Managing Production Risk When Demand Is Uncertain

The Problem

A pharmaceutical manufacturer was preparing production plans for high-value products with long lead times and strict manufacturing constraints. Production occurred in large batches, and changing course was costly and slow. At the same time, demand for the product varied over time.

The company’s operations team relied on traditional demand forecasts to guide production planning. These forecasts provided single predicted values for future demand, typically based on historical sales trends and recent market data. While the forecasts were often reasonable, they did not estimate the uncertainty surrounding those predictions.

This created a difficult situation for planners. A forecast might suggest that demand would increase over the next quarter, but the team lacked insight into how confident they should be in that prediction. If production was increased too aggressively, the company risked producing excess inventory. With limited shelf life, some product was discarded as waste. When production was too conservative, shortages could occur, disrupting supply to customers.

More accurate forecasts would not solve the problem. This company also needed to understand the uncertainty in the estimates.

QSC’s Modeling Approach

To address this, QSC’s modeling approach focused on forecasting the range of plausible demand outcomes, rather than producing a single point estimate. Using historical sales data and market indicators, we developed a probabilistic forecasting framework to estimate how demand might evolve over time. Hierarchical Bayesian modeling techniques were used to propagate uncertainty from all aspects of the data and model into estimates of predicted demand.

These forecasts allow the team to evaluate multiple possible demand scenarios and quantify the likelihood of each outcome. Instead of asking whether demand next quarter will be, for example, 10,000 units, planners can evaluate the distribution of projected demand.

This shift in perspective proved important. Production planning decisions can now be evaluated in terms of risk. For example, the team can estimate the probability that planned production levels will lead to shortages, or alternatively the probability that excess inventory will accumulate if demand falls below expectations.

Instead of point estimates of future demand, models provide a range of predictions with uncertainty estimates.

Decision Support in Practice

To support production decisions, the software presents the connection between demand uncertainty and operational constraints such as production schedules, equipment maintenance, and batch testing policies. These simulations help the team explore questions like:

• What is the probability that demand will exceed available inventory before the next production cycle?
• What is the shortage risk if sales increase by 10%?
• Under what conditions would increasing production by one batch meaningfully reduce shortage risk?

Rather than relying on a single forecast, the operations team can now see how different production strategies perform across a range of plausible demand outcomes.

This approach does not eliminate uncertainty. Demand for products will always involve factors that cannot be predicted precisely. What changed is the ability to reason about that uncertainty in a scientific way.

Outcome

With probabilistic forecasts and scenario analysis in place, the production planning process shifted from attempting precise predictions to managing operational risk more explicitly. Decisions about production increases, delays, or inventory buffers can be tied directly to the probability of different demand outcomes.

For the operations team, the most valuable outcome is clearer visibility into the risks associated with each production decision. Managers can focus on production plans on minimizing over- or underproduction risk.

This shift, from point forecasts to probabilistic demand modeling, allows the team to plan production with greater understanding.