Sensitivity Analysis Should Change the Decision, Not Decorate the Appendix

The purpose of sensitivity analysis is to discover where the decision breaks, not to demonstrate that someone varied three spreadsheet cells.

A business case can look precise while resting on uncertain foundations. Demand forecasts, implementation cost, benefit timing, adoption, operating savings, residual value and discount-rate assumptions may all be expressed to two decimal places, but precision in presentation is not the same as confidence in the underlying estimate.

Sensitivity analysis is intended to expose that difference.

The Australian Government's 2006 Handbook of Cost-Benefit Analysis describes sensitivity analysis as a key general technique for examining uncertainty. It recommends testing the effects of plausible better and worse assumptions on net present value and also discusses more extensive risk analysis where several variables are materially uncertain. The same handbook treats sensitivity testing as one way to uncover optimism bias.

The strategic implication is stronger than “include a sensitivity table”. Leaders should use the analysis to identify the variables that can change the choice, determine what evidence is worth buying, and decide whether commitment should be staged.

The Strategic Context

Most material investment decisions are made before uncertainty disappears.

An infrastructure project is approved before final construction cost is known. A digital platform is funded before adoption can be observed at scale. A manufacturing line is automated before actual throughput, downtime and maintenance performance are proven. A new service is launched before demand is fully visible.

That does not make the decisions irrational. It means leadership must understand the shape of uncertainty rather than hide it behind a point estimate.

The failure mode appears when a base case becomes psychologically privileged. Once the model displays a single NPV, payback period or benefit-cost ratio, discussion gravitates towards that number. The assumptions become background.

A decision-grade analysis reverses the emphasis. The number is an output. The assumptions, causal relationships and thresholds are where governance attention belongs.

What Leaders Commonly Misread

The first misconception is that sensitivity analysis is synonymous with pessimistic and optimistic scenarios. Three scenarios may be useful, but they can still conceal the variables that actually drive the decision.

The second is that every variable deserves equal attention. Changing dozens of inputs by arbitrary percentages can produce an impressive appendix while creating little insight. The important variables are those that are both uncertain and decision-material.

The third is that uncertainty should be handled by simply increasing the discount rate. The 2006 Handbook cautions that loading the discount rate is a crude way to assess risk because it affects all future benefits and costs regardless of which items are actually uncertain. Sensitivity analysis is more transparent because it allows decision-makers to see what is moving.

The fourth is that sensitivity analysis is only financial. A project may remain NPV-positive while becoming operationally unacceptable because implementation exceeds a shutdown window, adoption drops below a minimum viable level, safety assurance fails, or a regulatory approval is delayed beyond a strategic deadline.

The fifth is that the output should be reassurance. In fact, a good sensitivity analysis may make the decision less comfortable. That is its value.

Reframing the Issue

Sensitivity analysis should answer three executive questions.

What must be true? Identify the assumptions required for the preferred option to remain attractive.

Where does the decision change? Find the switching values at which the ranking, recommendation or viability changes.

What should we do about that uncertainty? Decide whether to obtain more evidence, redesign the investment, stage commitment, create contingency or accept the exposure deliberately.

This turns sensitivity analysis into an instrument of governance.

Strategic Analysis: Find the Decision-Critical Assumptions

Start with causal drivers, not spreadsheet convenience

Variables are often chosen for sensitivity testing because they are easy to change. That is backwards.

Begin with the value logic. What creates the benefit? What creates the cost? What determines timing? What determines adoption? Which dependencies could interrupt the chain?

For a hypothetical warehouse automation investment, the important uncertainties might be transaction volume, labour redeployment, integration downtime and equipment reliability. A generic ±10 per cent test on every cost line may tell leaders much less.

Distinguish uncertainty from vulnerability

Mesly's feasibility work draws a useful distinction between external risks and internal vulnerability. Applied to investment analysis, the same principle helps leaders move beyond forecasting.

Demand uncertainty is external. A business model requiring one narrow demand forecast to succeed is an internal vulnerability.

Supplier delay is external. A project architecture with no viable second source and no schedule contingency is a vulnerability.

Regulatory change is external. A strategy that becomes uneconomic after one modest regulatory change is a vulnerability.

Sensitivity analysis should therefore test not only what might vary, but also whether the design is unnecessarily exposed to that variation.

Look for switching points

A percentage range is less useful than a threshold when leaders need to make decisions.

At what installed cost does the preferred option cease to outperform the alternative?

How many months of delay erase the expected advantage?

What adoption rate is required to break even?

At what demand level should a staged investment expand?

What minimum benefit realisation justifies continuing the next tranche?

Switching values translate uncertainty into governance conditions.

Treat optimism bias as a system problem

The 2006 Handbook describes optimism bias as favourable estimates being presented as likely estimates, through overstated benefits or understated costs. It suggests remedies including sensitivity analysis, clearer assumptions and, in some cases, independent expert assessment.

The governance lesson is that optimism is rarely fixed by telling teams to “be more realistic”. Incentives matter.

Sponsors want approval. Vendors want contracts. delivery teams want momentum. Executives may want a strategic announcement to remain credible. Analysts are therefore operating inside a system that can reward favourable assumptions.

Independent challenge, transparent assumptions and pre-agreed evidence standards reduce this pressure.

Use ranges where the model justifies ranges

The correct response to uncertainty is not always a probability distribution. In many strategic decisions, there is not enough evidence to assign credible probabilities.

Leaders can still use bounded scenarios, break-even analysis, stress tests and explicit confidence judgements. The goal is not mathematical sophistication. It is to make the uncertainty visible at the level required for the decision.

Decision Framework

A useful sensitivity process can be organised into five stages.

Stage	Executive purpose
1. Map the value logic	Identify the assumptions connecting investment to benefits and costs
2. Rank uncertainty	Determine which assumptions are least evidenced
3. Rank materiality	Determine which assumptions can materially change the decision
4. Find thresholds	Calculate where recommendation, affordability or viability changes
5. Govern the response	Test, stage, hedge, redesign, defer or accept

The intersection of high uncertainty and high decision materiality is where leadership attention should concentrate.

This also supports a value-of-information decision. Not every uncertainty deserves more research. If an assumption can vary widely without changing the choice, further analysis may have little value. If a small change can reverse the recommendation, better evidence may be worth significant time and money.

From Strategy to Execution

Immediate action

Require investment papers to identify the five most decision-critical assumptions, not merely attach a generic sensitivity table.

For each assumption, show:

base assumption;
plausible range;
evidence source;
switching threshold;
owner;
next opportunity to update the estimate.

Where the threshold sits close to the base case, make that visible in the recommendation.

Medium-term capability building

Build independent challenge into material investments. This does not require a permanent internal audit exercise. It does require someone with authority to question demand, cost, benefit timing and implementation assumptions without being dependent on approval of the project.

Track forecast accuracy after approval. Compare estimated and actual cost, schedule, demand, adoption and benefits. Organisations cannot improve optimism bias if they never close the feedback loop.

Develop scenario templates that reflect the organisation's recurring uncertainties rather than arbitrary percentages.

Use decision gates to refresh sensitivities as evidence matures. A business case should become more informed over time, not merely more detailed.

Long-term strategic positioning

The strongest portfolios are designed to be robust rather than merely forecast to succeed.

That may mean modular architecture, staged capacity, diversified suppliers, contractual options, pilots, reversible commitments or explicit exit points. These features can reduce exposure to uncertain variables even if they slightly reduce the theoretical maximum return in the base case.

Strategic resilience often comes from sacrificing a small amount of optimised upside to protect against a large amount of downside.

Signals to Monitor

Watch for:

sensitivity tables where every variable uses the same arbitrary percentage;
base cases sitting just above an approval threshold with little discussion of switching values;
benefits that depend on adoption rates no one is accountable for measuring;
projects repeatedly using cost contingencies without improving estimate accuracy;
material assumptions with no named evidence source;
independent reviewers engaged only after the preferred option is effectively committed;
risk analysis that changes discount rates but leaves causal assumptions untouched;
scenario labels such as “pessimistic” and “optimistic” without explaining what drives them;
project teams unable to state which assumption would cause a stop, redesign or defer decision.

These are signs that uncertainty is being displayed rather than governed.

Questions for the Leadership Team

Which three assumptions can actually reverse this decision?
How far is the base case from each switching threshold?
Which uncertainty is worth resolving before we commit, and which can we safely carry?
Are our ranges evidence-based or chosen for convenience?
What design change would make the investment less sensitive to the most uncertain variable?
Are sponsor incentives encouraging optimistic assumptions?
When will we next have better evidence, and can commitment be staged until then?

Closing Perspective

Sensitivity analysis is not a technical appendix to a decision already made. It is a way of discovering the conditions under which the decision remains valid.

Used well, it tells leaders what must be true, what could break, what evidence matters and where a staged commitment is smarter than a full one. It exposes optimism without pretending uncertainty can be eliminated. It also shifts attention from forecast accuracy alone to investment robustness.

The critical question is not whether the spreadsheet changed when the assumptions changed.

It is whether leadership would change the decision.