Sensitivity analysis asks how a conclusion changes when assumptions or uncertain inputs change. Define plausible alternatives, recalculate transparently and identify switching points. It reveals dependence on assumptions; it does not assign probabilities to scenarios unless those probabilities have a defensible basis.
Expose the inputs doing the decision work
Write the decision rule before exploring scenarios. The rule might prioritise the project with the lowest expected staff hours per resolved issue, subject to a minimum service requirement. List the inputs, units and source dates. Separate observations from estimates, policy choices and placeholders. A precise spreadsheet cell can still contain an uncertain judgement, and hiding that distinction makes the resulting ranking look stronger than it is.
Focus on inputs that can materially change the recommendation. An uncertain arrival rate may matter more than a small rounding convention. Ask operational owners for plausible ranges and the evidence behind them. Do not select ranges solely to make the preferred option win or lose. When no defensible range exists, state the gap and show clearly labelled illustrative values rather than presenting them as measured uncertainty.
Change assumptions in a way readers can reconstruct
Start with one input at a time to locate switching points, then examine combinations that are operationally coherent. Some quantities move together: additional demand may increase both workload and staffing cost. Holding one fixed while moving the other can create an impossible scenario. Record these dependencies and distinguish a diagnostic calculation from a feasible operating plan.
Show the base case and alternatives using the same units and time horizon. Preserve full calculation precision internally and round displayed results sensibly. A scenario range is not automatically a confidence interval, and three named scenarios do not establish their likelihood. If probabilities are used, explain where they come from and how the recommendation depends on them. Include consequences that are not captured by the arithmetic, such as access or service continuity.
Turn fragility into a useful next step
A recommendation is fragile when a modest plausible change reverses it. That finding helps decide whether to gather more evidence, choose a reversible option or adopt a rule that performs acceptably across scenarios. Identify the particular input whose resolution would most improve the decision. More data about an irrelevant field does not make a fragile assumption more secure.
Publish the switching condition alongside the recommendation. A stakeholder should know when the analysis needs review, not only which option currently ranks first. Assign an owner and review date to the uncertain input. Keep the scenario workbook or calculation path with its versioned assumptions so that later changes can be interpreted without silently rewriting the original basis of the decision.
A fictional queue-improvement choice
A fictional scenario to explain the method, not a customer result or a market benchmark.
A fictional service team compares two process changes for a four-week pilot. Option A requires 40 setup hours plus 0.1 hour per case. Option B requires 10 setup hours plus 0.4 hour per case. At 80 cases, A uses 48 hours and B uses 42, so B requires six fewer hours under the stated model.
Setting 40 plus 0.1n equal to 10 plus 0.4n gives 30 equal to 0.3n, so the switching volume is 100 cases. At 140 cases, A uses 54 hours and B uses 66, making A twelve hours lower. The expected volume is uncertain between 80 and 140, and neither option's service quality has been tested. The team therefore describes the hour ranking as volume-sensitive, checks recent eligible arrivals and pilots quality safeguards. It does not convert the illustrative volume range into a probability distribution or claim guaranteed savings from either untested change. It also checks that case volumes refer to completed eligible cases rather than arrivals, because unresolved carryover would make the assumed workload boundary inconsistent between the two options.
TAKE THIS INTO YOUR NEXT CONVERSATION
A practical next-step checklist.
- Separate observed inputs from assumptions.
- Justify ranges and record dependencies.
- Calculate where the recommendation switches.
- Assign follow-up to the most consequential unknown.
Common questions.
Is a low, central and high scenario a confidence interval?
No. Scenario choices need not have statistical coverage or known probabilities.
Should I vary every input equally?
No. Prioritise plausible changes that could alter the decision or expose a material risk.
Further reading & source context.
These primary references provide background. They are not evidence for the fictional example above and do not endorse Accuralytix.
Found an error or a missing qualification? Send an editorial correction.
Check my data readiness.
Use a practical tool to turn this idea into a clearer starting point.
Check my data readiness