Sensitivity Analysis on Unit Economics: Which Levers Matter Most

Sensitivity analysis on unit economics shows how much profit per customer changes when one assumption moves, so that you know which numbers deserve the most attention and which deserve very little. A model with ten inputs seems to demand equal care on all ten. In practice, two or three of them drive most of the result, and the team often spends its time on the one that matters least. This guide shows how to run one-way and two-way sensitivities on a per-customer model, how to rank drivers with a tornado table, how to build scenarios, how to find break-even points and how to set up the calculations in Excel. All arithmetic is checked. It belongs to unit economics and budgeting and forecasting, builds on the unit economics template and unit economics dashboard, and leads to profitability reporting and board-level profitability metrics.

The base case

Use a simple per-customer model for the first twelve months. Assumptions (hypothetical):

• Contribution per order before marketing: $26.00
• Orders per customer in twelve months: 3.0
• Customer acquisition cost: $20.00

Profit per customer = 3.0 × 26.00 - 20.00 = 78.00 - 20.00 = $58.00.

The model is small on purpose. A realistic one has more lines, but the method is the same.

One-way sensitivity

Change one input at a time by the same percentage, hold the others at base and record the result. Use plus and minus 10 percent.

Check: at 2.7 orders, 2.7 × 26 = 70.20, minus 20 = 50.20; at 3.3 orders, 85.80 - 20 = 65.80; at $23.40, 3 × 23.40 = 70.20, minus 20 = 50.20; at $28.60, 85.80 - 20 = 65.80; at CAC of $18, 78 - 18 = 60; at $22, 78 - 22 = 56. The swing is the difference between the high and low results. Orders per customer and contribution per order each move profit by $15.60, nearly four times the swing from CAC ($4.00). At this base case, a team that obsesses over shaving 10 percent off CAC is working on the least powerful lever. Raising repeat purchase or unit contribution, through price, shipping cost, returns or product mix, offers a bigger return. The ranking depends on the base: in a business where CAC is $60 and contribution per customer is $80, CAC would dominate.

Ranking drivers by swing gives a tornado table: list the inputs from largest to smallest swing. In a chart, the bars form a tornado shape. Use equal percentage changes for comparability, and also consider realistic ranges. If CAC has historically moved by 30 percent and repeat rate by 5 percent, test those ranges instead, because what matters is the likely movement.

Two-way sensitivity

Two-way tables show how two inputs interact. A common one for a first-order view is price against CAC. Suppose variable cost is $34 per order, regardless of price, and each customer places one order. Profit per customer is price - 34 - CAC.

Check: at a price of $56, 56 - 34 = 22, so profit is 22 - 16 = 6, 22 - 20 = 2, 22 - 24 = -2 and 22 - 28 = -6. At $60: 26 - 16 = 10, 26 - 20 = 6, 26 - 24 = 2 and 26 - 28 = -2. At $64: 30 gives 14, 10, 6, 2. At $68: 34 gives 18, 14, 10, 6. A $4 increase in price offsets a $4 increase in CAC exactly, so the diagonal pattern is clear. The table shows the combination of price and CAC at which first-order profit turns negative, and it helps answer a question such as "how much more can we afford to spend on acquisition if we raise the price by $4?" (up to $4 for a customer who orders once). With repeat orders the effect of price is multiplied, which is why the earlier table showed orders as a major driver.

Break-even tests

A break-even test asks how far an input can move before profit reaches zero.

• CAC at break-even: profit is zero when CAC equals contribution per customer, 3.0 × 26.00 = $78.00. At the base CAC of $20, there is a wide margin of safety (a CAC almost four times higher would be needed).
• Orders needed to repay CAC: 20.00 / 26.00 = 0.77 orders. A customer who orders once already repays CAC.
• Contribution per order at break-even: 20.00 / 3.0 = $6.67. Contribution per order can fall by about 74 percent before the customer loses money.

These tests give a sense of resilience. In this simple case the business is robust to any single input. Real models add more cost lines and more dependence between inputs, and the margins of safety are smaller. The first-order version is the harshest test, since it ignores repeat purchases. Run both.

Scenarios

Sensitivities change one or two things at a time. Scenarios change several together to describe a coherent story.

Check: downside 2.2 × 22 = 48.40, minus 32 = 16.40; upside 3.5 × 27 = 94.50, minus 25 = 69.50. Notice that in the upside case CAC is higher than base ($25 against $20) because a scenario can include a plausible trade-off, such as more spending on acquisition to reach stronger customers. Write a one-line story for each scenario so that the numbers have a meaning. The downside has profit per customer 72 percent lower than base (16.40 / 58.00 = 0.283), a result that would show how little room is left if several things go wrong at once.

Choosing ranges

Ranges should come from evidence, not convenience. For each input, look at the history: its minimum, maximum and typical movement over the last eight quarters. Use judgment for inputs without history, such as a new channel. Document the source of each range. If the model uses a plus and minus 10 percent change, state that it is for comparison only, and add the realistic range next to it.

Building it in Excel

1. Put all inputs on one sheet in named cells, and build the base-case model with formulas that refer to them.
2. For a one-way table, copy the base model result into a row, and use a data table (What-If Analysis, Data Table) with the input as the column input cell. Alternatively write a small block of formulas that recalculates the result with the changed input.
3. For a two-way table, use a two-variable data table with the two inputs as row and column input cells, and the profit formula in the top-left cell.
4. For scenarios, create a table of inputs by scenario, and a selector cell that uses INDEX and MATCH to pull the chosen scenario into the live inputs. Add a summary that shows the three results side by side.
5. Add a check cell that confirms the base case equals the hand calculation.
6. Lock formulas and leave only input cells open for editing.

Data tables can slow down large workbooks, so set calculation to manual if needed, and recalculate before sharing.

Using the results

Translate the findings into actions. If orders per customer is the dominant driver, invest in retention and replenishment reminders. If CAC is the dominant driver, work on channel mix and conversion. Set targets for the top two or three drivers, assign owners and track them on the dashboard. See unit economics dashboard. Report the sensitivity table to the board along with the plan, so that decisions on hiring and spending are made with an understanding of the risks.

Common mistakes

• Moving all inputs by the same percentage and assuming it reflects reality.
• Ignoring dependence between inputs. Higher prices often lower repeat rates.
• Testing only the upside.
• Using a model that is too detailed to understand.
• Not documenting the source of ranges.
• Leaving the analysis in a file and not using it to set priorities.

FAQs

What is sensitivity analysis in unit economics?

A way of testing how much profit per customer changes when one or more assumptions change, so you can see which drivers matter most.

What is a tornado chart?

A chart that ranks inputs by how much they move the result, with the largest swing at the top. The same ranking can be shown in a table.

What is the difference between sensitivity and scenario analysis?

Sensitivity changes one or two inputs at a time. A scenario changes several together to describe a coherent set of circumstances.

How do I set the ranges for each input?

Use historical movements and evidence, and document the source. Compare drivers on equal percentage changes for ranking, and also test realistic ranges.

Can I do this in Excel?

Yes. Use named inputs, data tables for one-way and two-way analysis and a scenario selector built with INDEX and MATCH.