Choosing a Defensible CR-Lift Assumption for the CFO One-Pager
A practical method for choosing the CR-lift assumption in a CRO business case — the single line a CFO will attack first — using historical win rates, industry data, and a haircut for regression to the mean.
Quick answer
For a CFO-facing one-pager, anchor your blended CR-lift assumption at 5-8% over the sprint horizon — not per-test. Build it bottom-up: (historical win rate) × (median winning lift) × (0.7 regression haircut). Anything above 10% blended needs a documented reason, or the CFO will discount the whole case.
Defensible CR-lift assumption
The conversion-rate uplift number in a CRO business case that can be defended line-by-line against finance scrutiny.
A defensible CR-lift assumption is the single uplift percentage you put into a CRO business case — the input that drives projected CAC savings, payback period, and incremental revenue. It is not the lift of your best historical test, nor the headline number from a vendor case study. It is a blended, time-bounded estimate built from three anchors: your own historical win rate, a sober industry benchmark, and an explicit haircut for regression to the mean.
In a finance conversation, this line is attacked first because it has the largest leverage on every downstream output. If it survives, the rest of the one-pager usually does too.
Performance Managers lose CRO budget conversations on this one number more often than on any other. Pick it too high and finance discounts the whole model; too low and the payback period stretches past the budget cycle.
Why the number gets attacked
CFOs see three failure patterns repeatedly. First, the marketer cites a single winning test ("+22% on the PDP") and extrapolates it across the whole funnel. Second, the assumption is per-test instead of blended across wins, losses, and flats. Third, there is no haircut for the fact that lifts in a controlled test rarely hold at full traffic.
The fix is structural, not rhetorical. You build the number from components a finance partner can audit, and you show your work on the same page.
The 15% trap
If your one-pager assumes a 15%+ blended CR lift over a quarter, expect the CFO to ask why your historical results don't already show this happening. Either you have unexploited headroom you can prove from the pre-sprint audit, or the number is aspirational. There is no middle ground.
The three-anchor construction
Anchor one is your historical win rate. Pull the last 12-24 months of A/B tests on your Shopify or WooCommerce store. Calculate (winning tests / total tests shipped) and (median lift of winners). An online apparel store running 24 tests with 7 winners at a median +6% gives you 29% × 6% = 1.7% expected lift per test.
Anchor two is an industry benchmark for sanity-checking. Public data from Shopify Plus, Baymard, and CXL puts realistic per-sprint blended lifts in the 4-9% range for stores in the €1M-€15M band. If your bottom-up math lands far outside this range, the CFO will notice before you do.
Realistic blended CR-lift expectations by store maturity and sprint length
| Store maturity | 12-week sprint | 26-week sprint | Per-test median |
|---|---|---|---|
| First CRO programme (no test history) | 3-6% | 6-10% | 4-7% |
| Mature programme, untested funnel area | 5-9% | 9-14% | 5-8% |
| Mature programme, already-optimised area | 2-4% | 4-7% | 2-4% |
| Post-replatform / major UX overhaul | 8-15% | 12-22% | 6-12% |
The regression haircut
Anchor three is the haircut. Test winners regress when shipped to 100% of traffic — novelty effects fade, segment mix shifts, and the original measurement window was likely too short. The industry rule of thumb is a 30% haircut on measured lift; some teams use 40% for tests stopped early.
Applied to the apparel example: 1.7% expected lift per test × 0.7 haircut = 1.2% per test. Across 6 tests in a sprint, that compounds to roughly a 7.4% blended lift — a number you can defend, source line by line, and put on the one-pager.
What to show the CFO
Put the calculation on the page itself, not in an appendix. Three rows: (a) your historical win rate × median winning lift, (b) the regression haircut and why, (c) the resulting blended assumption. The transparency is what makes it survive — CFOs reject opaque numbers, not modest ones.
Experiment ideas to validate the assumption early
Run one or two high-confidence tests in the first three weeks of the sprint — ideally on the funnel step where the pre-sprint audit showed the largest drop-off. The goal is not to win big, it is to produce an early data point that either confirms or corrects your blended assumption while the budget is still flexible.
Once you have the validated lift, you can move directly into translating CR lift to monthly CAC savings at constant ad spend — the next line on the same one-pager. That conversion turns an abstract percentage into a euro figure the CFO can compare to the sprint cost.
Frequently asked questions
For a mature programme on an already-optimised funnel, no. For a first CRO programme or a post-replatform store, yes — but you need pre-sprint audit evidence of the headroom. Without that evidence, 10% reads as aspirational and the CFO will discount it.
Blended. Per-test numbers are misleading because they ignore flats and losers. A blended number across all tests in the sprint matches how finance models incremental revenue and is the only version that survives scrutiny.
Use the industry benchmark band (3-6% for a first 12-week sprint) and lean on the pre-sprint audit to justify where in the range you sit. Be explicit that you are using external data; CFOs prefer a labelled estimate over a confident-sounding guess.
30% is the common starting point. Use 40% if your tests typically run under two full business cycles or you stop them early. Use 20% only if you have a documented history of post-rollout lift matching test-measured lift.
Not as the primary anchor. Case studies are selection-biased toward big wins. Use them as a directional sanity check alongside your own historical data and a public benchmark — never as the sole source.
Stores in the €1M-€15M band typically see 4-8% blended lift in a 12-week first sprint, concentrated in checkout and PDP fixes. The variance is wide; the audit-driven number is more useful than the benchmark average.
Present a range, not a point estimate. Show base (your bottom-up number), downside (apply a 50% haircut instead of 30%), and upside (haircut removed). Then anchor the business case on the base — the range itself signals you've thought about the risk.
Partially. Two non-overlapping winners on different funnel steps compound roughly multiplicatively; two tests on the same step usually don't. Model it conservatively as additive minus 20% for interaction effects unless you can isolate the steps cleanly.
No — they must match line for line. If the full model uses 7.4% blended lift, the one-pager shows the same 7.4%. Mismatches are the fastest way to lose credibility in a finance review.
Quoting a single winning test as the assumption. The CFO immediately asks "and what about the losers?", and the rest of the conversation is spent defending arithmetic instead of strategy. Always lead with the blended number.
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.