Supply Chains Are Increasingly Volatile. Can Forecasting Keep Up?
It’s no secret manufacturing is facing an increasing challenge from market uncertainty. Changing consumer behaviors, continued COVID lockdowns around the world and other disruptions are forcing manufacturers to stay 10 steps ahead.
Data-driven predictions and forecasts can even fall short, if they are based mostly or entirely on internal data. The Institute of Business Forecasting and Planning states that demand forecasts are off by a staggering 37% on average, resulting in excess safety stock, poorly timed promotions and market-share loss.
While internal marketing, promotions and merchandising factors can affect production planning, customer demand is also driven by external factors such as supply chain bottlenecks, health crises, macroeconomic corrections and price index volatility.
However, thanks to the powerful combination of big data and cloud computing, intelligent forecasting is making it possible to better quantify and plan for external forces and fluctuating market dynamics.
Machine learning and artificial intelligence have further supercharged the ability to make accurate performance predictions based on millions of economic leading indicators and model refinements. Understanding the two key steps in quantifying market volatility makes this possibility even more of a reality.
Step 1: Identify the Economic Drivers
Intelligent forecasting starts by using artificial intelligence to correlate millions of economic and industry data sources to historical manufacturing performance. Several leading indicators will be identified as most predictive for the layer of business under review, including macroeconomic factors and supply chain changes, as well as social, industry, weather, financial and government trends over time.
This analysis might initially present hundreds or thousands of prospective indicators that can then be filtered by correlation strength and by context – for example, geography, industry, scope, channel. The outcome is 10-30 market drivers that are typically closely related to the business but can even present a few surprising factors that prove to be predictors. For example, wage increases before luxury good purchases and employment in garden supply stores can be indicators of propane tank sales, just as alcohol consumption can be an indicator of construction staffing levels.
The bottom line is that each manufacturer, each brand and each category have a unique set of leading indicators that most consistently happen in advance of the company’s own change in performance.
Step 2: Create, Test and Refine Prediction Models
The next step involves taking the leading indicators and building predictive models, which ultimately results in the generation of an economic baseline forecast.
Ideally, this is where manufacturers can leverage machine learning algorithms and computing power best. Predictive modelling helps identify the top five to 20 indicators to include in an econometric analysis to accurately determine a 12–18-month prediction, as well as a three-to-five-year compound annual growth rate (CAGR),
One approach is to experiment and manually build multiple models. This can be an arduous process, involving coding in R or Python and building/testing models one at a time.
AI-based analytics engines focused on external real-time insights can create hundreds of predictive models based on various permutations and combinations of those indicators. Machine-learning algorithms enable each subsequent model to learn from the last, reducing errors until the model with the best score is presented.
These forecasts, variances and leading indicator projections can then provide production management and leadership with invaluable insights to refine plans, resulting in inventory reduction, increased profitability and competitive foresight. From there, ongoing, real-time data-driven monitoring will keep manufacturing leaders up to date on any new internal and external changes to maintain operating conditions.
Intelligent forecasts offer a unique opportunity to tell a story, not just of the monthly or quarterly future data points, but of the upcoming winds in the economy. With insight into the causal factors of a forecast, strategic planners and executives can plan for upcoming headwinds and tailwinds.
Rich Wagner is CEO and Founder of Prevedere, a predictive analytics company that helps enterprises create highly accurate forecast models by incorporating the global economic leading indicators.
