Most of the tests didn't win. The booking flow still came out ahead.

Before touching either flow we pulled a month of GA4 data per landing page and per channel for both sites, covering every channel that landed on those pages. For the Dutch chain that showed a healthy funnel with uneven feeds: 6.58% of the homepage’s 98,242 September sessions ended in an eye-test appointment, 17.37% of the booking page’s 25,431, and the eye-test step of the booking flow converted at 25.96%. Behind each average sat very different visitors. Email was the largest source on the booking page, 10,981 sessions at 17.29%, and among the five largest sources the rate ran from 8.4% to 23.33%. We ran the same split for the German site.

We also kept an observation log from session recordings and heatmaps. Every observation that held up became a research question or a hypothesis in a fixed form: if we apply X, we expect Y for Z, because of W.

Dutch booking page, September 2024, all channels: 17.37% of 25,431 sessions ended in an eye-test appointment. Among the five largest traffic sources the rate ran from 8.4% to 23.33%.

The Dutch series worked on the step where a visitor picks the type of appointment. A redesign of that selection, re-run in April 2025, took completed appointments from 31.40% to 32.93%, 204 extra appointments net on 13,258 control sessions against 13,366. A follow-up that redesigned the appointment cards and carried the chosen type across them moved 30.67% to 31.13% on 26,401 sessions, too small a gap to call a win. A FAQ on the first step, blocked at first and then run as a smaller follow-up in January 2025, showed no clear difference either.

On the German site one of the clearest wins came from the simplest test. The hypothesis was that the footer under the booking form pulled people out of the flow, so the variant dropped it. VWO called the test after eight days: 4.72% more completed appointments at 99% confidence. The ledger over the full run shows 19,353 control sessions against 19,498 variant sessions, 20.51% to 21.56%, 204 appointments gained. We logged the caveat at the time: a change that moved a step in the flow went live in the same period and may have affected the result. Removing the asterisks from the appointment options won too, 21.04% to 22.18% completed appointments. The biggest relative lift in Germany was on the newsletter, not on bookings. Showing the loyalty club’s benefits on the confirmation page took opt-ins from 0.75% to 1.23% of sessions, 64% more, at 99% significance over 42 days.

The rest did not win. Of nine German experiments, three produced a clear winner: the footer, the asterisks and the confirmation page. A FAQ on the first step clearly lost, 24.02% to 22.60% completed appointments. The store-finder layout, labels on the appointment cards, a second attempt at limiting distractions, buttons outside the appointment accordions and a combined USP-and-asterisk variant showed no difference we could trust. On the Dutch side, a booking flow without header, menu and footer and a version that skipped the eye-test step both ran in 2024, and the team marked both as wins. Their follow-ups were dropped. A desktop-only version of the stripped flow did not win.

Dutch appointment-type selection, redesigned and re-run in April 2025: completed appointments from 31.40% to 32.93%, 204 extra appointments net on 13,258 against 13,366 sessions.

We read every test with the same extended significance calculator and logged it in one ledger: sessions and conversions per side, the delta, the net conversions gained. Each week the in-house team and we went through the running tests, the interim reads, and whatever could have skewed them. We stopped the losers and wrote them up with the same care as the winners, so the next hypothesis started from what we knew.

Taken one by one, the clear winners were modest: 31.40% to 32.93% on the Dutch appointment-type step, 4.72% more completed appointments without the footer in Germany, 21.04% to 22.18% after removing the asterisks. For an optician chain every appointment is a planned store visit, so small gains on a flow that already converted a fifth to a third of its visitors, and real ones. We set up the tests in VWO, read them and kept the ledger.

German footer test: after five days the testing tool read +2.96% at 85% confidence, after eight days +4.72% at 99%. Both reads went into the report, next to the step change that ran alongside.