Decisions Under Uncertainty: An Introduction

Every Decision Is a Bet

When a supply chain manager sets a reorder point, they’re betting on future demand. When a logistics director commits to a carrier contract, they’re betting on volume and rates. When a warehouse planner decides on staffing levels, they’re betting on throughput requirements.

Most people don’t think of these as bets. They think of them as plans, decisions, or standard operating procedures. But strip away the language, and every forward-looking business decision is a wager placed under incomplete information. The only question is whether you’re making informed bets — with eyes wide open about the odds — or blind ones.

This distinction is the foundation of everything we believe about decision-making in supply chain and logistics.

The Spectrum of Uncertainty

Not all uncertainty is created equal. It helps to think about a spectrum:

Known Certainty

You know exactly what will happen. This almost never exists in business, but it’s the implicit assumption behind most planning tools. When your MRP assumes a fixed lead time of 14 days, it’s treating an uncertain quantity as known.

Statistical Uncertainty

You don’t know exactly what will happen, but you have enough historical data to build a reliable probability distribution. Demand for a mature product with years of sales history falls here. You can’t predict next month’s exact demand, but you can characterize the distribution well enough to make quantitative risk decisions.

Deep Uncertainty

You don’t have enough data or structural knowledge to build a reliable distribution. New product launches, market disruptions, geopolitical events, and technology shifts live here. You might be able to define plausible scenarios, but assigning precise probabilities is questionable.

True Ambiguity

You don’t even know the full set of possible outcomes. “Unknown unknowns.” A global pandemic shutting down ports. A canal blockage halting 12% of world trade. These events are hard to plan for directly, but you can build systems that are resilient and adaptive in the face of surprise.

Most supply chain decisions live in the statistical uncertainty zone, which is good news — it means probabilistic methods work. But the best decision frameworks also account for the deeper uncertainties, primarily through robustness and adaptability rather than precise optimization.

Key Concepts for Decision-Making Under Uncertainty

Expected Value Is Not Enough

Expected value — the probability-weighted average outcome — is the most common metric for evaluating decisions under uncertainty. It’s also dangerously incomplete on its own.

Consider two options:

• Option A: 100% chance of earning $100,000
• Option B: 50% chance of earning $250,000, 50% chance of losing $40,000

The expected value of Option B ($105,000) is higher. But if that $40,000 loss puts you out of business, Option B is a terrible choice regardless of its expected value. Expected value ignores the shape of the distribution and your ability to absorb bad outcomes.

This is why we always evaluate decisions across the full distribution of outcomes, not just the mean.

Risk Is About the Downside

In everyday language, “risk” is vague. In decision science, risk should be precise: it’s about your exposure to outcomes you can’t afford. Value at Risk (VaR), Conditional Value at Risk (CVaR), probability of exceeding a threshold — these are the metrics that matter when the stakes are real.

A supply chain that looks great on average but has a 10% chance of catastrophic stockout during peak season is carrying hidden risk. Quantifying that risk explicitly — not burying it in an average — is the first step toward managing it.

Robustness vs. Optimality

An optimal solution is the best answer for one specific set of assumptions. A robust solution performs acceptably well across a wide range of assumptions. In uncertain environments, robustness almost always beats optimality.

Why? Because the assumptions underlying the “optimal” solution are wrong — guaranteed. The question is how far wrong. A robust solution sacrifices a small amount of performance in the best case to avoid catastrophic performance in realistic off-cases.

Adaptability: The Ultimate Hedge

The best strategies aren’t just robust at a point in time — they’re adaptive over time. They incorporate new information and adjust. They don’t lock in decisions further out than necessary. They maintain optionality.

A fixed annual plan is neither robust nor adaptive. A rolling policy that replans weekly based on updated data is both. The value of adaptability increases with the level of uncertainty — the less you know about the future, the more valuable it is to keep your options open.

The Bit Bros Approach

Our philosophy on decisions under uncertainty comes down to four principles:

1. Model Uncertainty Explicitly

Don’t hide uncertainty behind point estimates. Use probability distributions for demand, lead times, costs, and any other uncertain quantity. Make the uncertainty visible and quantifiable.

2. Evaluate Policies, Not Just Plans

A plan is a specific sequence of actions. A policy is a rule for choosing actions based on the current state and information. Policies adapt; plans break. Design and evaluate policies, testing them against the full range of uncertainty.

3. Surface Tradeoffs for Decision-Makers

Don’t hand stakeholders a single “optimal” recommendation. Show them the tradeoff curves: “Here’s the cost-service tradeoff. Here’s how risk changes as you move along it. Here’s what you gain and what you give up.” Let humans make the value judgments — that’s what humans are for. Let the analytics illuminate the landscape of choices.

4. Build Systems That Adapt

Static solutions decay. The environment changes, data drifts, new patterns emerge. Build decision systems with feedback loops: monitor performance, detect when the environment has shifted, retune or redesign policies when needed. The system should get better over time, not just run until someone remembers to update it.

A Practical Takeaway

Here’s something you can apply immediately, regardless of your tools or technical sophistication:

For your next major supply chain decision, don’t just present the expected outcome. Present three scenarios: a realistic upside, the base case, and a realistic downside. For each, show the outcome of the proposed decision. If the downside scenario produces an outcome you can’t live with, your decision needs to change — no matter how good the base case looks.

This simple exercise — asking “what if I’m wrong, and can I survive it?” — is the first step from blind bets to informed ones. Everything else we do is a more rigorous, scalable, automated version of that same question.