AI: how close are we to supply chains that can run themselves?

But if this is so, why are we not all putting out feet up as supply chain turns into the new spectator sport? Money is one reason. It costs money to put solid AI and lights-out solutions in place. On the other hand, surprisingly perhaps, concerns about unemployment are less of a reason. Artificial intelligence still works best when combined with human brainpower. Workforces may need to be retrained and reassigned. However, a business run entirely by robots is still a myth.

The First Question to Be Asked

New technology can sometimes be a solution looking for a problem. If AI is to be useful and profitable in the supply chain, it should be driven by issues that affect operations today. Of course, it can also open new opportunities for tomorrow. However, to get management support (aka budget) the fastest, the first question should be this one. “What Needs to Be Improved in Supply Chains right now?” The following are examples.

• Hard to plan for demand
• Excessive safety stocks and bullwhip effect
• Supplier unreliability
• Transport network unpredictability
• Demand by customers and partners for access to a “real person” (or a very good imitation)
• Seeing the real bottom-line impact of supply chain decisions

In different shapes and forms, AI can help with all these problems. However, AI is no panacea. To see where and when it might help, we can start by looking at the types of AI available.

Tools in the AI Toolbox

AI technology is not so new. As far back as the 1950s, technologists had visions of artificial intelligence. They saw it at the service of society and business. While the ideas were good, it took some time for reality to catch up. We can put AI tools into three categories.

• Tools to make smart decisions. A key example is the expert system. This AI tool is based on data and rules. A classic case was the project to create an expert system for the Campbell Soup Company. The firm’s leading expert on large-scale soup production was soon to retire. So, the idea was born of trying to capture that human expert’s knowledge in an expert system. The human expert estimated it would take a few hours to explain what he knew. The project finally took many months.
• Tools to act like humans. Machine learning is now gaining ground. With more data and greater processing power now available, computers that learn by themselves have become more feasible. Using ML, a computer automatically finds patterns in data. It can use them to draw conclusions from new data. Ecommerce recommendation engines do this. They find the patterns in how visitors navigate on the e-commerce site. They also note the end results of those navigations. Above all, they note if the visitor bought something and if so what. If a new visitor arrives and shows a similar pattern of navigation, the engine swings into action with “You might also like (insert name of frequently purchased product here)”.
• Tools to think like humans. Current research focuses on neural networks. The neurons in the human brain and the way they interact is reproduced in software. Human-like thought processes and reactions are the results. Image processing is an example. In logistics, an artificial neural network (ANN) that understands images can help self-driving vehicles to manoeuvre.

Putting AI to Work in Supply Chain

The first applications of AI may not be spectacular. However, they can be vital just the same. Price pressure and the need to make profits have left many supply chain organisations in a tight corner. Their answer has been to apply cost reduction and lean practices. This, in turn, has led to a smaller staff. AI can help by:

• Doing lower-level tasks faster and more reliably. This frees up their human counterparts for work that AI cannot do, such as creating a new supply chain strategy.
• Suggesting actions after reviewing analytics. Data analytics are part of a growing number of systems. Yet the insights they offer may overwhelm as a smaller team. AI can help pick out the insights with the most impact and suggest precise actions.
• Create a knowledge base that new workers can access, and based on the know-how of older workers, as in the Campbell Soup Company example.

AI Inspirations and Considerations for SCM

Despite being called “artificial”, AI for supply chain management and other domains is still rooted in the world of the living. IBM describes its AI-based technology for the supply chain as a swarm of bees, working together. Elsewhere, “ant colony optimisation” imitates the social habits of ants and their innate ability to find the shortest routes to food and shelter. For the supply chain, useful problems solved with this approach include vehicle routing and production process plan/selection.

Fuzzy logic is an AI solution for handling ambiguity and uncertainty. It can help make rules for using subjective criteria. One example is supplier performance evaluation. Other uses are finding out how expensive a product price is perceived to be, and controlling the cost of stocks. Agents are entities that act to achieve goals. They can also compete and cooperate with other agents. An AI agent can use the knowledge and deal with errors. It can also learn from what is around it, and talk to others in natural language. In the supply chain, agents can improve shop floor control, logistics planning, business negotiation, and customer relationship management (CRM).

AI techniques can also be combined for further effect. “Chatbot” AI software can use natural language processing (NLP) for a dialogue with humans. The chatbot can then translate a request from natural English into the instructions needed to drive a backend IT system. The chatbot converts the IT system’s output back into easy to understand language or graphics.

For example, a supply chain manager’s question might be the following: “What is the impact on the profitability of using only air transport for deliveries?”. With voice recognition, also an AI technology, the question does not even need to be typed in at a PC. It can simply be spoken.

However, avoid relying on AI to plug holes or paper over cracks that should not exist anyway. For instance, machine learning can be used to forecast the demand distortion that comes from the lack of collaboration of supply chain partners (the bullwhip effect). Yet a smarter solution might be to get the partners collaborating properly in the first place.

Braking Factors

So far, applications of AI have tended to be for problems that are well understood. Very complex problems or issues that are harder to define have been less present. Uptake of AI in the supply chain can also stall because of the following.

• The name. AI has often been presented as science fiction (thank you, “Terminator”!), or academic and abstract. Some rebranding might be useful. For example, “Chatbot” is a shorter and more intuitively obvious way of saying “autonomous, interactive agent with natural language processing”.
• Lack of skillsets. It takes certain skills to put data (including big data) and AI tools together for useful results. Some solutions manage to be effective while requiring little or no technical skills for use. Others still need a team of geeks.
• Poor usability. This is linked with the point above about skillsets. Even the geekiest can get tired of a clunky interface. New trends like chatbots and NLP are however making things easier.
• Machine stupidity. Everyday AI is driven by computer software, not by brains. It has no notion of “free will”, “initiative” or “creativity”. Workforce skills and knowledge of AI must continue to be developed to help AI to be used in a truly intelligent way.
• Too dumbed down. In trying to simplify, the risk is to strip AI of its real-world relevance. There is a happy medium to be found between rocket science and comic books.
• Glut of information. Market researcher IDC suggests supply chains have 50 times more data available to them today than five years ago. Yet only a smaller part is being analysed. The problem is often knowing where to start.
• Stale or bad data. Data often has a short shelf life. “Garbage in, garbage out” is true for AI too, especially when out of date data is being used.

Return on Investment

AI systems also require investment. Hiring data scientists to prepare and model data for the AI systems can be expensive, as can full lights-out supply chain operations. Often, the monthly ongoing expenses of a “standard” human workforce are easier to handle than a big investment in smart automation.

However, there are degrees of investment, just as there are degrees of AI. A macro in Microsoft Excel or Word is a primitive form of AI. You could call it the amoeba of AI, able to react in basic ways to the data around it. In addition, once you have a license to use Excel, the macros are free.

Macros are also a good example of machine intelligence helping people with tasks that must be done, but that takes a lot of time for not much value. AI, in general, has a big role to play in ensuring that repetitive tasks are done faster and more reliably. AI doesn’t get bored, skip a step, or forget to file the results.

It can’t think outside the box, but it can flag errors and anomalies for a person to then sort out.

Management consultancy McKinsey & Co put it this way. AI and more specifically machine learning is great at “relentlessly chewing through any amount of data and every combination of variables”. Based on what it has been told to look for and what it finds, it can then handle necessary but time-consuming tasks, freeing up people for other work with more added value.

Successful AI Implementation

The following points can help make a success of AI use in the supply chain.

• Focus on business objectives. Lowest internal costs and highest customer satisfaction are typical goals. If an AI solution cannot clearly link to either one, it may be ahead of its time or simply irrelevant. Instead, favour AI solutions that contribute to meeting relevant business goals and solving business problems.
• End-to-end thinking and data. Silo working in supply chains is bad news. Interdependencies often run from one end of a supply chain to the other. Correctly balancing the different components is crucial. If AI is restricted to isolated parts of the chain, the results may be no better than using standard SCM systems.
• Change management. Using AI is a change. AI is also changing fast, meaning still faster change for your enterprise. People often resist change or at least need time and help to feel at ease with it.
• Continuous improvement. Kaizen got it right, decades before AI was in vogue. In many cases, data is always growing and changing. Campbell’s Soups may have found success with a rule-based system because raw food ingredients do not change. In other sectors, change is the only constant. AI then needs to constantly examine the new data, learn from it, tune itself and stay up to date.
• Scalability Large supply chains can have millions of stocking locations. AI solutions must be able to make the right decisions, fast, and at scale.
• Good user experience. AI and human intelligence must work together. Whether in customer interactions, on the production line, in the loading bays, or on the road, people need to feel at ease with whatever AI tool is helping them to work better.