Data science and supply chain modeling teams support executives tasked with making decisions and maintaining relationships with suppliers and customers across the supply chain. One of the secrets behind supply chain planning and operations is maths, namely statistical and optimisation models to help plan everything from best transport routes, distribution channels, inventory volume and warehouse operations, to name some.
But behind this maths is another layer of maths helping drive the artificial intelligence (AI) major companies are using to optimise their demand forecasting in the face of a lot of undeniable challenges.
Supply chains have been in the news like never before. Global pandemic lockdowns, component shortages, a bullwhip effect in demand, rises in transport and freight costs, conflicts, surge demand triggered by social media, and various black swan events like the blockage of the Suez Canal all spring to mind, with manufacturers and retailers shifting between just-in-time to just-in-case approaches to inventory.
There are a lot of calculations behind supply chain planning, using mathematical models and techniques to optimise the flow of goods from suppliers to customers. An estimated 450 million people work in supply chains globally, with billions of tons of goods transported across thousands of container vessels, bulk carriers, freight and passenger planes, and heavy goods vehicles each year. The scale and variety of regional and global supply chains is a mammoth coordination task. Maths is needed and invaluable for supply chain leaders and the more technical teams who create the modeling and do the sums to support decision-making.
The two main branches of mathematics as mentioned, are statistical and optimisation models. That might sound a little complex, but these underlying models help an organisation with demand forecasting and supply chain optimisation. These models have evolved with the advent of cloud computing and advanced programming languages. The advanced models are more accurate, robust, and scalable.
Various optimisation models can be used to solve supply chain problems. Here are some of the most used ones:
- Linear Programming (LP): LP is a mathematical technique used to optimise an objective function subject to a set of constraints. In supply chain management, LP can be used to optimise production planning, inventory management, and transportation planning.
- Integer Programming (IP): IP is a variant of LP where the variables are constrained to be integer values. IP can be used to optimise supply chain problems where the decision variables are discrete, such as in facility location and production schedule.
- Mixed-Integer Linear Programming (MILP): MILP is a combination of LP and IP where some of the variables are constrained to be integer values. MILP can be used to solve complex supply chain problems that involve both continuous and discrete variables.
- Nonlinear Programming (NLP): NLP is a mathematical optimisation technique that optimises nonlinear functions subject to constraints. NLP can be used to optimise supply chain problems where the objective function or constraints are nonlinear.
- Stochastic Programming: Stochastic programming is used to optimise supply chain problems that involve uncertainty. It considers the probability of different outcomes and aims to minimise the expected cost or maximize the expected profit.
- Network Optimisation: Network optimization is a technique used to optimise the flow of goods and services through a network of nodes and links. It can be used to optimize transportation routes, facility locations, and supply chain networks.
Overall, the choice of optimisation model depends on the specific supply chain problem and the characteristics of the data available. It is the unconstrained consumer demand that supply chains are there to serve, and its consumer demand that can drive a speed up or slow down across a supply chain. An accurate demand forecast is the backbone of any organisation, considering the cascading impact on all downstream processes – mainly distribution and production planning for consumer packaged goods (CPG) companies, and buying, assorting, allocating, replenishing, and pricing for retailers.
And it’s here in the demand forecasting space that data scientists are making a big difference. AI-driven demand forecasting is the maths behind the maths for supply chains today, with well-known enterprise companies leveraging AI-driven demand forecasting to give their business a competitive edge and their customers the items they want.
Demand forecasting leverages various statistical techniques to generate an unconstrained consumer forecast. There are three main paradigms of statistical models:
- Traditional time series, e.g., exponential smoothing
- Machine learning tree-based, e.g., light gradient-boosting machine
- Artificial neural networks (ANN)
These modeling paradigms have strengths and weaknesses. It is important to choose the right model based on the data characteristics such as velocity, variability, and context of the problem at hand. AI/machine learning models can identify and estimate all demand drivers, such as seasonal fluctuations, trends, price and promo effects, and cross-product effects (halo, cannibalisation).
Various data elements and signals are ingested in the model – including the typical internal data elements like sales, inventory, price, promo, and orders from enterprise resource planning. And other external sources if available such as weather, social, competitor, and macroeconomics.
AI demand forecasting platforms provide ready-to-go, configurable, and extendable AI models and pipelines for solving essential forecasting needs. Unlike standard data science and ML platforms, data science teams have access to pre-built transformers and components purpose-built for the retail and CPG industries, empowering them to increase the speed of production.
Many companies have been able to build a forecasting model for one category of merchandise. The challenge is getting the forecasting solution to scale across all stock-keeping units, locations, and channels. Solutions on the market can deliver a scalable solution across millions of rows of data that can be re-forecast frequently to ensure decisioning is successful over time, even with constant market volatility.
Replacing a replenishment, planning, or allocation system without fixing the forecasting behind it may give leaders a better user interface, but with the same bad results. Behind all this is the now not-so-secret maths that are helping supply chain leaders work with data science and AI teams to make sure our shop shelves are filled with what we want when we want it.
