When Averages Fall Short: Understanding Uncertainty Through the Work of Dr. Sam Savage
Dr. Sam L. Savage, Executive Director of ProbabilityManagement.org, has spent much of his career encouraging leaders to look more closely at how uncertainty is represented in everyday decisions. His work doesn’t suggest that averages are meaningless, but that they often tell only part of the story when outcomes depend on variation, timing, or interconnected events.
“In many organizations, people are asked to compress uncertain futures into a single number, even when that number cannot fully reflect what might unfold,” he states. “The moment you’re forced to give one number in an uncertain world, much of the information may be lost.”
His work, therefore, points toward a more intuitive, human-centered way of thinking about uncertainty. It’s one he believes has existed for decades but has often been obscured by intimidating mathematics and specialized tools.
That perspective is grounded in a diverse and deeply technical background. Savage earned his PhD in computational complexity and began teaching management science at the University of Chicago before joining Stanford University, where he continues to serve as an adjunct lecturer in Civil and Environmental Engineering.
Across these settings, he noticed a recurring challenge that even when analytical models were sound, they often failed to shape decisions in practice. According to Savage, the barrier was communication. “If people can’t explain uncertainty to one another,” he says, “they can’t act on it together.”
These observations crystallized in his book The Flaw of Averages, where Savage introduced the idea that plans built on average assumptions can behave very differently from what those averages suggest.
“One of the illustrations I usually give is a statistician trying to cross a river that is, ‘on average,’ 3 feet deep,” he shares. “The average sounds comforting, but it hides the variation beneath the surface. Some parts of the river are much deeper than others. The statistician doesn’t face risks because the math is wrong, but because the math describes the river in a way that omits what matters most for survival.”
Savage brings this example into familiar planning scenarios. He says, “Consider a project that can’t begin until several approvals are complete. Each approval has an expected timeline, and leadership asks for a single start date. Plugging in average durations produces a clean answer, but it assumes that everything finishes on schedule at the same time.” Given his experience spanning energy firms, defense contractors, and the healthcare industry, Savage notes that what gets lost isn’t effort or competence but how uncertainty compounds across dependencies.
This idea is at the heart of ProbabilityManagement.org. Its mission is to make uncertainty actionable by treating it as data that can be stored, shared, and combined. Savage refers to this work as developing the “Hindu–Arabic numerals of uncertainty.”
Just as standardized numerals once transformed commerce and science, stochastic data, which expresses the language of chance, allows uncertainty to move through spreadsheets, models, and dashboards without being flattened into a single value. By preserving relationships across possible futures, uncertainty can be explored more fully rather than reduced prematurely.
These same principles extend into modern AI systems. Savage describes this as the stochastic data cycle. Stochastic data, he argues, is used to train most AI models, embedding knowledge about variability rather than just point outcomes. When results come back from the AI, the system may still be asked for a single number. And, if pressed, the AI will comply, often offering an average, such as the expected result of a die roll, which is a meaningless three and a half. “AI tends to be obedient,” Savage notes. “If you ask for one number, it will give you one.” The difference lies in whether the underlying stochastic data is preserved or discarded. When retained, it can flow into chance-informed applications, keeping uncertainty intact as decisions are made.
Savage expanded this thinking in his second book, Chancification, which describes how organizations can routinely work with chance. The idea mirrors other technological shifts: experts generate the complexity, while everyday users interact with it intuitively.
ChancePlan.AI, Savage’s commercial venture, applies this approach by linking stochastic models into coherent planning systems, using open standards developed by the nonprofit. The relationship between the two reflects a consistent philosophy that shared infrastructure enables flexible application.
Underlying all of this is Savage’s attention to how people actually learn. Experiences ranging from flying gliders to composing music shaped his belief that understanding emerges when intellect and intuition are connected. Interactive models that update instantly engage both.
Overall, the conversation about averages is about expansion. Averages can summarize, but uncertainty explains. By giving uncertainty a language people can share, across teams, tools, and even AI systems, Savage’s work creates space for clearer planning and more grounded decisions. Or, as he says, “When you can talk about chances without fear, better conversations tend to follow.”
