A few years ago I watched a team celebrate a test that had just cost them money.
They’d added a “Save 30% today” banner to a sales page. Conversion rate jumped from 3% to 3.9%, and the tool showed a big green checkmark.
Nobody looked at average order value. It had dropped from $97 to $68, because buyers skipped the order bump and picked the cheaper option. Revenue per visitor went down. They shipped the banner anyway, because the only number on the screen said “winner”.
That’s the problem with conversion rate as a primary metric in a funnel. It measures whether people clicked, not whether you got paid. With a bump, an upsell, a downsell or a subscription, roughly half of a visitor’s value happens after conversion rate stops looking.
Key takeaways
Conversion rate tells you how many people bought. Revenue per visitor tells you how much you made per visitor. They can move in opposite directions.
Discounts, urgency and scarcity usually trade order value for conversion rate. A test that only measures conversion rate can’t see the trade.
Use revenue per visitor as the primary metric on any page that sells something. Keep conversion rate as a guardrail.
Revenue tests need more traffic. Cover the gap with bigger changes, Bayesian stats, bandit allocation and outlier caps.
What is revenue per visitor?
Revenue per visitor (RPV) is total revenue divided by unique visitors. It equals conversion rate multiplied by average order value, so it captures both how many people buy and how much they spend.

RPV is CVR multiplied by AOV, so a change that moves one up and the other down can’t fool it.
In a funnel, “total revenue” should include everything the visitor bought in the session:
The main product
The order bump, upsells and downsell
The first payment of any subscription
Minus refunds
That’s the number your bank account sees. Anything less is an approximation.
The test that won and lost money
Here’s that discount test with all the numbers filled in.
Visitors | Orders | Conversion rate | Average order value | Revenue | Revenue per visitor | |
|---|---|---|---|---|---|---|
Control | 10,000 | 300 | 3.0% | $97 | $29,100 | $2.91 |
Variant B | 10,000 | 390 | 3.9% | $68 | $26,520 | $2.65 |
Variant B lifted conversions by 30%. It also cut revenue per visitor by 9%.
Ship it and you lose about nine cents on every visitor.
At 50,000 visitors a month, that’s over $4,000 a month gone, hidden behind a green checkmark.
The discount did exactly what discounts do. The measurement was the failure.
Why conversion rate misleads in funnels
Conversion rate works when there’s one yes-or-no outcome and every outcome is worth the same, like a newsletter signup. Funnels break both assumptions.
Not all conversions are equal. A $47 buyer and a $47 buyer who also took a $197 upsell are both “one conversion”.
The conversion you see isn’t the last one. A pushier order form can win on checkout rate and lose on upsell take rate.
The easiest levers trade AOV for CVR. Discounts, countdown timers and “only 3 left” produce fake wins when you only look at one side.
Where conversion rate still earns its place
Don’t stop watching conversion rate. It just has a different job.
Conversion rate moves fast, which makes it great at catching breakage:
A payment form that fails on mobile
A variant that loads slowly
A headline that quietly confused people
So treat conversion rate as a guardrail that can veto a result, and let revenue per visitor be the primary metric that decides it.

Mark CVR or refunds as a guardrail and Splitlab flags any variant that makes it significantly worse.
A variant that wins on RPV while conversion rate craters still deserves a hard look. You may have caught a few lucky big orders, not a real effect.
The honest cost of measuring revenue
Most tools default to conversion rate for a reason: revenue is noisier.
A conversion is 0 or 1. Revenue per visitor runs from 0 to your biggest bundle, and a few large orders can swing the average. A revenue metric can need several times more traffic to detect the same lift.
Four things make it workable.
1. Test bigger changes
A new offer structure or price point produces an effect you can see. A button colour doesn’t, on any metric.
2. Use Bayesian statistics
You get a probability that each variant is best, plus how much you’d lose by picking wrong. You can read both at any time. More in Bayesian A/B testing for marketers.
3. Let a bandit allocate traffic
If you’d rather earn during the test, a multi-armed bandit shifts traffic toward the higher-RPV variant as evidence builds.

Switch a test to Multi Arm Bandit from the traffic split menu.
Multi-armed bandit vs A/B testing covers when to use it.
4. Cap outliers
One $5,000 order in one arm shouldn’t decide a test on a $97 product. Cap order values at a sensible ceiling before the metric is computed.
Which metric for which page
The closer a page sits to money, the more the primary metric should be money.
Page you’re testing | Primary metric | Guardrails |
|---|---|---|
Opt-in / lead magnet | Lead rate (CVR) | Downstream purchase rate, if you can attribute it |
Sales page / VSL | Revenue per visitor | Checkout rate, refund rate |
Order form / checkout | Revenue per visitor (whole funnel) | Checkout completion rate |
Order bump | Revenue per visitor | Main-offer conversion rate |
Upsell / OTO | Revenue per funnel visitor | Take rate, refund rate |
Downsell | Revenue per funnel visitor | Take rate |
SaaS pricing page | Revenue per visitor or trial-to-paid | Trial starts |
Booking page (services) | Booked calls, or won-deal revenue | Form submit rate |
Conversion rate is only the honest primary metric on a lead capture page, where nothing is sold yet. Even there, attribute downstream revenue back to the variant when you can.
How to set this up
You need three things:
Revenue events that actually reach your testing tool
A way to tie each event to the variant the visitor saw
A tool that makes RPV the primary metric, with conversion rate as a guardrail
Here’s how it looks in Splitlab.
1. Connect your funnel or payment platform
The ClickFunnels 2.0 integration sends every order, subscription and opt-in by webhook. Refunds and failed payments are deducted, so you optimize net revenue, not gross.

Connect ClickFunnels from the Integrations page. Revenue then arrives by webhook.
Running ClickFunnels? How to A/B test ClickFunnels 2.0 pages walks through the full setup.
2. Star revenue per visitor as the primary metric
RPV, AOV and CVR are already there as standard metrics. Star RPV and leave the other two as guardrails.

Every Splitlab test comes with RPV, AOV and CVR built in. Star one to make it the primary.
3. Decide how to count subscriptions
First payment is the conservative default. Predicted lifetime value is possible, but it makes the metric slower and noisier.
4. Set an outlier cap
Cap order value if you sell anything with a wide price range.
5. Read the result as a decision
Skip the p-value. Look for a sentence like this: “Variant B has a 96% probability of higher revenue per visitor, and if we’re wrong we expect to lose about $0.04 per visitor.” You can act on that.
The short version
Conversion rate answers “did they click?”. Revenue per visitor answers “did we make more money?”.
Use RPV as the primary metric on money pages. Keep CVR as a guardrail.
Revenue tests need more traffic. Cover the gap with bigger swings, Bayesian reads and bandit allocation, not by measuring the wrong thing very precisely.
Frequently asked questions
Is revenue per visitor the same as average order value?
No. Average order value only looks at people who bought. Revenue per visitor divides total revenue by everyone who arrived, so it accounts for conversion rate too. RPV equals CVR multiplied by AOV.
Should I use revenue per visitor on a lead generation page?
Usually not as the primary metric, because no money changes hands there. Use lead rate as the primary. If your CRM or funnel platform can attribute closed deals back to the visitor, track downstream revenue as a secondary metric.
How much more traffic does a revenue test need?
It depends on how spread out your order values are. One product at one price behaves almost like a conversion metric. A $27 front end with a $997 upsell may need several times the sample. Bigger effects, Bayesian analysis and outlier caps all bring it down.
What about refunds?
Count them. A variant that sells more and refunds more is not a win. Deduct refunds from the variant that earned the original charge, and add refund rate as a guardrail when the test changes the offer.
Can conversion rate and revenue per visitor point in opposite directions?
Yes, and it’s common. A lower price, a discount or a nudge toward a cheaper option tends to raise conversion rate and lower AOV. Whether RPV goes up or down depends on which effect is bigger, so measure it directly.
My final thoughts
I still think about that team celebrating a test that cost them money. Nobody did anything wrong. They just measured the number that was easiest to see.
Conversion rate is a useful signal, and on opt-in pages it’s often enough. But on any page that leads to a sale, revenue per visitor should decide the test, because it counts both how many people buy and how much they spend. Track both, and let revenue make the call.
Your test report should agree with your bank account.
Ready to test for revenue instead of clicks? Splitlab credits every order, bump, upsell and refund to the variant that produced it. Run your first test or book a demo.
Built for revenue
Test for revenue, not just clicks.
Run A/B, multivariate, and multi-armed bandit experiments — with real revenue tied to every variation.
We’ll increase your conversion rate by 30% or you don’t pay. Terms apply.
✓
Revenue per variation
✓
A/B & MVT
✓
Multi-armed bandit
Test for revenue, not just clicks.
A/B, multivariate and multi-armed bandit experiments, with real revenue tied to every variation.


