Step 1 — identify the two or three load-bearing inputs. Build a quick tornado chart by flexing each input one at a time by its plausible economic range (not by an arbitrary fixed percentage) and measuring the impact on the output. Most models have two or three inputs that account for 80% of the variance, and a long tail of inputs that move the answer by less than 5%. Sensitivity work belongs on the first group; the rest is noise.
| Sensitivity exercise | What it shows | Verdict |
|---|---|---|
| Flex every input by plus-or-minus 10% | That the model has eight inputs | Cargo cult — uninformative |
| Tornado chart: each input flexed across its plausible range, one at a time | Which two or three inputs the answer is actually sensitive to | Useful — drives further work |
| Two-variable table on the load-bearing inputs | Where in input-space the answer flips conclusion | Useful — the indifference frontier is the decision boundary |
| Monte Carlo on every input with assumed distributions | A confidence interval, often falsely precise | Useful only if the input distributions come from actual evidence |
Worked example — Crestline Tools (fictional industrial supplier). Tornado chart shows the DCF output is dominated by (a) 2026E EBITDA margin and (b) terminal growth rate; revenue growth, capex, working capital and discount rate each move the answer by less than $1.50 per share within their plausible ranges. The two-variable sensitivity table flexes EBITDA margin from 14% to 19% (the band of last-five-year actuals plus management guide) against terminal growth from 1.5% to 3.5%. The intrinsic value crosses current price of $42 along a diagonal line — the indifference frontier. Above-and-right of the line, the stock is a long; below-and-left, it is a short or pass. The analyst reports: "I am long because I think EBITDA margin recovers to 17% (above the 16% indifference line at my terminal-growth assumption); my kill criterion is Q3 margin below 15%."
The indifference frontier — the combination of load-bearing inputs where the intrinsic value equals the current price — is the single most decision-useful object a sensitivity exercise produces. It converts "is the model right?" into "which side of this line do I believe?". The latter is a question the analyst can actually answer; the former is a question that produces endless tinkering.
Sit with the ideas.
An analyst's sensitivity table on Crestline Tools shows the DCF output flexed by plus-or-minus 10% on every single input — revenue growth, gross margin, SG&A, capex, working capital, terminal growth, discount rate, share count. The table fits on one page. What is the right critique?