A lift seen before the sample is large enough is usually variance rather than effect: with a small sample a handful of orders can push conversion up, then the next batch pulls it back. The first rule of read-out discipline is admitting 'it is not time to read yet' and treating interim numbers as process notes, not conclusions.
Write the planned sample size and read-out date in the experiment log and withhold judgement until then; when a mid-run check is needed, look at data quality and whether tracking is sound, not at the conversion figure. Record the time and number of every interim look so you can later verify whether the call was premature.
This article covers read-out discipline when the sample is short only; it issues no compliance verdict on any experiment. Record interim numbers as they are, write 'sample not reached' when that is the case, note flat numbers as flat, and do not use them to push an early launch or a change of direction.