I worked with the educational vertical for an audience from northern Europe, traffic there is usually cautious, and any unnecessary action on the feed immediately cuts CR. On the first test cycle, the indicators froze at about 3%, and the cohorts looked loose. I started to dig deeper: I looked at the session sections, and saw that users often return, but do not complete the form. I removed one step and transferred part of the flow to a softer prelander.
This review reflects the subjective opinion of the user, and not the official position of the editors.
You can see a deep understanding of the audience and careful work with the funnel. It is really critical for Northern Europe to minimize unnecessary steps, and switching to a softer blender in such a situation seems like a logical solution.
The analysis of session sections and returns provides a more accurate picture of behavior than the dry CR figures, and allows you to optimize the user's path without pressure on traffic. The platform is designed for such point-by-point changes and quiet refinement of bundles.
It's nice that the tools allow you to flexibly adapt to the specifics of GEO and work on quality, rather than on sudden jumps in performance.