Probabilistic Forecasting – a Primer
At ToolsGroup we have been big advocates of probabilistic forecasting (sometimes also known as stochastic forecasting). Understanding and employing this relatively simple principle can take your forecasting and supply chain planning from “good” to “great.” This primer explains how.
Single-Number Forecasting and the Downsides of Traditional Forecasting
There are two ways to make a prediction.
The first is predicting that one specific thing will happen.
For example: the horse “Secretariat” will win the Kentucky Derby.
Since Secretariat is the most successful racehorse of all time, you might place a single bet on him to win. In the world of supply chain planning, this kind of prediction is called a ‘single number’ forecast.
With this, planners aided by simpler systems (often spreadsheets or legacy planning systems) forecast one number for a particular item.
Single number forecasting can work in those circumstances where you are confident that an established pattern will be repeated – such as with fast-moving, commodity items.
For example, you might have 3 years of history of selling 100 standard USB chargers every week, give or take a few. In this case, forecasting 100 units is a pretty safe bet.
Most products, of course, aren’t like that, just like most racehorses aren’t like Secretariat.
Even the most successful, healthy horses with able jockeys are subject to many unforeseeable variables that affect their actual outcome. They could have a collision, develop a sudden injury, or just have an off day.
Let’s stay with the horse race analogy here for a moment.
Serious gamblers often review the range of possible outcomes and then apply their own knowledge before settling on a bet. They may also place multiple bets to ‘hedge’ against losses that would result from making a single bet.
This scenario is somewhat analogous to probabilistic forecasting.
In supply chain planning, advanced algorithms are used to analyze multiple demand variables to identify the probabilities of a range of possible outcomes, one of which is the most likely.
It’s a much more reliable way to make predictions where demand patterns are variable, where there’s limited order history (as in the case of new product introduction), or when factors like seasonality come into play.
Even if aggregate weekly or monthly demand for an item stays relatively consistent, when you drill down into daily demand for that item by location, there is usually considerable volatility at this more granular level.
And in a distribution network, looking at aggregate demand is not enough. To meet service levels, you need a plan that ensures you get the right number of items to the right locations.
A probabilistic forecast that takes uncertainty into account helps you manage risk.
It’s not just about improving average demand predictions but assessing the entire range of possible outcomes—including demand volatility, which has the biggest impact on service levels.
With probabilistic forecasting, you still get one number that’s associated with the highest probability. However, banded around this number, you get a range of other possible outcomes, each with a different probability attached.
The Benefits of the Probabilistic Method
Returning to the horse racing analogy, betting on Secretariat is likely pretty safe.
But where there’s little risk, there’s also less reward.
Similarly, in business, there’s usually less upside in selling predictable commodity products.
Most thriving companies profit from carrying ‘long-tail’ products in their portfolio.
Viewed in this light, probabilistic forecasting is much more than a nerdy statistical method. It allows you to consistently place better inventory bets than your competitors for those harder-to-forecast items.
By freeing up working capital and improving service levels at the same time, this tried and tested approach can provide the sustainable competitive advantage to take your company from good to great.
And it can restore the trust in your forecasting.
When supply plans or safety stocks are based on wrong assumptions about demand uncertainty, targets go unmet and supply chains go into firefighting mode. Trust in the planning process erodes.
When planners stop trusting forecasts they usually err on the side of holding too much safety stock. This leads to excessive costs, waste and obsolescence.
Better to hedge your bet through probabilistic forecasting.