Thinking about AI in Supply Chain Operations and Planning

Ketan Varia FCILT and Dr. Abdul Ali CMILT

Chances are you've either used ChatGPT or know someone who's raved about AI and mostly recently Generative AI. In the supply chain world, AI is already being utilised by organisations to optimise processes and drive efficiencies. Coca-Cola(1) recently signed a $1B deal with Microsoft to deliver AI services to it supply chain and operations process. DHL(2) recently announced the use of AI to optimise it warehouse services.

AI can optimise production planning and scheduling by analysing demands, capacities, resources, and order priorities, leading to reduced bottlenecks and efficient resource allocation. Additionally, AI can enhance logistics by optimising route planning, reducing emissions, improving delivery times, and predicting supply chain disruptions.

But we need to understand how neural networks that are behind AI work, so that our profession uses this technology wisely and responsibly.

A neural network is best thought of as a connected network, but one that has a mechanism for learning by making rapid adjustments. This adjustment process involves a hidden layer.

Consider a very simple network above. The hidden layer is a combination of inputs and weights (w1 to w6 in the diagram above), and the weights that can be adjusted. However, the beauty of these hidden layers is that they need no programming or formulae for learning. During training the neural network, all you do is take the output, understand the errors compared to what you expected, and simply adjust the weights. This process is repeated until you have the desired effect, i.e., the output matches what you expect.

The best example is to think of how babies learn to walk. Walking is a very complicated movement. A baby does not think about mathematics when it falls; but has a clear purpose that is to stand and move. Each time the baby falls, it makes a readjustment, falls again, readjusts etc…. until it learns to walk.

This relearning (called backpropagation in AI) is where the power of neural networks lies. The fact that they are multiple connections allows adjustment of many weights to deliver the expected output.

And once the neural network has learnt something, just like when a baby has learnt to walk, it does so without having to remember any complex formulae,. Think of the hidden layer(s) as  ‘muscle memory’.

So what does this mean in the use AI in Supply Chain Operations and Planning? We will look at two areas: demand planning and warehouse management:

1. AI For Enhancing Demand Forecasting and planning

Historically, ERP systems used fixed rules to predict demand. These used statistics (e.g., averaging over a time period) and/or human rules based on experience/collaboration.  However, AI works differently. We can use AI to include a wider variety of input data aspects, considering all factors that generate demand. Through more complex neural networks the AI learns how these different factors can predict demand. And all this without the needs for any mathematical formulas or predictive theory.

Plus we now have access to a vast amount from various sources such as historical sales data, market trends, social media sentiment, and even weather patterns. In fact, many companies have these data lakes readily accessible. Moreover, AI can continuously update and refine these forecasts with new data, ensuring that businesses can respond to changes in demand in real-time more swiftly and effectively.

2. AI for Advancements in Warehouse Management

The impact of AI-powered robots is well documented with companies like Ocado at its forefront. In the warehouses, Asda uses a system from Swiss automation firm Swisslog and Norway's AutoStore. In the US, Walmart has been automating parts of its supply chain using robotics from an American company called Symbotic(3). However, AI is more than just robotics: Traditional warehouse design relies on good design skills to optimise layout but usually considers only a few configurations. AI can create limitless configurations, especially with Generative AI which can help design layouts that minimize travel distances for workers, suggest optimal picking routes, and account for future growth. This can significantly improve picking efficiency and order fulfilment speed. 

Based on your desired outputs (e.g. minimising pick time, machinery use, and congestion) AI can be used to create optimum layouts. It can dynamically adjust the placement of inventory based on order types, weather frequency, machine and labour availability.

In 2023 during the wet July, readymade roast potatoes sales went up by nearly 50%. The power of AI is to include more factors than a human can comprehend.

Ethical Considerations in AI Data Usage

The use of data to make predictions in supply chain operations does present ethical challenges. External data needs to be obtained ethically and internal data cannot breach confidentiality. This highlights the importance of considering the 'human in the loop' approach when working with such a powerful tool.

Other Applications of AI in Supply Chain

1. AI in Optimising Production Workflows: AI could analyse production processes to identify bottlenecks and inefficiencies, suggesting improvements that streamline workflows and enhance productivity.
2. AI for Transportation and Logistics Optimisation: AI can optimize routes, reducing fuel consumption and delivery times. It can also predict and mitigate potential disruptions in the supply chain, such as delays caused by weather or traffic conditions. It can help companies with Load planning for balanced transportation and logistics.
3. AI for enhancing supply chain visibility: AI can work with Internet-of-Things (IoT) sensor inputs to provide visibility into supply chains.
4. AI for managing inventory: AI can help optimise inventory turns and decrease stockouts, supporting retailers and manufacturers to recognise the seasonality of stock-keeping units (SKUs). As AI can read barcodes, text, and other information from images and match them with data a Warehouse management system it can provide real-time insights on inventory.
5. AI for sustainability: AI-powered tools can cleanse and integrate data from disparate sources to facilitate carbon emission measurement and reporting. AI-based solutions can also enable more energy-efficient sea voyages.  Through AI companies can garner insights from their returns data and recognise patterns and underlying reasons helping them with reverse logistics planning and minimising product returns.
6. AI in Ethical Supply Chains: AI can help ensure ethical practices in the supply chain by monitoring compliance with labour laws and environmental regulations. It can track the sourcing of materials and ensure that suppliers adhere to ethical standards.

In conclusion, AI is here to stay. Within Supply Chain and Operational Planning we need to use both our expertise and experience in combination with AI which offers high speed learning and predictability to enhance our processes to ultimate lead to improved efficiency and higher quality. A new set of skills will be needed by the profession to work with AI.

Are you working on a project that is using AI within your organisation? We would love to hear form you.

Ketan Varia and Abdul Ali both support various groups within the Chartered Institute of Logistics and Transportation.

References:

1.   Supply Chain Dive. “Coca-Cola Signs $1B AI Deal with Microsoft to Enhance Supply Chain and Operations.” 

2. DHL Press Release. “DHL Supply Chain Partners with Robust.AI to Drive the Future of Warehouse Automation.” Link.

3. BBC News. “Ocado and Symbotic's AI-driven Robotics in Supply Chain.” Link.

4. Harvard Business Review. “5 Forces That Will Drive the Adoption of GenAI.” Link.