Is CRO Worth It? ROI, Costs and Realistic Expectations

Conversion rate optimisation (CRO) sells itself on a tidy promise: keep the same traffic, squeeze more revenue out of it, repeat. The promise is real, but it is conditional, and most pages that rank for this question are written by people who want to sell you a tool or a retainer. So here is the honest version. CRO is worth it when you have enough traffic and enough testing capacity to find and ship wins faster than they decay. Below a certain volume, paying for a formal A/B testing programme is a way to set money on fire while feeling rigorous. This page explains the maths, the costs, the win rates nobody quotes, and a decision framework you can apply to your own site.

The maths that makes CRO attractive

Start with why anyone bothers. If your site converts at 3% and you move it to 4%, that is not a one-point change; it is a 33% increase in revenue from the same traffic, with zero extra acquisition spend. One percentage point of conversion rate can be worth a third more revenue. That is the headline that makes CRO look irresistible.

Improvements also stack multiplicatively rather than additively. A 20% lift at one funnel stage combined with a 10% lift at another gives you roughly 1.20 multiplied by 1.10, which is about a 32% gain overall, not 30%. Chain several genuine wins together and the compounding is what pays for the programme. Our CRO revenue uplift calculator lets you plug in your own numbers to see what a given uplift is worth before you commit to anything.

There is a catch the marketing pages skip: winners decay. A test that lifts conversions by 10% is a depreciating asset, not a permanent gain. Competitors copy the change, novelty fades, and the measured uplift shrinks. Treat every win as something you have to keep replacing.

What an honest win rate looks like

Here is the part that determines whether the maths above ever materialises. Most tests do not win.

• Optimizely analysed roughly 127,000 experiments across 1,100 companies and found only about 12% won on the primary metric. Win rates for revenue-tied experiments sat around 10%.
• Industry benchmarks from other platforms land higher, around 25 to 30%.
• At Microsoft’s Bing, the engineers who pioneered large-scale online experiments report that roughly a third of ideas come out positive and significant, a third flat, and a third negative and significant. In a heavily optimised product, improving a top-line metric can be closer to a one-in-thousands event. The detail is in Ronny Kohavi’s lessons from twelve years of running A/B tests.

Round it off: somewhere between 70% and 90% of your tests will not produce a winner. The returns come from volume of attempts. You need to run a lot of tests to bank a few wins, which means test velocity is the binding constraint, and test velocity depends on traffic.

The size of the wins is sobering too. Optimizely’s data puts the median winning test at about a 2.77% revenue-per-visitor uplift, with the top quartile of winners around 5.21%. The 20 to 50% uplifts you see in case studies are the exceptions that get written up, not the median you should budget for.

The traffic threshold that decides everything

A/B test results only mean something if the test reaches statistical significance, and significance needs sample size. The rough rule: a page needs at least a few thousand monthly visits and around 100 conversions per variant before a test is viable. Some practitioners push that to 350 to 400 conversions per variant for solid 95% confidence, the industry-standard threshold.

Smaller uplifts need far larger samples. If you are hunting for a 2% improvement, you need a lot more traffic than if a 15% improvement would still be worthwhile. Most A/B tests take two to four weeks to reach significance. If a test would take longer than six to eight weeks, you either widen the effect you are willing to detect, narrow the scope, or send more traffic to it.

Below the threshold, testing gives you the feeling of rigour without the conditions for rigour. Underpowered tests, especially when someone peeks early and stops the moment a result looks good, manufacture false confidence. You ship a “winner” that was noise, your developers build it, and you have paid to make your site worse. For more on baselines and what counts as healthy, see what is a good conversion rate.

Test velocity: the number programmes live or die by

A good CRO programme is defined by steady velocity, tests launched, completed and shipped month after month, not by one celebrated win. The reality is modest: most practitioners run only a couple of tests a month, and fewer than one in ten run more than twenty. If you can manage one test a quarter, the wins will never compound faster than they decay, and the programme cannot pay for itself. Building that cadence is the real work; our guide on how to build a CRO programme covers the operational side.

What CRO actually costs

There are three cost models, and they suit very different sizes of business.

Tools. A/B testing platforms are tiered. Entry tiers sit in the low-hundreds-per-month band; fuller suites run to the high hundreds. Enterprise platforms move into annual contracts in the tens of thousands. A market note worth knowing for 2026: VWO has been retiring its permanent free plan, with new sign-ups offered a trial rather than a free-forever tier, and the testing-tool market consolidated when VWO and AB Tasty merged under common ownership. Check current terms on vwo.com before you commit, since pricing and plans are moving. You do not need a heavy platform to start; GA4 with Google’s testing tools, PostHog and other lighter options let you begin without a large licence.

In-house. A dedicated CRO manager, an analyst and a designer is a six-figure annual commitment before you have paid for a single tool. Add engineering time to build the winners and the loaded cost climbs further. This only breaks even with serious volume behind it.

Agency. Retainers typically run from a few thousand to the low tens of thousands a month, with most mid-tier engagements in the mid four figures. For mid-market brands, a good agency can deliver output equivalent to an in-house team at a fraction of the loaded cost, which is why it is the common middle path.

How teams waste money on CRO

The failures are predictable. Avoid these and you have removed most of the downside:

1. Testing on pages without enough traffic, then trusting noise as if it were signal.
2. Peeking and stopping tests before they reach significance.
3. Testing everything instead of prioritising high-value hypotheses, burning velocity on trivial changes.
4. Building and shipping “winners” that were never real, wasting developer and design time.
5. Chasing 20 to 30% overnight uplifts when real median winners are low single digits.
6. Buying an enterprise testing platform before having the traffic to use it.
7. Skipping qualitative research, so every test is a guess. Heatmaps, session recordings and surveys come first; our CRO best practices page goes deeper on sequencing.

What to do if you do not have the traffic

Low traffic does not mean you ignore conversions. It means you stop trying to “prove” things statistically and start reducing uncertainty instead. Fix obvious friction, sharpen your value proposition and calls to action, cut unnecessary form fields, add credible social proof. Use heuristic and qualitative methods that do not need a large sample: heatmaps, session replay, user surveys, and rapid five-second tests with people who match your customer profile.

Every visitor is worth proportionally more on a low-traffic site, so these fixes still pay. Just be honest that they are informed judgement, not validated experiments. The broader method sits in our conversion rate optimisation guide and the structural work of conversion funnel optimisation.

A decision framework

Run through these before you spend anything:

1. Volume. Can a key page generate roughly 100 to 350 conversions per variant within four to six weeks? If not, do heuristic CRO, not A/B testing.
2. Baseline. Is your conversion rate already strong for your sector? UK ecommerce averages sit around 3.4%, with live SME panels such as IRP Commerce running lower at the median. A lower baseline means more headroom and a better case for CRO.
3. Margin per conversion. High average order value or lifetime value means even a tiny uplift repays the cost. Low margin and low volume rarely does.
4. Velocity capacity. Can you ship two or more tests a month with developer support? One test a quarter will not compound.
5. Cost model match. Under 100k visits, stay DIY or heuristic. Mid-market, an agency retainer fits. Above 100k visits and £1M+ in revenue, in-house starts to make sense.
6. Budget for failure. Expect roughly one in four or five tests to win, with median wins in the low single digits.

So, is CRO worth it? If you clear the traffic and velocity bar, yes, because compounding modest wins on existing traffic is one of the highest-return activities available to you. If you do not clear it, formal testing is not worth it yet, and the smart move is heuristic improvement until your traffic grows into a real programme.

Frequently asked questions

Is CRO worth it for a small business or low-traffic site? Usually not as a formal A/B testing programme, because you will not reach statistical significance. It is still worth improving conversions through heuristic fixes, qualitative research and clearer calls to action; just treat those as informed judgement rather than validated experiments.

How much traffic do you need to A/B test? As a rough guide, at least a few thousand monthly visits to the page and around 100 conversions per variant, with 350 to 400 per variant giving more reliable 95% confidence. If a test would take longer than six to eight weeks to reach significance, you do not have enough traffic for that test.

What percentage of A/B tests actually win? Far fewer than people expect. Reported win rates range from about 12% on primary metrics in large datasets up to 25 to 30% in some industry benchmarks. Budget for 70% to 90% of tests producing no winner, which is why test volume matters so much.

How much does CRO cost? It depends on the model. Tools run from a low-hundreds-per-month tier up to enterprise contracts in the tens of thousands a year. Agencies typically charge a few thousand pounds a month upwards. A full in-house team is a six-figure annual commitment before tooling, which is why it only suits high-traffic, high-revenue sites.

What uplift is realistic from a winning test? Modest. Median winning tests deliver roughly a 2.77% revenue-per-visitor uplift, with the top quartile around 5.21%. The 20 to 50% gains in published case studies are outliers, so plan around small, compounding wins rather than one transformative result.

When is CRO not worth it? When you cannot generate enough conversions per variant to reach significance in a reasonable window, when your margin per conversion is too low to repay the effort, or when you can only run a test or two a year. In those cases the money is better spent growing traffic and fixing obvious friction first.