A relaunch always ends with the same question: is it better now? Anyone who wants to measure before the relaunch usually hears that there is no time for it, and that the numbers are in the tools anyway. The second half is even true. Those numbers simply carry an expiry date, and the deadlines are shorter than almost anyone assumes: two months, forty weeks, sixteen months. The rebuild does not delete the past, the retention rules of the tools do. This piece covers which four numbers you have to capture beforehand, how large a difference has to be before it is more than noise, and when you are allowed to measure at all after the switch.
Why do the numbers not come back after a relaunch?
Because each of the three tools that would have to supply the answer has its own memory, and none of them reaches as far back as the comparison would need.
Google Analytics offers two values for event data retention: 2 months or 14 months, with the longer periods reserved for the paid 360 edition. Anything older disappears, and the wording is explicit: "When data reaches the end of the retention period, it is deleted automatically on a monthly basis." There is a qualification that rarely gets mentioned and that works in your favour: the setting "does not affect standard aggregated reports (including primary and secondary dimensions)", it only touches explorations and funnel reports. Your monthly totals survive. The free-form analysis of the period before, with your own segments and paths, is gone one quarter later if the setting sits at two months.
The field data on how fast your site is comes from the Chrome User Experience Report. Google describes it as a "28-day rolling average of aggregated metrics", so always a rolling mean of the last 28 days, never a daily figure. The history behind it reaches back 40 weeks, with 25 weeks returned by default. That is a little over nine months, so less than a year. And for many small sites the number does not exist at all: anything below the threshold is not included in the dataset, since inclusion requires "a minimum number of visitors".
Search Console has the longest memory. Google puts it plainly in its own help pages: "Search Console keeps data for the last 16 months." That is the largest of the three windows, and even that is tight as soon as the comparison is meant to cover two full annual cycles.
Which four numbers do you capture?
First: clicks, impressions and position per individual URL. Not the total for the domain. A relaunch changes how pages are cut and often changes addresses too, and the total will not tell you afterwards which page is carrying the loss. The export of the last twelve months from Search Console, row by row, is the basis for every later statement. It is also the only one of the four numbers that carries the annual cycle with it, which is the answer to the objection "that is just seasonality".
Second: LCP, INP and CLS from field data, not from a lab test. The lab test measures a simulated session on a simulated device, field data measures your actual visitors. Google is clear that the value counts, and equally clear about its limits: "Core Web Vitals are used by our ranking systems", while there is "no single signal". A relaunch swaps out the technology underneath the site, so speed is the value most likely to move, in either direction.
Third: the number of conversions per month, in absolute terms. Not the conversion rate. The rate looks better quickly after a rebuild if fewer people arrive at the same time. And the absolute number decides whether a difference is detectable at all, more on that in a moment.
Fourth: what one conversion is worth. That is the rate at which an enquiry turns into an order, and the average order value. This number has no expiry date because no tool stores it. It sits in the CRM or in the accounts, in most cases never worked out. It is the only one of the four that lets you tell a rise in enquiries apart from better business.
How large does the difference have to be before it is more than noise?
This is the part almost every relaunch review skips. Enquiries arrive irregularly. Two months on the same website never produce the same number, and that gap has nothing to do with the rebuild. How far the two values may sit apart without anything but chance behind it depends solely on the absolute number of conversions.
We ran the numbers for the normal case of a service business, using the standard random model for counted events and the standard 95 per cent interval. At 20 enquiries a month before and after, the count has to almost double before it separates from the noise: plus 86 per cent. At 50 it is plus 48 per cent, at 100 plus 32 per cent, at 250 plus 19 per cent, at 500 plus 13 per cent.
Turn the calculation around and it gets uncomfortable. Proving a 30 per cent improvement with the usual 80 per cent power takes roughly 228 conversions per period. At 20 enquiries a month that is 11.4 months before and 11.4 months after, close to 23 months in total. That is longer than any of the three tool windows from the first section and a multiple of the two months an untouched Analytics setting keeps.
The honest conclusion is not that you should give up on it. It is this: with small counts, the month-on-month comparison is not proof, it is a hint. What counts then are the intermediate measures that occur far more often: sessions, form views, abandoned forms, clicks on the phone number. And a comparison within the same period, as in A/B testing, always beats a comparison between two periods, because it holds season, weather and market constant.
When are you allowed to measure after the switch?
Not in the first week. On moving a site with URL changes, Google writes that a small to medium project needs "a few weeks for most pages to move", larger ones longer. And further: "the visibility of your content in Search may fluctuate temporarily during the move. This is normal and a site's rankings will settle down over time."
For speed there is a hard technical limit: because field data is a mean across 28 days, every value in the first four weeks after the rebuild still contains sessions on the old site. The first clean look at LCP, INP and CLS is possible 28 days after the switch at the earliest, before that old and new are mixed.
What you do have to look at in the first days are not success figures but failure figures: pages that no longer resolve, missing 301 redirects, lost forms, a noindex accidentally carried over from the staging environment. That check is not measurement, it is damage control, and it is the second reason why the figures from week one say nothing about the quality of the new site. Which items keep running after launch is covered in Website costs: building it is the smaller number.
What do you do when the numbers are missing?
That is the normal case, and it is no reason to postpone the rebuild. Three things can be done today, even if the relaunch starts next week.
Set Analytics retention to the higher value: it takes two minutes and only works forwards, so the earlier the better. Pull the URL export from Search Console and park it as a file, independent of any tool. And work out the fourth number once, since it is written down nowhere anyway: how many enquiries turned into orders last year, and what an order was worth on average.
If even that is out of reach, the honest option remains: write down which number is missing, and later do not pretend you had it. A baseline from four weeks is thin, but it is honest and its weakness can be named. An invented comparison figure is worse than none, because decisions get built on it.
Three levers before the next relaunch
First: do the export before the switch, not after. Twelve months of Search Console data per URL as a file, plus the monthly conversion totals. That is half an hour of work and the only part that cannot be caught up later.
Second: write the comparison question down beforehand. Which number should change by how much, and by when. Put that in writing first and the calculation from the third section immediately shows whether your own case numbers can answer the question at all.
Third: take the intermediate measures with you. Form views, abandonments, clicks on the phone number and the email address occur ten to a hundred times more often than conversions. With small counts they are the only route to a defensible statement in reasonable time.
If a rebuild is coming up and you want to know which of your numbers are about to expire, we are happy to look at it together 🙂