Approach to Machine Learning in Supply Chain Management

Below is the four types of Analytics which is performed to address wide ranges of questions and problems statement are:

a) Descriptive Analytics :
This is the starting point for any business intelligence. It uses data aggregation and data mining to collect and organize historical data, producing visualization such as line graph, water fall charts, bar charts, pie charts. Descriptive Analytics presents on what happened in the past, but it does not make interpretation/recommendation to solve the problem statement at hand. Today majority of organization and also supply chain profession use this analytical technique to get the business intelligence. This is due to lack of understanding/knowledge of various statistical tools available to perform advance analytics.

b) Diagnostic Analytics:
This is the form of advance analytics to examine data or content to answer the question “Why did it happen” . This is done using techniques such as drill-down, data-discovery, data mining and correlations. One can also bring in outside datasets to be more fully informed analysis. Very few organizations are using diagnostic analysis along with descriptive analytics.

c) Predictive Analytics:
Note more than 5% organization or supply chain professions have ventured into this. It is an area of statistics that deals with extracting information from data and using it to predict trends and behavior patterns. Predictive analytics helps predict the likelihood of a future outcome by using various statistical and machine learning algorithms. The data is pooled with historical data present in the CRM systems, POS Systems, ERP, and HR systems to look for data patterns and identify relationships among various variables in the dataset. When you know what happened in the past and understand why it happened, you can then begin to predict what is likely to occur in the future based on that information.

d) Prescriptive Analytics:
Handful of companies have ventured into Prescriptive Analytics. Prescriptive analytics is the next step of predictive analytics that adds the spice of manipulating the future. Prescriptive analytics advises on possible outcomes and results in actions that are likely to maximize key business metrics. It basically uses simulation and optimization to ask “What should a business do?” . It is a combination of data and business rules. The data for prescriptive analysis are internal and external eg. Social media tweets, feedback, sentiment analysis which uses text mining.
Now that we know the types of analytics, let us understand what the supply chain framework is. A typical professional in supply chain always have a very narrow view of supply chain and always explores some part of supply chain as a full-blown supply chain activity.
Our objective here is to understand how ML can be used in supply chain and for that its important to understand the various legs of supply chain. Below is the range of activities which the supply chain covers. In each of the activities which comes under supply chain gambit, every organization needs to explore the area where predictive or prescriptive analysis is required, for which the ML algorithms can be deployed.

Let us also look at the supply chain matrix which is given below :
Once we understand the SC Planning Matrix, it can be further broken down into long-term, mid-term and short-term activities under each areas of supply chain. We now need to further understand the problem which needs to be solved under each activity.
It's important to understand the business needs with clear priority before the journey on Machine Learning and models are applied. In the early planning stages, a business will typically identify what data it has access to and outline specific goals that it will use data to achieve. These goals may be hard metrics or KPIs the business wants to improve — like perfect order rate, cash-to-cash cycle time, or on-time shipping rate, right plant location, supplier delivery prediction, supplier inventory forecasting, Product mix allocation to plants, Plant capacity fulfillment, lot sizing, etc.

One can also be able to use data to identify suppliers that may be vulnerable to disruption, or otherwise improve your business’ ability to analyze risks and respond to supply chain threats. No matter what the goal is, you must have something clear and specific to work toward. A well-defined objective will help track success and evaluate data use to drive supply chain management.

Once the business need is documented, one can then explore which ML model can be applied to achieve the desired outcome. Below diagram provides the structure of Machine learning (refer earlier article Machine Learning to transform Supply Chain | LinkedIn). Three major components of ML is Supervised Learning, Unsupervised Learning, and Reinforcement Learning. At present, Reinforcement learning is kept out and we will be focusing more on supervised and unsupervised model. Classification and Regression are two major prediction problems which is dealt in Data Mining. Classification and Regression is that classification maps the input data object to some discrete labels. On the other hand, regression maps the input data object to the continuous real values.

Eg of classification problem:
a) To predict whether the shipment will be on time or not
b) To predict whether the supplier delivery will have quality issues or not
c) Companies involved in umbrella manufacturing would like to predict whether it's going to rain or not.

Eg of Regression problem:
a) Predict the inventory levels across warehouses.
b) To predict the on-time %
c) To predict the demand of a product
d) To predict the sales in a particular channel or predict the overall sales
e) To predict the customer returns Cluster analysis or clustering is the task of grouping a set of objects in such a way that objects in the same group (called a cluster) are more similar (in some sense or another) to each other than to those in other groups (clusters). Association rule learning is a method for discovering interesting relations between variables in large databases. Finding frequent patterns, associations, correlations, or causal structures among sets of items or objects in transaction databases, relational databases, and other information repositories.

Eg of Association Rule Mining:
a) Basket Data Analysis - buys(x, “diapers”) ® buys(x, “beers”) [0.5%, 60%]
b) 98% of people who purchase tires and auto accessories also get automotive services done
c) Home Electronics Þ * (What other products should the store stocks up?)

Ef of Clustering Analysis :
a) Segmenting the distribution partners basis their credit worthiness or sales patterns or specific product category sales
b) Segmenting customer basis their demographics
c) Segmenting the supplier Risk basis, the quality of material supplied to plants

Now that we understand types of analytics, canvas of supply chain, creation of matrix to map the activities, problem which needs to be solved and segregating it under supervised or unsupervised learning model requirement, it will become more easy to work on applying the right ML model to get the desired results.