CRO ROI is the extra gross profit your winning tests produce, minus what the program costs, divided by that cost.
A program with enough traffic usually pays back because a small share of winners carry the rest.
Industry data puts that share at about 12% of tests in Optimizely's analysis of 127,000+ experiments, and testing also stops you shipping changes that would have cut revenue.
Below is the formula, a worked calculator with three scenarios, and real results from our database of A/B tests showing where the return comes from and where it gets overstated.
The ROI of A/B testing: formula and worked calculator
Use gross profit, not revenue. A test that adds $10,000 in sales at a 50% margin adds $5,000 you can set against the cost of finding it.
| Input | What it means |
|---|---|
| Monthly revenue (R) | Revenue from the traffic you can test on. Use revenue per visitor × visitors. |
| Winners per year (W) | Tests run per year × win rate. |
| True lift per winner (L) | The lift you expect to keep after launch, not the lift the test reported. |
| Months live (M) | How long each winner runs inside the period you measure. |
| Retention (D) | Share of the lift that survives over time. |
| Gross margin (G) | Turns revenue into profit. |
| Program cost (C) | Agency or staff, tools, design and development time. |
The formulas in plain text:
- Incremental profit = W × R × L × M × D × G
- ROI = (Incremental profit − C) / C
- Break-even lift = C / (W × R × M × D × G)
Worked example: a store with $1M a month in revenue
Take a store with 500,000 visitors a month at $2.00 revenue per visitor, so R = $1,000,000. It runs 24 tests a year.
At the 12% win rate in Optimizely's benchmark, that gives about 3 winners (W = 3).
Winners ship through the year, so on average each is live for 6 months of year one (M = 6).
Optimizely found revenue uplifts kept 91% of their first-month value after a year, so D = 0.91. Gross margin is 50%.
For cost, assume $8,500 a month: an $8,000 retainer plus a $500 testing tool, or $102,000 a year.
That sits inside the $800 to $10,000 a month WebFX quotes for agency CRO (as of November 2026).
| Scenario | True lift per winner | Year-one extra revenue | Extra gross profit | ROI |
|---|---|---|---|---|
| Cautious | 2% | $327,600 | $163,800 | 61% |
| Middle | 4% | $655,200 | $327,600 | 221% |
| Optimistic | 8% | $1,310,400 | $655,200 | 542% |
The middle row works out as 3 × $1,000,000 × 0.04 × 6 × 0.91 = $655,200 in revenue, then × 0.5 = $327,600 in profit.
($327,600 − $102,000) / $102,000 = 221%. Break-even here is an average true lift of about 1.25% per winner.
Year two is usually better. Winners from year one run all 12 months, and you only pay for the program, not for the lifts it already found.
The same program on a small store
Now take a store with 50,000 visitors a month and $100,000 in monthly revenue.
It can run far fewer conclusive tests, so assume about one winner a year at the same 4% lift and the same $102,000 cost.
One winner live for 6 months adds $21,840 in revenue and $10,920 in profit.
ROI is about −89%. The math only flips if the cost drops to a few thousand dollars, which points this store toward a one-off CRO audit and shipping obvious fixes rather than a full testing retainer.
Why traffic sets the ceiling
Revenue is a noisy metric, so revenue tests need a lot of visitors.
In Kohavi and colleagues' KDD 2007 guide, detecting a 5% change in revenue in their example needed over 1.6 million users.
A 5% change in conversion rate needed under 500,000, and a 20% change needed 30,400.
That is why smaller sites run fewer tests and need bigger lifts to see anything.
Check your own numbers with the sample size calculator before you budget for a test count you can't reach.
Where the ROI estimate goes wrong
- Using the reported lift. Winning tests overstate their effect, especially at low power. This is the winner's curse, and it is why the calculator asks for true lift, not test lift.
- Annualizing month one. Lifts fade. Optimizely's data shows most uplifts kept 80% of their first-month value after a year, with revenue holding at 91%.
- Adding up every winner. Lifts on the same page overlap, so five 3% wins rarely add up to 15%. A holdout group measures the combined effect directly.
- Using revenue instead of profit. A discount test can grow revenue while cutting margin.
- Leaving out internal costs. Developer and design hours count even when no invoice arrives.
Conversion optimization ROI in practice
Real programs earn their return in two ways: a few winners that pay for many flat tests, and losers caught before they ship.
Our database shows both, along with how practitioners report the money.
Most tests won't win, and that's built into the math
At Microsoft, Kohavi and colleagues reported that only about 1/3 of ideas improved the metrics they were meant to improve.
Optimizely's benchmark puts winners at around 12% on the primary metric.
Clean Commit publishes its full test library. Of its 173 tests in our database, 16 won, 35 lost and 122 were flat, a win rate of about 9%.
Our win-rate research covers why published case studies look far better than this.
If you budget on the assumption that every test wins, the program will look like a failure by month three.
Budget for roughly one winner in eight to ten tests.
Losses avoided are part of the return
Every losing test is a change you didn't ship to all your traffic.
That value is real, even though it never shows up as a revenue line.
- Clean Commit tested a single-row header with an image mega-menu. It lost: −22.5% conversion rate and −12.3% revenue per visitor, with a 99.5% probability that the control was better.
- Dave Diederen tested removing the email popup to "clean up" a store. Revenue per visitor fell 20.51%.
- Oliver Kenyon found a live chat feature that was costing revenue. Removing it recovered about $28,000 a month.
On the $1M-a-month store above, shipping a −12.3% revenue change untested would cost about $123,000 a month until someone noticed.
One caught loss like that covers more than a year of the program.
The same idea can win or lose
Copying tactics is the cheap alternative to testing, and our data shows why it's risky.
Dave Diederen added a countdown timer to checkout and saw +3.16% revenue per visitor.
On another store, the same idea returned −7.1% revenue per visitor.
Without a test, you can't tell which of those two stores you are.
How practitioners report the money
Many results in the database are reported in dollars, which makes ROI easy to discuss but easy to inflate. Read how each number was built.
- Dave Diederen's Best Sellers section at the bottom of the product page reports +$488,000 extra yearly revenue. That is an annualized revenue figure.
- Kasper Dolk's hero benefit bullets that answer survey objections report +$763,679.14 profit/year. Profit is the right base for ROI.
- Oliver Kenyon's brand-aligned scarcity text below Add to Cart reports $44,751/month, labeled as forecasted.
Forecasts like these assume the test lift holds.
Discount them for decay and the winner's curse before you put them in a business case.
@oliverkenyon
@DaveDiederen
@kasper_dolk@DaveDiederen@DaveDiederen
Clean CommitAgency, in-house or audit
The cost line in the formula depends on who does the work.
Our CRO pricing guide breaks down retainers, audits and performance deals.
| Option | Cost reference | Fits when |
|---|---|---|
| Audit, then ship fixes | One-off fee | Traffic is too low for steady testing |
| Agency retainer | $800 to $10,000 a month (WebFX) | You have traffic but no testing team |
| In-house specialist | Indeed average $111,259 a year for web optimization specialists, plus tools and dev time | You can keep a test pipeline full all year |
Prices as of November 2026. Whatever you choose, ask for results reported as profit, with lifts discounted for decay, and a way to prioritize tests by expected value.
How to raise the return
- Test where the money is: checkout, cart and product pages usually carry the most revenue per visitor.
- Base ideas on research, like Kasper Dolk's survey objections, rather than on tactic lists.
- Kill flat tests on schedule so you don't burn traffic.
- Track the whole program, not single wins, and learn the basics in our A/B testing guide.
If you'd like someone to model this for your store and run the tests, our CRO experts directory lists practitioners whose results are in the database.