Maja Voje opens with a line worth stealing outright. Let us A/B test it is one of the most expensive phrases in go to market, because it usually translates to I have an opinion and I want to spend four weeks of engineering time proving it.
This one is co-written with Casey, so we are a party to it rather than a neutral observer. The reason it earns a place here is the framework in the first half, which is about deciding what to test at all.
The framework
Most companies fail the first check
Match traffic source to messaging A visitor arriving from social knows far less about the product than one arriving from a retargeting ad. Where awareness is low you have to show the platform, or people will assume you are something you are not.
Use a unified variable set Start from one objective, say reduce complexity on the pricing page, and make every change fold back into it. Fewer line items, one pricing metric, nothing above the plans. You would not add a quiz just because it might help.
Ask whether the test is meaningful A meaningful test introduces new information or new direct value. Book a Demo against Get Demo does neither. Kit adding subtext on trial length, no credit card and free migrations introduces information. Sage adding an industry and size selector that reroutes you introduces value.
CRO should not be about opinions. It should be about engineering context that converts.
Pattern one
Specificity in capability is the answer
Casey read the headlines of thirty companies selling agents. It took until the eighth before one said anything concrete. The rest were variations on deploy, orchestrate and govern fleets of specialized AI agents.
The one that worked named the job and the number: an agent that handles qualification, discovery and booking, with teams seeing 30 percent more qualified pipeline from the same traffic.
Across hundreds of tests on AI messaging, from OpenAI to IBM to GoDaddy, vague positioning loses and capability oriented language wins. Support agents provide 24/7 ticket coverage in 42 languages beats AI powered anything.
Pattern two
Segmentation is moving to the top of the page
MongoDB runs an experience selector tied to its two ICPs. Pick developer and the hero CTA goes to documentation. Pick business leader and it becomes pricing. Sage does the same by industry and size rather than role.
Most sites bury who this is for far down the page. For enterprise brands selling into a committee where a developer and a CFO want opposite things, moving it up is the whole point.
Pattern three
Cost signalling, borrowed from biology
Ethan Decker, a biologist turned marketer, gave Casey the frame. An elk spends real resources building antlers, which is precisely why they signal anything. A signal that is cheap to produce carries no information.
The easier it is to fake, the less credibility it carries.
Which explains why logo bars, star ratings and quote testimonials lose while case studies and video testimonials win. A video of a named executive at Orangetheory, as Webflow uses, cannot be conjured. A logo can.
The closing advice is for teams without Adobe traffic. Consistency beats volume: a thousand visitors a month from a stable source gives cleaner signal than ten thousand from viral spikes. And every pattern here was checked on mobile, where the results mirrored desktop.