Invisible Inventory: Why Supply Managers Need Machine Learning

In field of Operations Planning, few questions are less jarring and harder to solve than those of Inventory Forecasting. Several methods have been tried and tested, which often result in lack luster accuracy due to reasons that very few Supply Managers can account for. The current methodology is simple, easy to understand and often wrong:

1. Create a sales forecast against a list of sales units
2. Break the sales units down into their base element, or Bill of Materials.
3. Forecast the lead times for each element to arrive
4. Determine ending inventory per period through the following arithmetic:
   1. Beginning Inventory + Inbound Elements – Outbound Elements (Production) = Ending Inventory

To be frank, this methodology fails to take into account the numerous elements that affect a supply chain outside of sales and lead times. Confounding variables such as weather, order habits of supply planners, historic forecast accuracy, geographic hurdles, supplier bottlenecks and a host of other variables are left out of the equation at the determent of the business.

The result? An array of business meetings that often result in Supply Managers meandering through inventory files wondering how it was possible that their forecasts resulted in out of stocks in critical units and overstocks in units the business would soon discontinue.

It is a comedy of errors that plays every financial period in every Fortune 500 company and it’s about time someone minimized all 30 tabs of Microsoft Excel and asked the crucial question: “Is there a better way?”.

Fortunately progress in Data Science and Machine Learning are providing answers that can finally account for hidden patterns in Inventory Data which can begin shedding light (and hopefully inefficiencies) in the way large companies forecast stock. Though the applications have not yet garnered mass adoption amongst larger firms, companies such as Flieber have begun implementing Machine Learning solutions to solve for SMEs inventory predicaments. Quick and easy libraries such as Scikit Learn now provide an opportunity for would-be Data Science oriented Supply Managers (such as myself) to begin implementing solutions that seek hidden correlations in Inventory Data. Mythical queries into how weather, sales forecast accuracy, truck availability and even planner attitude can now be tested in the crucible of correlation. Key variables and relationships can now be isolated, and their weighted effect on accuracy can now be measured.

I am confident that, if correctly applied, Machine Learning can begin solving the woes of large corporations and spread a movement that will eventually rival and replace the overly simplistic understanding of the Inventory Forecast Process.