Why CFOs Discount CRO Forecasts (and How to Present a Number They'll Accept)
CFOs discount CRO forecasts because they're in gross revenue, not contribution margin, and never net the cost of losing tests. Here's the framing — and the spreadsheet — that survives a finance challenge.
Quick answer
CFOs discount CRO forecasts because the numbers arrive in gross revenue, assume every test wins, and skip the cost of running the program. Re-present the same forecast in contribution margin, multiply by your historical win rate, subtract the fully-loaded program cost, and include a payback month. That's a finance-grade number — and it's the one that gets signed off.
Why CFOs Discount CRO Forecasts
The systematic haircut finance applies to CRO ROI projections that use gross revenue, assume 100% win rates, and ignore program cost.
A CRO forecast that says "a 0.4pp lift on €8M of traffic is worth €320k" almost always lands on the CFO's desk and gets cut in half — or rejected outright. The reason isn't skepticism about experimentation. It's that the number is built the wrong way for a P&L owner. Finance plans in contribution margin, probability-weights every initiative against base rates, and expects a payback month. Most CRO business cases skip all three. This page shows the mechanism behind the haircut, how to detect which of your forecasts are at risk, and the exact restructure that turns a gross-revenue lift estimate into a number a CFO will defend in front of the board.
The gap is rarely about whether CRO works. It's about translation. A growth team's natural unit is incremental revenue; a CFO's natural unit is incremental contribution dollars, net of risk, with a date attached.
Why the haircut happens
Three structural errors show up in almost every CRO forecast that gets discounted. They're not analytical mistakes — they're framing mistakes that make the number incomparable to every other initiative the CFO is ranking.
First, the forecast is in gross revenue. On a Shopify apparel brand with a 38% contribution margin, a €320k revenue lift is €122k of contribution. The CFO is comparing it to a paid-media spend that's already netted of COGS, returns, payment fees, and fulfilment — so the CRO number looks 2.6× bigger than it actually is in their unit.
Second, the forecast assumes the test wins. Industry win rates sit at 15-25% for a mature program. If you forecast €320k of lift per test and you only ship one in five, the expected value per test is closer to €60-80k — and that's before subtracting the revenue lost on losing variants that ran at 50% traffic for two weeks.
Third, the forecast omits cost. A two-person CRO function plus tooling plus design time runs €180-280k a year on a brand in your revenue band. If that line doesn't appear on the slide, the CFO mentally adds it — and usually overestimates it. Better to show it.
The three errors, compounded
Gross revenue × assumed 100% win rate × no program cost = the standard CRO forecast. A CFO who multiplies through by 0.38 (margin) × 0.20 (win rate) and subtracts €220k of run-cost lands at a very different number than the deck claims. They're not being conservative — they're being consistent with how every other line on the P&L is forecast.
How to detect which forecasts are at risk
Read your last CRO business case and check four things. Does the headline number use gross revenue or contribution margin? Is there an explicit win-rate assumption with a citation, or does it implicitly assume 100%? Does the cost side include people, tools, and design hours fully loaded? Is there a payback month, or only an annualised figure?
If three or four of those answers are missing, your forecast will get haircut. The pattern is consistent across e-commerce finance teams: anything that doesn't show up in their format gets converted into their format on the fly, and the conversion is rarely generous. The companion page on CRO ROI walks the underlying math; the payback-period guide for DTC brands covers the date side.
How to present a number they'll accept
Restructure the forecast in four lines. Start with the gross revenue lift from your test design — say €320k annualised on a beauty SKU expansion test. Multiply by contribution margin (38% → €122k). Multiply by historical win rate (20% → €24k expected value per test). Multiply by planned test volume per year (24 tests → €585k loss-adjusted contribution).
Then subtract fully-loaded program cost (€240k → €345k net contribution) and divide by program cost to get a payback multiple (2.4×). Add a payback month from the first compounding test. That's six numbers, and every one of them maps to a line the CFO already uses. The Marketing ROI Calculator can sanity-check the same restructure against your paid-channel forecasts so the numbers sit in one comparable frame.
The line that closes the room
"At our historical win rate, this program returns 2.4× its fully-loaded cost in contribution margin, with payback in month 7. If our win rate is half what we've shown, it still breaks even." That sentence — with the downside case built in — is what gets the headcount approved. CFOs sign off on initiatives where the worst case is articulated, not hidden.
Experiment ideas to harden the forecast
Before the next board cycle, run three internal exercises. Pull your last 12 months of tests and compute your actual win rate, average winning lift, and average losing lift. Most teams discover their win rate is lower than they cite and their average winning lift is higher — both numbers improve forecast credibility. Then back-test last year's forecast against actual realised contribution and present the variance; CFOs trust teams that show their own track record.
Finally, agree the contribution-margin number with finance before you forecast against it. The single fastest way to lose credibility is to use a 45% margin in your deck when finance uses 38% in theirs. One pre-meeting eliminates the entire category of "that's not how we count it" objections.
Frequently asked questions
Start with gross margin minus payment fees, fulfilment, and a returns reserve — typically 30-42% for apparel and beauty, 18-28% for electronics. Footnote your assumption and ask finance to correct it. The act of asking usually surfaces the real number within a week.
No — it's matching the rest of the P&L. Paid acquisition forecasts assume some campaigns will underperform; product-launch forecasts assume some SKUs will miss. A CRO forecast that assumes every test wins is the outlier, and that's why it gets haircut.
Use the industry base rate of 18-22% for a mature program, then haircut it to 12-15% for year one to reflect learning curve. Cite the source. Commit to replacing it with your own data by month nine — and actually do it.
Yes, but it's usually small. A test running at 50/50 traffic for two weeks with a losing variant -2% vs control costs roughly 1% of two weeks of revenue. Across 24 tests, that's a real number — €40-80k on an €8M brand — and naming it before the CFO does is a credibility win.
Six to nine months is defensible for a brand in the €1-15M band; under six months is exceptional and probably worth challenging. The CRO program payback period guide for DTC brands walks through how each input moves the date.
Conservatively. Assume each winning test holds for 12 months and ignore stacking gains across tests in the first forecast. CFOs heavily discount compounding assumptions because they rarely survive contact with seasonality and traffic-mix shifts.
Forecast in the metric the test targets, then convert to revenue at the end. A checkout test forecasts conversion rate; a bundle test forecasts AOV. Mixing them in the same line is what creates the unit confusion CFOs distrust.
Annualising a single winning test. Saying "this one test is worth €320k a year" implies the lift is permanent, universal, and uncontested by future tests on the same page. CFOs have seen that movie. Forecast the program, not the test.
One slide, six numbers: gross lift, × margin, × win rate, × test volume, − program cost, payback month. If it takes more than that, the math isn't tight enough yet. Appendix the assumptions.
A shared spreadsheet with finance is enough for year one. By year two, use whatever surfaces your actual win rate and average lift automatically from your test history — that data, not the model, is what makes the forecast defensible.
Test ideas before you ship them
Run unlimited A/B tests, attach hypotheses to outcomes, and build a searchable archive of what works — and what doesn't.