What-if analysis is one of those practices that finance teams believe they're doing well right up until someone presses them on it in a board meeting and they realize the scenario they prepared doesn't actually answer the question being asked.
The mechanics of sensitivity analysis are straightforward — change one input, observe the output change. The discipline required to make it genuinely useful for decision-making is much harder to maintain. Three failure patterns show up consistently in mid-market FP&A work, and they're worth naming explicitly because they're each correctable with a specific practice change.
Mistake 1: Too Many Variables, No Variable Ranking
The most common what-if analysis failure is building a sensitivity table that varies too many inputs simultaneously without establishing which ones actually drive most of the output variance. A model with 15 inputs in the sensitivity table isn't more rigorous than one with 5 — it's usually less useful, because the signal-to-noise ratio drops and decision-makers lose the ability to identify which lever to pull.
The correct starting point for sensitivity analysis is a driver ranking exercise. Before building scenarios, run a one-at-a-time sensitivity: hold every input at its base case value, change one input by a defined amount (say, ±10%), and record the resulting change in your key output metric (ending ARR, EBITDA, or free cash flow, depending on what the decision is about). Rank the inputs by absolute output sensitivity.
In most mid-market SaaS models, 3–4 variables will account for 70–80% of total output variance. For a subscription business, those are typically: new ARR (or top-of-funnel close rate), gross churn rate, average contract value, and headcount growth rate. Everything else is secondary. Your what-if analysis should concentrate on those high-impact variables and treat the rest as secondary scenarios or footnote assumptions.
We're not saying the other variables don't matter. Infrastructure unit cost shifts can have real P&L impact in at-scale businesses. Payment terms affect working capital meaningfully. But in an FP&A context where the goal is decision-useful analysis rather than exhaustive modeling, identifying and focusing on the high-impact variables first is the discipline that separates useful sensitivity analysis from performative complexity.
Mistake 2: No Base Case Discipline
What-if analysis requires a credible, locked base case. This is less obvious than it sounds. The base case isn't "the plan" — it's a specific set of assumptions that everyone has agreed represents the most likely outcome given current information. Without that shared anchor, what-if analysis becomes a debate about whether the base case is right rather than a structured exploration of how outcomes change as conditions vary.
The practical failure mode looks like this: the CFO asks to see what happens if the sales team misses Q2 by 15%. Someone on the FP&A team builds the scenario, but the base case they're starting from still includes optimistic ramp assumptions for two new AEs who haven't actually started yet. The downside scenario looks less severe than it should because the base case was already understating likely near-term performance.
Base case discipline means: the base case reflects current actuals through the most recent close period, uses confirmed headcount (not approved headcount), uses updated win rate data from the current quarter rather than annual averages, and has been reviewed and signed off by the CFO before any scenario analysis begins. Any scenario — upside, downside, or stress — is built as a delta from that agreed starting point.
The mechanism that enforces this in practice is a base case freeze: once the base case is signed off, you lock it and track any subsequent changes to assumptions as explicit scenario deltas. This prevents the slow drift where assumptions get updated mid-analysis without anyone noticing that the comparison baseline has shifted.
Mistake 3: Scenarios That Don't Survive Contact with Business Reality
The third failure pattern is building scenarios that vary one driver in isolation when that driver doesn't, in reality, move in isolation. A churn scenario that applies a higher gross churn rate to the revenue model without adjusting the customer success headcount requirement, the CS-sourced expansion pipeline, and the support cost assumptions is mathematically coherent but operationally nonsensical.
Consider an FP&A team at a growing B2B software company modeling a "churn acceleration" scenario for their board deck. They set gross churn from 9% to 16% annualized and show the resulting ARR impact. What they didn't model: a 16% churn rate implies the CS team is failing to retain accounts at roughly 2x the planned rate. That implies either the CS team is understaffed (and needs additional headcount), or the product has a fundamental issue that will also suppress expansion ARR. Neither of those consequences appears in the model — so the scenario shows a revenue decline but doesn't show the cost response (more CS hiring) or the compounding effect (suppressed expansion). The board looks at the scenario and underestimates how bad the actual outcome would be.
The fix is to build what practitioners call "narrative-consistent scenarios" — scenarios where all the first-order consequences of a given driver change are modeled together. If churn goes up, CS capacity gets stressed, expansion slows, and potentially marketing spend needs to shift toward new logo acquisition to offset the lost NRR. Those secondary effects should be wired into the scenario branches, not left as assumptions for the audience to infer.
How to Present Sensitivity Output to a Board
The output format matters as much as the analysis quality. The most decision-useful format for sensitivity analysis at the board level is a two-variable sensitivity table (sometimes called a data table in Excel) that shows the output metric across a grid of two key driver values.
For example: a table with gross churn rate on the horizontal axis (7%, 9%, 11%, 13%) and new ARR closed in the period on the vertical axis (80% of plan, 100%, 120%), showing ending ARR in each of the 16 cells. The board can immediately read which quadrant they're in based on current actuals and see how bad the bottom-left corner looks vs. how good the top-right looks.
Complement this with a tornado chart — a horizontal bar chart ranking each driver by its impact on the output metric, showing the upside and downside range from a standard driver shift (typically ±10%). The tornado immediately communicates the relative importance of different risks and upside levers, which helps the board focus discussion on the variables that actually matter.
The one format to avoid: a table of 8–10 named scenarios, each with a different combination of driver assumptions, presenting full P&L output for each. By the time you've presented 8 scenarios, the board can't hold the differences in their heads simultaneously. Three to four scenarios at most, with the two-variable sensitivity table as a supplement for drill-down.
The Operational Rhythm: When to Run What-If Analysis
Sensitivity analysis isn't just a board presentation artifact. It should be part of the ongoing forecasting rhythm at three points in the quarter:
At the start of each quarter, when the operating plan is freshly set, run a driver ranking exercise to identify the 3–4 inputs with the most output sensitivity. Document those as the "key risk and upside levers" for the quarter — these are what you'll monitor and update in monthly forecasts.
At the mid-quarter forecast update (typically week 6–7 of the quarter), re-run the sensitivity on those key drivers using actual-to-date performance as the updated base. If actuals are tracking against plan assumptions, the sensitivity distribution stays similar. If actuals are diverging, the sensitivity output will show a more concentrated risk in the direction of the divergence — an early warning that the board will want to see.
Before any major resource allocation decision — a significant hiring batch, a budget reallocation, an acquisition assessment — run a decision-specific sensitivity analysis focused on the variables most relevant to that decision. This is where the discipline of parameterized driver-based models pays off: the same model infrastructure that supports the quarterly planning cycle can answer ad-hoc strategic questions without a full model rebuild.
The goal isn't more analysis — it's faster, more targeted analysis that arrives in time to change decisions rather than confirm them after the fact.