Somewhere in your company there is a dashboard nobody opens anymore. Not because the business stopped caring about numbers, but because the last three times someone opened it, the numbers argued with each other and the meeting ended with a shrug.
This guide is about fixing that. Not with a new tool, and not with a 12-month data transformation program. With a standard: every number you report has a known definition, a verified collection path, and a documented margin of error. Miss any of the three and you don't have analytics, you have decoration.
Why do your dashboards disagree?
Because they are supposed to. GA4, your ad platforms, and your store each count different events, under different attribution rules, over different time windows, with different exposure to consent rejections and ad blockers. Two dashboards agreeing to the digit would actually be the suspicious outcome.
The problem is not the disagreement. It is that nobody in the room can explain it. Once you can say "Facebook claims every purchase it touched within its window, the store counts orders, and GA4 sits in between because roughly a third of visitors reject tracking," the same numbers stop being a crisis. I walk through the mechanics in why GA4, Facebook and your store disagree on purchases.
Can analytics ever be 100% accurate?
No. Consent rejections, ad blockers, browser privacy features, and plain script failures mean a real share of your visitors never gets measured, and that share differs by audience. Anyone selling you "complete data" is selling you the word, not the thing.
What you can have is data that is accurate enough to bet on, with a known direction of error. "Revenue in the store is exact; sessions in analytics undercount, mostly among privacy-conscious desktop users" is a sentence a business can act on. "The dashboard says 42,000" is not.
There are ways to claw back some of the loss, and they cost money. Server-side tracking is the honest version of that conversation, including when it is not worth it.
What does "data you can trust" actually mean?
Three tests, in order of how often they fail:
- Verified, not configured. The tag manager's preview pane says an event is set up. The network tab says what actually fired, and where it landed. Only the second one counts. Most setups I open have events that pass the first test and fail the second.
- Defined, not assumed. "Conversion" means something different in every platform in your stack, and usually something different to every person in the meeting. Every reported number needs one written definition, including what it excludes.
- Legal, not just installed. If your consent banner blocks some tags and waves others through, part of your data comes from people who said no. That is a compliance problem and a data-quality problem at once. Start with what your consent banner isn't blocking.
Where do you start?
Not with a migration. Tool choice matters less than tool honesty, and a broken GA4 setup migrated to a new platform becomes a broken new-platform setup. If you do suspect the tool itself is wrong for you (data residency, product analytics, cookieless measurement), the tool selection guides compare the real options one pair at a time.
Start by finding out what your setup actually does. Open the network tab, pick your three most important events, and check whether they fire, once, with the values you expect, to the destinations you expect, under the consent states you expect. If that sentence sounds like work: that is the work. It is exactly what my Clarity audit does hands-on in three to five days, and the report is written so any competent developer can fix what it finds, with or without me.
The pages under this guide each take one narrow question and answer it honestly, including the answers that cost me projects. That is the point of the whole exercise: analytics you can trust starts with an analyst you can trust.