How do I get started building AI-based demand forecasting as a capability?

Question

Accepted Answer

Building a new capability into an existing one is tricky. A planned, business-backed approach from the beginning is vital to integrating data-driven decision-making that can be expanded to new use cases and business problems. It is an investment that will yield enduring results in the future—an MHI report surveyed supply chain leaders and more than 60 percent of those polled said that they had or planned to invest in AI capabilities in the next five years. The process to get started building and deploying an AI-based forecasting model looks something like this:Define objectives and cases.Reach clarity on who will use the solution and how they will use it.Lay the groundwork with data preparation and engineering.Develop the model, test, and iterate.Why is the groundwork important?When building AI models and other advanced predictive algorithms, the underlying data is the crucial foundation for success. From data collection and cleaning to transformation and feature engineering, the data that an organization has access to can be a gold mine. All of the indicators that predict demand “train” the model to make predictions as new events happen. Transitioning to AI-based forecasting requires analyzing a range of sources such as marketing, consumer foot traffic, product specifications, and more. Building the modeling engine is much smoother when the historic sales, indicators, and product information is easily accessible, particularly considering the wide variance in AI project difficulty. The constant iteration during development, the updates done for continuous improvement, and thorough validation all depend on preparation.Unifying these disparate sources is an opportunity to collate institutional knowledge. However, just because web traffic data (or whatever data you have) is not directly applicable to demand this time does not mean that data cannot be used to solve other problems! By properly documenting data and using engineering capabilities to build a “library” or data repository, you unlock the ability to drive change with facts.Realizing your potential with new uses of AIOnce the “groundwork” of creating a data library that acts as a repository of information to make decisions has been laid, the real fun can begin. The deep understanding and insights gained from studying demand and its relationship to indicators can be used for channel strategy, product assortment, marketing, footprint optimization, and more (see figure). Sensing product cycles and obsolescence before it happens, optimizing channel and assortment strategy to let customers get the best access to their needs, and identifying attractive markets for entry/exit are all decisions that organizations need to make. Predictive AI models can give answers to these questions and enhance the organizational decision-making process.AI-based demand forecasting methods can incur significant investment costs, but the result is a best-in-class capability that not only solves the problems at hand, but also future challenges. Once this has been accomplished, the last step is making it work for you.