Deciding Wisely When the Future Is Uncertain
Traditional planning assumes one future. Risk and decision analysis acknowledges multiple possible futures — and builds decisions that hold up across them.
Every consequential business decision is made under uncertainty. Revenue forecasts can miss. Costs overrun. Competitors act unexpectedly. The techniques in this set do not eliminate that uncertainty — they make it visible, quantifiable, and manageable.
The key insight: the goal is not to predict the future accurately. It is to make decisions that perform well across the range of futures that could plausibly occur — and to know which assumptions matter most.
simulations
decision trees
Scenario Analysis
Evaluates how outcomes change when multiple assumptions change simultaneously. Best Case · Base Case · Worst Case. Strategic planning, investment decisions, business continuity.
What-If Analysis
Tests the impact of a single specific hypothetical change. "What if costs increase by 15%?" Fast, simple, ideal for operational decisions and quick stakeholder briefings.
Sensitivity Analysis
Changes one variable at a time to measure its effect on the outcome. Identifies the critical assumptions — those that most influence the result.
Tornado Charts
Visualises sensitivity analysis results as horizontal bars, ranked by impact. The widest bar = the variable that matters most. Executive-friendly and immediately actionable.
Monte Carlo Simulation
Runs thousands of scenarios simultaneously, each with randomly sampled inputs from probability distributions. Produces a probability distribution of outcomes — the most rigorous uncertainty quantification technique.
Decision Trees (Expected Value)
Maps out decision alternatives and their consequences using probability-weighted payoffs. The Expected Value calculation selects the option with the highest probability-weighted outcome.
Risk Matrix
Qualitative/semi-quantitative tool for prioritising risks by Likelihood × Impact. Used for risk registers and rapid risk communication when full quantification is not feasible.
These techniques work best in sequence: Scenario Analysis (broad uncertainty exploration) → What-If Analysis (specific event testing) → Sensitivity Analysis (identify key drivers) → Tornado Chart (prioritise for stakeholders) → Monte Carlo (full probability distribution if needed). Each builds on the previous, moving from broad uncertainty to focused action.
Scenario Analysis
Not one future — three. Best Case, Base Case, and Worst Case reveal the range of possible outcomes.
Scenario analysis evaluates how outcomes change when multiple assumptions change simultaneously. Unlike sensitivity analysis (which changes one variable at a time), scenarios reflect the fact that real-world conditions move together. A recession doesn't just reduce revenue — it simultaneously increases costs, tightens credit, and changes competitor behaviour.
Ask: Which scenario worries management most? and Which assumptions drive the largest differences? These questions focus the conversation on what matters — not just the numbers, but the decisions those numbers imply.
The Scenario Development Process
| Variable | Best Case | Base Case | Worst Case |
|---|---|---|---|
| Annual Sales (units) | 15,000 | 10,000 | 7,000 |
| Selling Price ($/unit) | $110 | $100 | $95 |
| Unit Cost ($/unit) | $55 | $60 | $70 |
| Profit | $825,000 | $400,000 | $175,000 |
The range is $175K–$825K — a 4.7× spread. This tells management that the investment is profitable in all scenarios, but the level of profitability is highly uncertain. The conversation shifts: "Which scenario is most likely, and what could we do to prevent the worst case?"
Industry Applications
| Industry | Typical scenario variables | Primary use |
|---|---|---|
| Finance / Investment | Revenue growth, margins, discount rate, capex | DCF valuation, capital allocation |
| Operations | Demand volume, supplier costs, lead times, capacity | Capacity planning, inventory decisions |
| Telecommunications | Customer churn, data usage, network costs, energy costs | Network expansion, pricing strategy |
| Human Resources | Headcount growth, salary inflation, attrition rate | Workforce and compensation planning |
| Strategy | Market share, competitor response, regulatory environment | Market entry, M&A, business continuity |
What-If Analysis
What happens if X changes? Fast, specific, and instantly communicable to any audience.
What-if analysis examines the impact of a single, specific hypothetical change to one input variable. It answers the most common question in business planning: "What happens to our result if this one thing changes?"
Unlike scenario analysis (which changes multiple variables simultaneously) or sensitivity analysis (which systematically varies one variable across a range), what-if analysis tests a single specific event — making it the fastest and most accessible of the three techniques.
Current situation: Revenue = $1,000,000 · Costs = $700,000 · Profit = $300,000
Question: What if operating costs increase by 15%?
New Costs = $700,000 × 1.15 = $805,000
New Profit = $1,000,000 − $805,000 = $195,000
A 15% cost increase reduces profit by 35% (from $300K to $195K). This single calculation reframes the conversation: the business is highly sensitive to cost increases. What cost controls should be in place, and at what cost level does the business break even?
Common Business What-If Questions
| Department | What-If Question | Variable changed |
|---|---|---|
| Sales | What if sales fall by 10% in Q3? | Revenue volume |
| Procurement | What if fuel costs increase by 20%? | Input cost |
| Operations | What if supplier delivery times double? | Lead time / inventory holding cost |
| HR | What if travel expenses increase by 20%? | Departmental cost line |
| Finance | What if interest rates rise by 150 basis points? | Financing cost |
| Marketing | What if customer acquisition cost increases by 30%? | Marketing efficiency |
Quick to perform — a single calculation or formula change in Excel. Easy to explain — anyone can follow "if X changes to Y, then Z becomes W." Immediately actionable — the result either crosses a threshold that triggers action or it does not. Use what-if analysis for rapid operational decisions and for setting up the more detailed sensitivity analysis that follows.
Sensitivity Analysis
Change one variable at a time, hold everything else constant, and measure the impact on the outcome.
Sensitivity analysis systematically tests how the outcome changes as each input variable is varied across a defined range — one at a time, holding all others at their base values. This identifies the critical assumptions: those inputs where a small change produces a large change in the outcome.
The defining discipline of sensitivity analysis: change only one variable at a time. Keep all other variables at their base case values. This isolates the unique effect of each variable. If you change multiple variables simultaneously, you cannot determine which change drove the result — that is scenario analysis, not sensitivity analysis.
A project has Base Case NPV = $500,000. The analyst tests the sensitivity of NPV to sales volume changes, one level at a time:
| Sales Volume Change | NPV | Change from Base |
|---|---|---|
| −20% | $100,000 | −$400,000 (−80%) |
| −10% | $300,000 | −$200,000 (−40%) |
| Base (0%) | $500,000 | — |
| +10% | $700,000 | +$200,000 (+40%) |
| +20% | $900,000 | +$400,000 (+80%) |
Interpretation: A 10% change in sales volume produces a 40% change in NPV. Sales volume is a highly sensitive variable — management should monitor it closely and build contingency plans if early sales data falls below plan.
Common variables tested in sensitivity analysis
Demand & Price drivers
Revenue · Selling price · Customer demand volume · Market share · Price elasticity · Exchange rates
Cost drivers
Labour cost · Material cost · Energy costs · Financing / interest rates · Overhead allocation · Supplier costs
Test each variable across a consistent range (typically ±10%, ±20%, ±30%). Document all base-case assumptions clearly before running sensitivity. After completing all sensitivity tests, produce a tornado chart to rank the results by impact. The combination of sensitivity analysis + tornado chart is the standard risk communication package for executive audiences.
Tornado Charts
Which risk matters most? Tornado charts answer this visually — making sensitivity results instantly actionable for any audience.
A tornado chart visualises sensitivity analysis results as horizontal bars ranked from largest impact (top) to smallest (bottom). The resulting shape resembles a tornado — wide at the top, narrowing toward the base. The widest bar at the top is the variable that most influences the outcome. Management should focus attention and mitigation effort there first.
Reading the chart: Sales volume has the widest bar — it dominates the outcome. A focus on volume risk mitigation (pipeline monitoring, contingency capacity) delivers far more protection than addressing utilities (narrowest bar). Selling price is the second priority.
Executive-friendly: A bar chart ranked by size is instantly legible at every level of the organisation. Fast interpretation: The top 2–3 variables are obvious at a glance. Prioritises resource allocation: Management effort should be proportional to bar width — invest heavily in monitoring and mitigating the top 2 variables, less so in the bottom 2.
Monte Carlo Simulation
Instead of three scenarios, run ten thousand. The result is a full probability distribution of possible outcomes.
Monte Carlo simulation replaces fixed input values with probability distributions — each input is described not as a single number but as a range with associated probabilities. The simulation then randomly samples from each distribution thousands of times, calculating the output for each combination. The resulting collection of outputs forms a probability distribution of the outcome.
Instead of "profit will be $400K in the base case," Monte Carlo gives you: "there is an 80% probability that profit will be between $280K and $620K, and a 10% probability that profit will fall below $200K."
The Monte Carlo Process
A retail chain models first-year profit for a new location. Three uncertain inputs are each assigned triangular distributions:
| Input | Minimum | Most Likely | Maximum |
|---|---|---|---|
| Customer Volume (daily) | 180 | 260 | 340 |
| Average Spend per Visit ($) | $22 | $31 | $42 |
| Operating Costs (annual $K) | $580K | $650K | $790K |
After 10,000 iterations: Mean profit: $418K · P10: $142K · P90: $695K · P(loss): 8.3%
The 8.3% probability of a loss is the number that matters for the board's risk appetite. If the organisation's threshold is "no more than 5% chance of a loss," this project is marginal — and the analysis has quantified exactly how marginal.
| Distribution Type | When to use | Parameters |
|---|---|---|
| Normal | Variables with symmetric uncertainty around a mean; measurement errors | Mean, Standard Deviation |
| Triangular | When you have expert estimates of min, most likely, and max — most common in business | Minimum, Mode, Maximum |
| Uniform | Equal likelihood across a range; when you truly have no information beyond bounds | Minimum, Maximum |
| Lognormal | Variables bounded at zero (prices, costs, durations); right-skewed distributions | Mean and SD of the log |
| Discrete | When the variable takes specific values with known probabilities | Value-probability pairs |
Decision Trees & Risk Matrices
Mapping choices, probabilities, and payoffs — then choosing the path with the highest expected value.
Decision Trees — Expected Value Analysis
A decision tree maps out decision alternatives, chance events, and their outcomes. Expected Value (EV) is the probability-weighted average payoff: the sum of each outcome multiplied by its probability. Decision-makers choose the option with the highest EV — unless risk tolerance dictates otherwise.
Interpretation: EV of entering = (0.4×$800K) + (0.4×$200K) + (0.2×−$300K) = $340K. Since $340K > $0 (don't enter), the expected-value maximising decision is to enter. But note: there is a 20% chance of losing $300K. If the organisation cannot absorb that loss, it might rationally choose not to enter despite the positive EV.
Risk Matrix — Likelihood × Impact
For risks that cannot be fully quantified, a risk matrix provides a structured qualitative assessment. Each risk is rated on Likelihood (1–5) and Impact (1–5). Risk Score = Likelihood × Impact. Risks in the high-score zone require immediate mitigation; low-score risks are accepted or monitored.
| Risk Score (L×I) | Zone | Action required |
|---|---|---|
| 15–25 | 🔴 Critical | Immediate action — mitigation or contingency plan required before proceeding |
| 8–14 | 🟡 High | Active monitoring and mitigation plan. Escalate to senior management |
| 4–7 | 🔵 Medium | Standard monitoring. Include in risk register. Review quarterly |
| 1–3 | 🟢 Low | Accept and document. Review if circumstances change |
Microsoft Excel
Scenario Manager, Data Tables, Goal Seek — everything you need for risk analysis without any add-ins.
The Scenario Manager stores multiple named scenarios (Best/Base/Worst) with different input values and produces a summary table comparing all outcomes at once.
Data Tables calculate how a formula result changes across a range of input values automatically — far faster than manually changing inputs one at a time.
Goal Seek finds the input value needed to achieve a specific target output. The reverse of a normal calculation: "What sales volume do we need to break even?"
Python — Scenario, Sensitivity & Monte Carlo
Automate all risk analysis techniques — from scenario tables to full Monte Carlo simulations with thousands of runs.
For analysts who prefer Excel over Python, @RISK (Palisade) and Crystal Ball (Oracle) are industry-standard Excel add-ins that integrate Monte Carlo simulation directly into spreadsheet models. You define distributions by right-clicking cells and selecting distribution types — no formulas or code. They produce output histograms, S-curves, and sensitivity (tornado) charts automatically. Widely used in finance, engineering, and consulting.
Three Simulators
Build intuition for scenario analysis, sensitivity analysis, and Monte Carlo through live interaction.
Simulator 1 · Scenario Builder
A telecom company faces declining profitability. Build three scenarios by adjusting the five key variables. Watch how profit changes under each scenario.
Simulator 2 · What-If Calculator
Starting from a base business model, apply specific what-if changes and see the immediate profit impact.
Simulator 3 · Monte Carlo Explorer
Run a live Monte Carlo simulation for a product launch. Adjust the uncertainty on key inputs and watch how the profit distribution changes.
Common Mistakes & Knowledge Check
Six mistakes that undermine risk analysis — followed by five assessment questions.
Presenting a single-point forecast without scenarios
"Revenue will be $4.2M next year." Without a range or confidence interval, this implies false precision. Any business plan based on a single number is a plan without risk awareness. Always accompany forecasts with a spread of scenarios.
Building scenarios that are not internally consistent
A "Worst Case" where demand falls 30% but prices stay flat is not a coherent scenario — in a real downturn, pricing pressure typically increases too. Scenarios must reflect how variables actually co-move in the real world.
Confusing sensitivity analysis with scenario analysis
Sensitivity analysis changes one variable at a time. Scenario analysis changes multiple variables simultaneously. Mixing them produces results that cannot be cleanly interpreted. Maintain discipline: one variable at a time for sensitivity; coherent multi-variable stories for scenarios.
Treating the Expected Value as the "most likely" outcome
Expected Value is the probability-weighted average outcome — not the most probable one. In the market entry example, EV = $340K but the most likely individual outcome may be either $800K or −$300K. Communicate this distinction clearly to stakeholders.
Assigning distributions in Monte Carlo without justification
Picking "Normal" for every variable because it is familiar is not analysis — it is assumption. Each input distribution should reflect the actual uncertainty characteristics of that variable. Costs are often lognormal; subjective estimates are often triangular. Document the choice.
Never updating the analysis as new information arrives
A scenario built in January with January's assumptions becomes stale as the year progresses. Risk analysis must be a living tool — re-run when key assumptions change significantly. A static risk register reviewed once a year is a compliance exercise, not a management tool.