Which design helps people
take the next step?

A follows the reference. B tests a more direct visual hierarchy. C tests a warmer presentation. Each page has a matching form, the same fields, offer, images and CTA wording. No version promises an unverified subscription or booking.

What the numbers mean

• CTA click-through: visits with at least one tracked primary CTA or “See how it works” click divided by visits. Repeated clicks in one visit count once.
• Form completion: visits with a durably recorded request divided by visits that opened the form. It separates curiosity from willingness to contact us.
• Visit-to-request: submitted-request visits divided by visits. A submitted request is not yet a qualified lead or a sale.
• Active time: time on a visible page while the visitor has interacted in the last 30 seconds, capped at ten minutes per visit. Idle tabs do not keep adding time.

What design science can tell us

Make the action legible, easy to target and clear about what happens next. Our buttons target at least 44 CSS pixels in height with visible focus states. Contrast and target size are accessibility considerations; they do not prove a particular button color or corner radius increases sales. W3C WCAG 2.2.

Evaluate outcomes alongside diagnostics. More clicks with fewer completed forms can mean the page attracts attention but sets the wrong expectation. More time can mean engagement or confusion. Review form failures, page speed and lead quality before acting on a rising metric. Microsoft’s trustworthy experimentation guidance.

How allocation changes

New outside browsers begin with equal random allocation. Their first-party cookie keeps their assigned version on refresh and return visits. Employees use verified company logins and cycle A → B → C on refresh; employee traffic is excluded server-side, but submitted test inquiries remain visible in CRM.

After each version has at least 100 visits and seven days of data, a directional score uses 50% click-through rate, 40% request rate and 10% active time normalized to two minutes. New visitors are allocated 60% to the leading score and 20% to each other design. Returning visitors keep their version.

This is a heuristic for traffic allocation, not a statistical significance test. A hundred visits is a warm-up floor, not proof. Repeat visits are not independent people, and adaptive allocation changes the traffic mix. For a formal winner, run a fixed-allocation test with a sample-size calculation based on the observed baseline and the minimum improvement worth acting on. Compare qualified leads downstream, too.

The form and page vary together, so this experiment tests each complete design. A follow-up test can isolate CTA wording, button shape or form layout one at a time. Never attribute a bundle’s result to a single visual choice.

Known limits: cookie deletion and another device create new assignments; rate limits reduce abuse but do not prove human traffic; blocked scripts or a tracking outage can leave untracked inquiries. Metrics contain no form contact details. Source attribution stays on the private CRM record. Raw visit and budget expiry values are metadata until a retention policy is deployed.