Dr. Cyrus Hadavi is the CEO of Adexa, a leading AI powered supply chain planning company with Fortune class clients in five continents.
Artificial intelligence (AI) and machine learning (ML) have impacted our lives in many ways, though the impact may not be visible to us. Social media sites, e-commerce marketplaces, retailers and advertisers, among others, are benefitting by having a better understanding of our habits and lifestyles. This is attributed to systems becoming more intelligent and learning to adapt by sensing and analyzing data. Supply chain planning systems are also becoming more intelligent. This is manifested in people augmenting decisions made by the system rather than the traditional way of systems augmenting the human decision-making process.
Given the proliferation of data and the ability of computers to analyze it, systems can perform reasoning to recognize deep relationships between parameters that cause certain outcomes and trends. Much like humans, systems can evolve to observe, learn and predict patterns. Systems can also “guess.” They do so by looking at the probability of events based on observed patterns of the past.
Just like humans, systems can also get it wrong. Unlike humans, elements of emotion and compassion are not part of systems’ decision-making process. Systems are not affected by economic and social factors. This trait may be good or not so good depending on the task at hand. To this end, we can deploy systems to do what they do best: perform billions of calculations based on billions of pieces of data and discover what might not be so obvious to humans. They can perform optimization functions and can sense trends and predict outcomes — just like a mathematical formula.
Providing A System With Relevant Data
In order to reason and conclude outcomes and predictions, intelligent systems need to be exposed to relevant data. To this end, a supply chain planning system cannot be effective if it is exposed only to planning data. It needs to have a deep understanding of all other relevant data, from the shop floor to weather and even natural disasters and geopolitical issues, in order to find cause-and-effect relationships. The system needs to have the capability to sense changes in real time and be able to communicate with other parts of the system and/or people in order to alert and ask for help.
Users may ask questions as to why a certain decision was made by the system. This may require natural language understanding by the system and responses alike. Users may ask for recommendations and options that are available to them in case of a disruption. Having a global model of all the different options would depend on how much data is available to the system regarding suppliers, the capacity of contract manufacturers, tariffs and tax implications, demand trends, and regional issues, such as transportation and infrastructure.
Systems need relevant data to self-correct their models and keep them up to date, reasoning capability to self-improve their decisions and design new policies, and intelligence to self-optimize their own performance and fine-tune algorithms. These are all in contrast with the simplistic models of today’s sales and operations planning (S&OP) solutions, which are really an extension of traditional planning.
Moreover, a must-have capability of a planning system is to generate plans that are executable — in other words, the system has planning and execution of the plan in the same environment without manual intervention by the users. Plans generated by S&OP solutions are far from this capability. They can only produce a long-term rough plan.
A Road Map To An Autonomous Supply Chains
There are three ingredients to the digitalization of supply chains: model accuracy, data and intelligence.
In the absence of the ability to represent a true and accurate model of the supply chain, no amount of intelligence can help. Imagine how difficult it would be for an autonomous car to navigate if it could sense the exact number of cars around it, the cars’ speeds and sizes, and the position of actual lanes and curves on the road. A rough model of an S&OP system is exactly that. Making intelligent or even good decisions in such a rough model is almost impossible.
Secondly, the availability of timely data from the past, present (real time) and future (planning data) is crucial. Data is used to build models, sense events, predict events and find causal effects. Data can be from IoT devices, systems of records or streaming solutions, all of which can be relevant to how the system plans and predicts the operations.
Finally, through an accurate representation and relevant, timely data, the system can be in a position to make intelligent decisions and predict potential outcomes. This is accomplished through the use of AI/ML, operations research algorithms and the fast processing speed of CPUs.
If you are on the path to digitalization, it is imperative to understand what systems can do for you. Design your processes based on a system’s capabilities now and in the future, and ensure timely availability of all kinds of relevant data. The paradigm shift is already here, and the speed at which you adapt is your competitive advantage.