Image (Midjourney)
As hurricane season approaches, emergency managers across coastal regions face a familiar and urgent question: how much is enough? Deciding how many truckloads of bottled water, ready-to-eat meals, and medical supplies to stockpile—and where to send them—can mean the difference between communities getting the help they need or going without.
Estimate too low, and critical shortages can leave families without clean water or care. Estimate too high, and limited budgets are drained by unused supplies that may expire or go to waste.
For decades, planners have relied on two dominant approaches to make these decisions: one uses historical data to predict future demand, while the other prepares for worst-case scenarios. In practice, each can produce undesirable results when its assumptions do not match the operating environment.
Karmel S. Shehadeh, a WiSE Gabilan assistant professor in the Daniel J. Epstein Department of Industrial and Systems Engineering at the USC Viterbi School of Engineering, has developed a framework to help decision-makers navigate this uncertainty more effectively.
“Decisions should account for uncertainty, but not overreact to rare extremes,” Shehadeh said, emphasizing the need to balance data-driven resource allocation and planning with caution.
Her paper, “The Trade-Off Between Optimism and Pessimism in Multi-Resource Planning and Allocation under Demand Uncertainty,” was recently published in the European Journal of Operational Research.
Why traditional planning approaches fall short
One of the most widely used methods, stochastic programming, optimizes resource planning and allocation decisions using an estimated probability distribution of demand, often constructed from historical data or simulated demand scenarios.. Although this approach is grounded in available demand information, it assumes that the estimated demand distribution will remain representative of future demand. When actual demand differs, the resulting plan can be optimistically biased and perform poorly in practice, a growing concern in an era of increasing uncertainty.
“When reality diverges from what we’ve seen before, whether it’s a more severe storm, a longer drought or a sudden surge in demand, these models can underestimate what’s actually needed,” Shehadeh said. “The result is a plan that looks good on paper but may not perform well in practice.”
At the opposite end of the spectrum is distributionally robust optimization, which optimizes decisions against the worst-case distribution within a set of plausible demand distributions. Although this provides stronger protection against distributional uncertainty, it can lead to overly conservative plans and higher procurement costs.
“The worst-case distributions these models hedge against tend to emphasize a small number of extreme outcomes that are rarely observed in practice,” Shehadeh said.
In settings such as food banks, this could mean overstocking perishable goods that spoil before they can be used. In healthcare staffing applications, it could translate to overstaffing during quiet periods while still struggling to respond effectively to actual demand fluctuations.
A flexible framework for real-world uncertainty
Shehadeh’s model, called TRO-SRAP, allows decision-makers to move between optimism and caution rather than choosing one or the other. At the center of the framework is a trade-off parameter, theta, which determines how much emphasis the model places on performance under the available demand data versus protection against the worst-case distribution in a chosen set of plausible distributions.
At one end, the model fully trusts the available demand data; at the other, it plans for the worst case. In between, it blends both perspectives. For example, a theta value of 0.4 places 60% of the weight on expected performance under the empirical demand distribution and 40% on the worst-case expected performance. .
Rather than producing a single “best” plan, the model generates a spectrum of options, each representing a different level of robustness. “Decision-makers can select a procurement and allocation policy that best aligns with their operational settings,” Shehadeh said.
This flexibility is valuable across different operating environments. When planners expect demand to remain close to the distribution represented by the available data, lower theta values may be appropriate. When substantial demand shifts or extreme outcomes are plausible, higher theta values provide greater protection. Intermediate values can offer a useful balance under moderate shifts.
“The key advantage is that you’re no longer locked into one extreme,” Shehadeh said. “Decision-makers can select a level of conservatism that reflects their operating environment and tolerance for uncertainty.” .
Tested across disaster response and food systems
To evaluate performance, Shehadeh tested the model using two real-world case studies based on data from disaster preparedness and food bank operations.
The results showed that solutions based on intermediate theta values often achieved lower out-of-sample total costs than classical stochastic, robust, or distributionally robust approaches, especially under moderate distributional shifts. These savings resulted from balancing procurement costs with penalties associated with shortages and unused resources
.“By considering a range of possible scenarios, we can achieve solutions that are both cost-effective and resilient,” Shehadeh said.
Expanding applications and future impact
Looking ahead, Shehadeh plans to extend her framework to more complex planning problems and develop user-friendly tools that make it accessible to practitioners without specialized training.
“The framework has a very broad range of applications,” she said. “It can be used in disaster response, healthcare, transportation, supply chains and many other areas where decisions must be made under uncertainty.”
Beyond this study, Shehadeh’s broader research examines how to better model uncertainty in real-world optimization problems and incorporate fairness considerations into decision-making, particularly when vulnerable populations may be affected.
Ultimately, her goal is to help organizations move beyond rigid planning strategies and make more informed, balanced decisions.
“I want to help people make better decisions that improve access to critical resources while reducing unnecessary waste,” Shehadeh said. “That’s especially important in situations where resources are limited, and the stakes are high.”
Published on August 4th, 2026
Last updated on August 4th, 2026