The funnel
Share of visits reaching each step, in order: a visit counts at a step only if it reached the step before first. Dates start on 25 November, because add to cart was not recorded for most of November.
Groups of visits
25 November to 31 January. "Like-for-like" puts every group on the same mix of new and returning visitors, because returning visitors buy about five times as often and some groups have more of them. "Steadied" pulls small groups toward the overall rate so they don't top the list by luck. These are patterns, not effects: nothing here was randomly assigned.
Test backlog
Each change is aimed at one group of visits, sized on January traffic. The lift is an assumption, not a finding: nobody knows what a change will do until it is tested. Ranking is extra orders a month divided by effort.
How long the first test needs
Files
The backlog file has each change at the 2%, 5% and 10% lifts used in the report. The group file has every group in every grouping, with raw, steadied and like-for-like rates and the rates at each step. Every figure on this page and in the report is in the JSON file. The raw visit data is not shared here; the study folder has the queries that rebuild it from Google's public sample.