What Is Sensitivity Analysis

Image
What Is Sensitivity Analysis
Learn what sensitivity analysis is, with a clear definition, the main types, a worked example, Excel steps, and how it applies in SAFe.
Blog Author
Published on
Nov 17, 2026
Views
2223
Read Time
8 Mins
Table of Content

Every plan rests on numbers you're not sure about. The useful question isn't whether they're right; it's which ones would change your decision if they were wrong. I'm a leading SAFe certification trainer at StarAgile, and I teach teams to ask that question before they commit budget. In this guide, I'll explain the definition, the steps, the types, examples, and how it applies in SAFe.

What Is the Sensitivity Analysis Definition?

The standard sensitivity analysis definition is a method for measuring how changes in one or more input variables change an output, under a clearly stated set of assumptions. In plain terms, you take a model, nudge its inputs one by one, and watch what happens to the result. If a small nudge to one input swings the answer wildly, that input deserves your attention. If a big nudge barely registers, you can stop worrying about it.

What Questions Does It Answer?

  • Which input has the biggest effect on the outcome?

  • How far can a number move before my decision changes?

  • Where would better data be worth paying for?

  • How fragile is this plan or model overall?

Where Is It Used?

The sensitivity analysis definition is the same everywhere, but the use cases differ:

  • Finance: testing how a bond price responds to a 1% rise in interest rates

  • Engineering: checking how a design behaves when tolerances shift

  • Environmental modelling: studying complex systems, such as climate models, where the exact link between inputs and outputs isn't well understood

  • Project management: finding which cost or time assumption could sink a budget

  • Product and portfolio decisions: seeing which estimates drive a ranking

If a decision rests on a forecast, an estimate, or an assumption, this kind of testing has something to offer.

How Do You Perform Sensitivity Analysis Step by Step?

You perform it in six steps: define the model and outcome, pick the key inputs, set ranges, choose a method, run it, and then rank and communicate the results.

What Are the Six Steps?

  1. Define the model and outcome: Decide what you are measuring, such as total cost, profit, or delivery date.

  2. Pick the key inputs: Choose the assumptions that are uncertain or likely to matter.

  3. Set a range for each: Give every input a low, base, and high value, or a probability distribution if you have the data.

  4. Choose a method: Pick a type that suits your model (see the next section).

  5. Run it and record the results: Change the inputs systematically and note the output each time.

  6. Rank, communicate, and decide: List inputs by impact, share the findings, and decide where to firm up data or add buffers.

For example, if your base assumption for monthly sign-ups is 1,000, you might test 800 as the low case and 1,200 as the high case, then record the profit each time. Treat it as a continuous loop. When new data arrives or the model changes, run it again.

 
 
 
 
Enroll in Leading SAFe Certification Today!

What Mistakes Should You Avoid?

  • Testing too many inputs at once, which buries the few that matter
  • Using ranges that are far too narrow or wildly unrealistic
  • Ignoring inputs that move together, such as price and sales volume
  • Forgetting to record the base case you are comparing against
  • Showing a chart with no recommendation attached
  • Treating the results as a forecast instead of a guide to where the risk sits

How Do You Do It in Excel?

In Excel, the tools sit under Data, then What-If Analysis. CFI's guide describes two ways to set up the inputs:

Approach

How It Works

Best For

Direct method

Substitute different numbers straight into an assumption, such as 0%, 5%, and 10% growth

Quick checks on a single driver

Indirect method

Add a percentage-change cell to the formula, such as growth equals 10% plus X

Flexing many assumptions from one control area

Data Table

Shows the effect of changing up to two inputs at once

Two-way results, such as price against volume

Goal Seek

Works backwards to find the input value that hits a target

Break-even and "what would it take?" questions

Keep every assumption in one place, colour-code the inputs, and test only the ones that matter.

What Are the Types of Sensitivity Analysis?

The main types of sensitivity analysis are one-at-a-time, local, global, and Monte Carlo approaches, and they differ in how many inputs change together and how widely each one moves.

How Do One-at-a-Time, Local, Global, and Monte Carlo Approaches Compare?

Type

How It Works

Limitation

One-at-a-time

Changes one input while holding the others constant

Can miss interactions between inputs

Local

Tests small changes around the base case

Says little about extreme values

Global

Varies all inputs together across their full ranges

Needs more time and computing effort

Monte Carlo

Samples inputs from probability distributions over many runs

Needs credible distributions to start with

Which Type Should You Choose?

  • A quick check on a few inputs: one-at-a-time is usually enough

  • Small tweaks around a trusted base case: local works well

  • Inputs that interact, or results that behave unevenly: go global

  • Real probabilities available: use Monte Carlo and read the spread of outcomes

If you're unsure, start simple. A one-at-a-time check takes minutes and often shows you whether a heavier approach is worth the effort.

What Does a Sensitivity Analysis Example Look Like?

A simple sensitivity analysis example is a project budget where you flex each cost driver across a realistic range and see which one moves the total most.

What Does a Simple Revenue Example Look Like?

Adapted from CFI's example, imagine a seasonal retailer that sold 500 packs at £20 each last year, giving £10,000. Testing shows that every 10% rise in shopper traffic lifts sales volume by 7%.

Traffic Change

Sales Change

Revenue

+10%

+7%

£10,700

+20%

+14%

£11,400

+40%

+28%

£12,800

+100%

+70%

£17,000

Once you know the relationship, you can read off any scenario without rebuilding the model.

How Do You Read a Tornado Diagram on a Project Budget?

A tornado diagram lists inputs from most to least influential, so the chart narrows like a funnel cloud. Here are illustrative figures for a £500,000 project:

Cost Driver

Baseline

Assumed Range

Swing on Total

Labour

£300,000

±10%

±£30,000

Materials

£125,000

±15%

±£18,750

Equipment

£50,000

±10%

±£5,000

Subcontractors

£25,000

±10%

±£2,500

Labour dominates, so that is where better estimates pay off first. Subcontractor costs barely matter, so they don't need the same scrutiny. When you present a chart like this, show the ranked drivers, name the top two, and attach one clear action to each, such as fixing a labour rate in the contract.

How Is Sensitivity Analysis in Scaled Agile Framework (SAFe) Used?

Sensitivity analysis in Scaled Agile Framework work means testing how far your estimates can be wrong before a decision changes, most often when ranking epics and features with WSJF. SAFe doesn't prescribe the technique by name, but it gives you natural places to use it.

How Does It Help With WSJF Prioritisation?

WSJF divides the Cost of Delay by job size, using relative estimates. Cost of Delay combines user-business value, time criticality, and risk reduction or opportunity enablement. Because the inputs are rough, a small change can reorder the backlog. Here is an illustrative check:

Feature

Cost

of Delay

Job Size

WSJF

A

20

5

4.0

B

16

3

5.3

C

22

8

2.8

Feature B ranks first. But if its job size is really 5 rather than 3, its score drops to 3.2 and Feature A overtakes it. That ten-minute check is a sensitivity analysis example any Lean Portfolio Management team can run, and it tells you which estimate to refine before committing. A good routine is to run the check on your top five items and look for pairs with close scores, because those are the decisions most likely to flip. SAFe itself notes that portfolio epics need better-informed estimates in the Lean business case than WSJF alone provides.

Where Else Does It Fit in SAFe?

  • Lean business case: flex the revenue, cost, and duration assumptions behind an epic to see which one the case depends on
  • PI Planning: ask what happens if team capacity is lower or a dependency slips, then link the answers to your ROAM risks
  • Portfolio funding: check how an epic's ranking changes under different budget guardrails
  • Forecasting: use Monte Carlo simulation for delivery dates, a topic Scaled Agile's own podcast has discussed

Sensitivity vs Scenario Analysis: What's the Difference?

Sensitivity testing isolates the effect of individual inputs, while scenario analysis changes several inputs together to describe one coherent future.

How Do They Compare?

Aspect

Sensitivity

Scenario

Focus

The effect of individual inputs

The combined effect of a defined set of changes

Key question

Which input matters most?

What happens if this future arrives?

Output

A ranked list of drivers

A handful of described outcomes

Typical use

Prioritising data collection and risk drivers

Strategic planning and stress testing

Neither is better. They answer different questions, and the strongest work often runs one after the other.

When Should You Use Each?

  • Use sensitivity testing to find the few inputs that deserve attention
  • Use scenario analysis to prepare for major shocks, such as a market shift
  • Use both, since drivers found first can feed the scenarios you build next
 
 
 
 
Sample Banner For Blog

What Are the Benefits and Limitations?

Its main benefit is showing where to focus, and its main limit is that it is only as good as the model behind it.

What Are the Benefits?

  • Identifies the key drivers of an outcome
  • Adds credibility by testing a model across many possibilities
  • Informs better decisions and supports risk management
  • Focuses data collection where it pays off most
  • Makes the robustness of a plan easy to explain to stakeholders

In short, it turns "we're not sure" into "here is exactly what we're unsure about, and how much it matters."

What Are the Limitations?

  • Depends on an accurate, complete model
  • One-at-a-time tests can miss interactions between inputs
  • Choosing sensible ranges and distributions is hard
  • Global and Monte Carlo runs can be slow
  • Shows how outputs respond, not what will happen
  • Can lead to paralysis by analysis if you test everything

Conclusion

Testing your numbers before you trust them is one of the cheapest risk controls you'll ever use. Start with one outcome and the handful of inputs most likely to move it. Rank them, then spend your effort firming up the top few. Run the same check on your WSJF scores before the next planning session. If you want to practise this on real portfolio decisions, a SAFe Course is a sensible next step. Share what you find early, so others can challenge the assumptions. The aim isn't a perfect model; it's a decision you can defend.

Frequently Asked Questions

1. What is the main purpose of what-if analysis? 

To show which inputs most affect a result, so you know where an estimate must be right and where it can be rough. It also shows where better data would be worth collecting, which saves time and money on estimates that were never going to change the answer.

2. How many inputs should you test at once? 

Start with the three to six you think matter most, one at a time. Move to a global or Monte Carlo approach when inputs interact, or many are uncertain. Testing everything rarely helps, because it hides the few inputs that truly drive the result.

3. Is it the same as scenario analysis? 

No. It isolates the effect of individual inputs, while scenario analysis changes several together to describe a possible future, often building on the drivers it identifies.

4. What tools can you use? 

Excel's Data Table and Goal Seek cover most business needs. For larger models, Monte Carlo add-ins such as @RISK or Python libraries such as SALib are common choices.

5. Can it predict the future? 

No. It shows how outputs respond to changes in inputs, not what will happen. Pair it with forecasting and scenario work for a fuller picture, and revisit it whenever your assumptions change.

About Author
Ishwin Khokhar

Corporate Trainer

Experienced Agile Coach with more than a decade of experience in transforming organizations through Agile methodologies. Specializing in SAFe (Scaled Agile Framework), I guide teams to drive continuous improvement, enhance collaboration, and achieve business agility at scale. Passionate about fostering a culture of innovation.    

Are you Confused? Let us assist you.
+1
Explore Leading SAFe® Agilist!
Upon course completion, you'll earn a certification and expertise.
Image Image Image Image

Trending Articles

The most effective project-based immersive learning experience to educate that combines hands-on projects with deep, engaging learning.
WhatsApp