Your customers don’t care if you have a great demand forecast. They care if you have the inventory they want when they want it. But carrying too much inventory can reduce profits and might not be practical for some items such as fresh goods. How do you turn a good forecast into a great inventory plan?
About 15 years ago, many forecasting solutions turned to multi-echelon optimization. This process considers the calendar of manufacturers, the timing of transportation, customs, item processing, order cycles, lot sizes, minimums, and shelf life. While this is great in theory, the MEIO (multi echelon inventory optimization) weakness of those legacy-built systems is that they are built on a flawed foundation. The older MEIO systems use a static forecast methodology; often this is a basket of formulas that many call “best fit”. Most of the “best fit” formulas were created over 50 years ago; examples are Holt Winters, Croston’s Method, Poisson Distribution, Exponential Smoothing, etc.
For MEIO to work as intended, it needs a strong demand forecast foundation that can understand inventory demand potential in the modern world, with all its strange anomalies.
Most planners and buyers that I encounter tell me that they spend a lot of time expediting inventory. This is precisely what a strong MEIO system is supposed to reduce…the expediting.
What does multi-echelon inventory optimization need to function properly?
Optimized inventory should come from a better foundation. According to Gartner, this foundation should not start and stop with each change of the formula selected. And, from my findings working in the supply chain industry for 30 years with over 1,000 clients, the forecast foundation should handle slow, lumpy, long tail, seasonal, trending, promotional, causal, and new items. By using a modern approach that leverages more data related to demand, all of these item types, or even the life cycle of an item that shifts from type to type, can be better forecast.
For MEIO to work as intended, it needs a strong demand forecast foundation that can understand inventory demand potential in the modern world, with all its strange anomalies.
A solid foundation can segment demand signals. Demand is not what you shipped in the past. Demand is the number of customers, what they order, when they order or consume products, the weather outside, supply calendars, promotional activities, social media, new product introduction/retirement, trends, and more. Any system that cannot segment and identify these components into their effects is not taking advantage of modern artificial intelligence (AI) and machine learning computing power.
How do you automate demand planning decisions?
As a planner, I spent a lot of time trying to identify problems before they happened (or as soon as possible once they did). This meant I had to track shipments, check days supply of inventory for my key customers, move aging inventory, and take many proactive actions based on my experience. With probabilistic forecasting, a lot of that “chasing” is automated. When planners have “help” in the form of machine learning, smart AI, and a solid baseline forecast that can handle unusual demand patterns, the planners can focus on activities that drive value for an organization. Planning becomes a modelling exercise, not a guessing game.
By setting a solid foundation, inventory planning becomes a modeling exercise. Model for what can happen and apply a proper strategy and the chasing game is virtually gone.
Can your demand forecasting system answer, “what if?”
Legacy “best fit” formulas rely on a spread from a norm. This is a simple math formula that is supposed to account for weather, promotions, trends, seasonality, and other factors. Using such a method is very archaic given the power of modern computers and the abundance of available data. Probabilistic forecasting is a bit of a mouthful but the concept is pretty simple: it provides a range of possible demand outcomes, with their probability of occurrence. A probabilistic forecast takes uncertainty into account and helps you manage risk by better understanding the ‘what ifs’: What if I add a couple more customers? What if my customers buy more or less? What if something goes on promotion on a Monday in February versus a Friday in December? What if this summer is hotter than last? Probabilistic forecasting isolates and models each of these (and more) possibilities into optimized inventory to meet specific strategies.
Maybe it’s time to re-examine your forecasting foundation and bring it into the modern era of machine learning, and probabilistic forecasting.
Download our brief:
How to Calculate Inventory and Service Level Improvements from Supply Chain Planning Software