Probabilistic Planning in Business Forecasting

Line Art

How supply chains mitigate risk with probabilistic planning

Business forecasting has historically been a field defined by error.  Forecast errors that can lead to probabilistic planning, can haunt planners years into the future, when even small discrepancies between forecasts and actuals can cost companies significant amounts of resources and lost profit.  Planners who employ point-forecast techniques are using a deterministic planning method, which more or less places all bets on one number.  This is a plan with great risks; any variance between forecasts and actuals come at a cost due to either under or over-production of goods.

Better suited, especially in supply chain planning, is a probabilistic planning model, which accounts for uncertainty of both demand and supply, and gives a range of potential outcomes (quantified by probability of occurrence).  When planners use probabilities in their forecasting models, they see an immense uplift in the usability of their short-, mid-, and long-range projections.

Figure 1:  Representation of historical sales data flowing into a probabilistic forecast for the upcoming year

Probability Distribution

Probability distribution is a superior approach to modeling uncertainty in business forecasting.  It can determine the likelihood, and conversely, the risk levels associated with each potential future outcome.  In the example above, we can see that in April of the upcoming year, the following potential sales numbers are quantified by likelihood of occurrence:

Number of Units Projected to SellLikelihood of OccurrenceAssociated Risk if Committed
500%100%
4550%50%
3795%5%
3050%50%
250%100%

Figure 2:  In this scenario, we commit to building 37 units for April sales knowing this value has the least associated risk level

Building a Probabilistic Forecast

There are various approaches to deriving a probability distribution, either from historical data or using sales team insight.  We will focus on the statistical models, including:

Coefficient of Variance (CoV):  SKUs are categorized into high-stability/low-risk or low-stability/high-risk using the amount of historical variance between forecast and actuals.

Forecast decomposition:  extracting the baseline, trends, seasonality, cycles, and residuals from a time series forecast to determine the lowest risk component (baseline) and the highest risk component (residuals).

Better Planning with Probabilities

probabilistic approach to business planning allows decision-makers to understand associated levels of risk with each commitment.  In the example in Figure 2, the planners would know that committing to 45 or more orders for April would come with a 50% risk of overproduction.  This analysis enables supply chain leaders to evaluate the increasing levels of risk on both sides of the baseline forecast.  This means committing to quantities with confidence and optimizing inventory levels across the business.