One of the greatest industrial applications of artificial intelligence is improving a global supply chain.

Specifically, the use of AI to optimize enterprise resource planning (ERP) for companies will be big business in the coming years.

As this HBR article notes:

Companies are cutting supply chain complexity and accelerating responsiveness using the tools of artificial intelligence. Through AI, machine learning, robotics, and advanced analytics, firms are augmenting knowledge-intensive areas such as supply chain planning, customer order management, and inventory tracking.

A supply chain is the process of taking raw materials into a finished product - the key players in the supply chain include manufacturers, distributors, and transportation companies. In this case, we are using machine learning to make the whole process of goods being exchanged more efficient.

Here are a few steps in the supply chain that can be improved with machine learning:

• Demand forecasting - the goal is to ensure the right amount of raw material is created to meet the projected demand
• Allocation & replenishment - the goal is that produced goods will then be allocated to the optimal transportation networks to meet the forecasted demand
• Assortment & space planning - the goal is to optimize the distribution phase when products are sorted, packaged, and stored for the retail centers

The supply chain process is a long and complicated process with many factors that affect profitability for businesses, luckily machine learning is perfect for these tasks.

## Machine Learning Algorithms in the Supply Chain

Let's look at simple examples of how machine learning algorithms can be applied in the supply chain.

#### Supervised Learning

We could apply labels to our data and make this a supervised learning prediction problem. This is a simplified version of the process, but all we have to do is decide what our target variable is - for example, it could be predict the freight cost given all these other features. We would then:

• Regularize our data so it's all numerical - currently the dataset has dates, weather conditions, names, etc.)
• Split our dataset into training and testing subsets using scikit-learn's train_test_split function. All of the features would be our $X$ values and $y$ would be our prediction variable (freight cost).
• We would then feed our training data to a neural network since they've been shown to outperform all other machine learning algorithms
• We're using the neural network to predict the mapping between the X and y's
• We would then test the algorithm with new features (future price, weather conditions, etc.) and it will tell you what the likely freight cost will be

#### Time-Series Forecasting

While supervised learning could help us make predictions in the supply chain, a better solution could be time-series forecasting.

An example of a question we could answer with time-series forecasting is: "what date should I send a shipment so that the price is the lowest?"

We want to be able to explain our conclusion to companies, and one of the best ways to do this is with data visualization.

As an example, since there are many features in our dataset, we could use dimensionality reduction using PCA or t-SNE to get a two dimensional graph with all our features on the y axis and dates on the x axis. We are then looking for the lowest point in the graph, which represents the date on which the price is the lowest.

The tricky part with supply chains, however, is that there is not just one dataset that you can access to see the whole process, there are many companies and datasets involved.

## 7 Applications of AI in Supply Chains

### 1. Chatbots for Operational Procurement

• These can be used to inform key employees of a company about potential disruptions in the supply chain, retrieve relevant data from the companies database, and provide possible solutions to the disruption

2. Supply Chain Planning & Predictive Analytics

An example of a company doing this is ClearMetal, which offers end-to-end supply chain visibility in order to:

• Improve inventory forecasting
• Eliminate detention and demurrage fees
• Increase personnel efficiency
• Reduce buffer stock

3. Machine Learning for Warehouse Management

An example of a company solving this problem is Rubikloud with their intelligent decision automation software.

4. Autonomous Vehicles for Logistics and Shipping

As self-driving cars are making their way into society for consumer use,  they will more likely start providing value in manufacturing and supply chains. As this article describes:

Autonomous vehicles have a much more serious and quantifiable impact on supply chain logistics and operations than their potential for transporting people. The article notes that many companies believe the market for autonomous vehicles in the logistics space will become a \$1 trillion market.

5. Natural Language Processing

Aside from chatbots and customer communication, natural language processing can be used for many use cases in the supply chain such as:

• Extracting important data from customers, suppliers, and documents
• Interpreting or querying huge datasets

The feedback provided by scanning through supply chain contracts, purchase orders, and other documents can be incredibly beneficial for optimizing supply chain management.

6. Predictive Analytics for Supplier Selection & Relationship Management

Since sourcing the right suppliers is such a crucial part of supply chain sustainability, companies have bee using machine learning to supplier selection and risk management.

This can be done by using machine learning algorithms to analyze and provide predictions based on things like supplier assessments and audits, given a users specific criteria.

7. Robotics

Of course robotics and the AI that powers it will be incredibly valuable for all aspects of a supply chain, as this article describes:

The robots automate picking and packing processes in large warehouses. Supply chains and robots form a natural relationship, automating essential, but often monotonous, tasks.

## Case Study: Cognitive Logistics Solution

If you want to see an example of a cognitive logistics solution that analyzes real-time data, provides intelligent recommendations, and has a monitoring dashboard - check out the IBM Logistics Wizard demo, the code for this demo can be found here.

In this case the example company has:

• 5 distribution centers
• 5 active shipments
• 4 retail centers
• 1 shipment completed

When we hit "simulate storm" we get a weather alert for a "moderately severe snow storm", which leads to supply shortages in the affected cities.

The simulator then uses AI to suggest shipments which you can either approve or reject.

## Summary: Artificial Intelligence for Supply Chains

To summarize, the supply chains of tomorrow will have an end-to-end fully autonomous systems for managing supply chain logistics.

Other technologies of mention that will will play a huge factor in these systems are the internet of things (IoT), and of course, blockchain.

These cognitive logistics systems will be self aware, self governing, self determining, and self-optimizing.

If you want to learn more on AI in the supply chain, check out these resources: