Conversion Rate Optimisation (CRO) fundamentals are basically solved.

The maturity data for the CRO industry says it’s in rude health. VWO's 2025-26 Experimentation Maturity Benchmark¹ puts 90% of organisations in the top two tiers: 54% in Strategic and 36% in Transformative. No one remains in Beginner or Aspiring. VWO's own conclusion is blunt: experimentation is now treated as a mandatory growth engine, not an optional feature.

That is a genuine shift. The 2024 VWO x Speero benchmark² had 45% of companies in the top two tiers and headlined its own finding as stagnation. The foundations that used to separate good programmes from average ones (tooling, process, basic rigour, executive buy-in) are now table stakes.

And yet 56% of organisations still ship only one or two tests a month¹.

That is the contradiction worth sitting with. Maturity has never been higher, and throughput has barely moved. VWO call it the execution gap and name it the defining challenge of 2026. They’re right, but the more useful question is why it exists at all, because the obvious culprit has already been dealt with.

Three things moved at once. First, tooling matured from platforms that just run experiments into ones that can plan, build and analyse them end to end, so the toolchain stopped being the bottleneck it used to be. Second, personalisation went from an enterprise-budget nice-to-have to something accessible to programmes of any size. And third, CRO itself has been relocating, moving out of Marketing and into Commercial and Product teams, closer to where the roadmap decisions actually get made.

Underneath all that, the psychology of testing shifted too. Teams are now noticeably more comfortable talking in terms of ROI and incrementality than in raw win rate, and more willing to be bold and accept failed tests as part of the process rather than something to hide. In VWO's benchmarking¹, "feeling safe to fail" came out as the strongest-scoring category of all, at 85%, a clear sign the stigma that used to attach to a "losing" test has largely gone.

Two forces are doing most of the work. AI has lowered the cost of designing and building experiences, so testing more no longer means paying proportionally more. VWO's own data shows how deeply that has embedded: teams are not experimenting with AI casually, they rely on it to unblock manual work, from rapid copy generation through to automated friction discovery inside session recordings¹.

And budget pressure has forced a level of new rigour into programmes that wasn't there when testing budgets were more forgiving. You can't justify a low-conviction test anymore, so the average test getting shipped is simply better thought through.

AI has removed the production constraint. Copy, creative, build and analysis are all faster and cheaper than at any point in the discipline's history. And yet less than half of organisations still do more than two tests a month¹, the very same bottleneck the industry has complained about for a decade.

If making things got dramatically cheaper and throughput did not move, then making things was never the bottleneck.

The constraint sits upstream of production, in the decision layer: what gets chosen, what gets challenged, what gets signed off, what gets killed. That is slow, political, and largely untouched by the tooling revolution of the past three years.

VWO's data points at exactly this. Prioritisation is the lowest scoring of all 17 capabilities measured, at 61%¹: the weakest link in an otherwise mature industry. We industrialised the factory floor and left the planning office exactly as it was.

Which reframes the whole velocity question. The answer to "we can only run two tests a month" is rarely more resource. It is usually that nobody can agree quickly enough on which two.

If the fundamentals are solved, the next gains won't come from doing more of the same thing better. They come from changing what you're optimising for.

First, look at prioritisation. It's consistently the weakest of the 17 operational capabilities measured in VWO's benchmarking¹. A formal ICE or RICE model is the fastest available route to higher ROI. A working prioritisation framework is a few weeks of honest work, not a replatforming. And because almost nobody has done it properly, it is one of the few competitive advantages in this discipline still sitting unclaimed.

Second, look at the metric itself. Win rate is a vanity metric. It rewards volume over value, and it's a poor proxy for what a programme is actually worth to the business. The strongest functions run win rates around 10%; if yours is much higher, that's a sign you're testing too safely, not that you're better at it. The metric worth reporting instead is incremental annual return: what the programme actually added.

This resolves the velocity issue. Most companies are stuck bottlenecked at a couple of tests a month¹, nowhere near enough throughput to compound learning at the rate the top tier does.

None of these are things a testing tool can fix. They're organisational choices, not technical ones. And here's why prioritisation carries the most leverage of the three: if AI has already made design, build and analysis fast, whatever's left in a test's timeline is decision time. That’s the weeks spent debating what to test next, defending an idea and waiting for sign-off. A working prioritisation framework compresses that stage specifically. It provides a pre-scored backlog, agreed criteria instead of a fresh negotiation every time. Shrink the slowest part of the cycle and the rest follows. 

Get the sequencing right, and the 2-test-a-month problem largely solves itself.

¹ https://wingify.com/ebooks/experimentation-maturity-benchmark-report/
2 https://wingify.com/ebooks/experimentation-program-maturity-report-2024/