects such underlying trends and keeps updating the model, always maintaining a true digital twin.
Having an accurate model is essential but not sufficient. Domain expertise is also needed, making it possible to create optimal plans and respond well to disruptions. Therefore, systems need to self-improve to be able to optimise policies and procedures, such as deciding the best safety stock levels due to seasonal variations, product mix changes or product life cycle.
Lastly, self-optimising algorithms work on improving their own efficiency to provide better results faster. In planning, there are many interactable problems, such that as the problem size grows the run-time for the prescriptive algorithm increases exponentially. By learning from past searches, they can quickly arrive at the result for a new search avoiding the dead ends that were discovered previously.
In general, a supply chain planning system needs to be adaptable to changes in the physical model and changes in the business and its priorities and policies. An initial model becomes irrelevant unless it can constantly adapt itself and learn. To do so, techniques such as deep neural nets and pattern recognition are used to detect trends in demand as well as supply and operations, ensuring that more accurate decisions are made. The older generations of sales and operations planning (S&OP) solutions fail to do this. As a result, they require intervention and adjustments by humans, resulting in sub-optimal plans and inaccurate financial projections.
The two essential elements needed, therefore, are model accuracy and intelligence. Model accuracy requires S&OE solutions. Intelligence comes from a system’s ability to improve itself using artificial intelligence and machine learning. These two ingredients enable manufacturing companies to take a quantum leap ahead of their competition by providing faster and better service at much lower cost.
Cyrus Hadavi is CEO of Adexa
This article was originally published in the Autumn 2021 issue of Technology Record. To get future issues delivered directly to your inbox, sign up for a free subscription.