Step 1 (days 1–2): baseline, always-gated list and thresholds in Setup
The first step with a Conversion FTE is to record the baseline and fix the rules before any test. The baseline is revenue per visitor, conversion rate and checkout completion by device, read from the store's analytics for the last ninety days, plus average order value from orders. The always-gated list is written now: price, discounts, checkout fields, payment options, legal copy, and anything else the owner wants a person to decide.
The owner also sets the smallest lift worth detecting and the significance threshold they accept, because those two numbers decide how long every test will run at the store's traffic. The Conversion FTE never asks for what onboarding already captured, so Setup is short: the gated list, the thresholds, the testing tool, and who approves.
Step 2 (days 3–7): review the drop-off map in the Funnel tab
The second step is the drop-off map, which the Conversion FTE builds from analytics and orders and reports in the Funnel tab. It shows sessions arriving and leaving at product page, cart and checkout by device, the revenue at stake at each leak using the store's own order value, and the signals behind each: exit pages, error events, speed, help-desk pre-purchase questions and session recordings where the store has them. Hypotheses are ranked beneath it by revenue at stake and evidence.
The owner's job is to read the ranking and set the order. Accept it, move a step the business cares about up, or exclude a step from testing for now. The owner also answers the map's requests, most often for access to session recordings or the testing tool.
Step 3 (days 8–14): approve the first experiment in Approvals
The third step is the first experiment proposal in the Approvals tab. It contains the hypothesis with its evidence, the exact change previewed on mobile and desktop, the metric, the sample size required at the store's traffic for the chosen minimum effect and threshold, the expected duration, the stop rule and the rollback. The owner approves, edits the change or the stop rule, or sends it back with a note.
At the store's traffic level the sample size may imply weeks, and the proposal says so plainly. A store with modest traffic runs one or two tests in month one, on the largest leak, and that is the right number; ten tests read early would be ten stories.
Step 4 (days 8–30): let the test run to its planned read date
The fourth step is patience, enforced by the mechanism. The Conversion FTE does not read a running test before its planned sample size, and the morning brief in the Today tab shows the test as running with its planned read date, not with an interim result to be tempted by. It does monitor for breakage: a variant that errors on one device or a tracking failure is reported immediately and the test is paused with a proposal.
While the first test runs, the FTE keeps the drop-off map current and adds evidence to the next ranked hypotheses, so the queue for month two is ready when the first result is read.
Step 5 (day 30): the review in Performance and the autonomy decision
The fifth step is the monthly review in the Performance tab. If the first test reached its sample size, the review shows the result with the observed difference, the confidence interval and the revenue effect over the period, and names a no-lift result as a loss. If it has not yet reached sample size, the review says so and shows the read date. The review also shows revenue per visitor and checkout completion against the baseline, with the caveat that a single month is not a trend.
The autonomy decision follows the approval record. With one or two tests it is usually too early to promote anything, and the review says that. Copy and layout tests inside the remit are the class most likely to earn autonomy later; the always-gated list never does.
What does done look like after 30 days?
Done after thirty days with a Conversion FTE looks like a drop-off map the owner trusts, one experiment run and read honestly, win or loss, a queue of ranked hypotheses with evidence for month two, and a remit that has grown by the rules the send-backs taught it. It looks like a store that has stopped redesigning on opinion.
Done does not look like a headline lift. One valid result is worth more than a month of before-and-after comparisons, and the first review is honest because it says exactly what was learned.