There is a sentence I refuse to put on my own website, even though it would convert: "we guarantee 100% data accuracy." I refuse it because it is impossible, everyone technical knows it is impossible, and an industry that keeps saying it anyway is the reason clients arrive at my door suspicious of everything, including me.
So here is the honest version of the accuracy conversation, the one worth having before any tool decision or agency contract.
Why 100% is structurally impossible
Web analytics runs on a script in a browser you do not control, on a network you do not control, governed by a consent choice you do not make. Each layer takes a bite:
- Ad blockers and tracking protection stop the measurement script outright for a meaningful slice of visitors, skewed toward technical and younger audiences. Your site's exact share depends on who visits it; a developer-tools store and a garden center live in different worlds.
- Consent, when honestly implemented, means declined visitors are measured minimally or not at all. That is not a bug. That is the system working. (And when it seems like you kept full data after installing a banner, check what the banner actually blocks.)
- Browsers themselves cap cookie lifetimes and trim cross-site memory, so returning visitors reappear as new, and journeys fragment.
- Your own implementation drifts with every release, which is the only loss category fully in your control and the one nobody budgets attention for.
- The tools add their own fog on top: GA4 applies thresholding to hide small segments and sampling on complex queries, so even collected data is sometimes shown to you approximately.
Multiply the layers and a flawless count of humans is off the table permanently. Any vendor pitch built on restoring it (several "cookieless, 100% accurate" products lean this way) deserves the follow-up question: measured against what, exactly? To claim 100%, you would need a perfect reference count, and if a perfect reference count existed you would not need the product.
What accuracy is actually achievable
Decision-grade accuracy: a setup where the numbers are consistent with themselves over time, the error is roughly stable, and its size is known. That is genuinely reachable with disciplined implementation, and it is enough for nearly every decision a business makes:
- Trends hold. If traffic really rises, measured traffic rises. The undercount subtracts a stable-ish layer; it does not invert directions.
- Comparisons hold. Channel A versus channel B, this landing page versus that one: both sides carry similar bias, so the ranking survives.
- Ratios mostly hold. Conversion rates are dented (blockers cut the numerator and denominator unevenly), but movements in them remain meaningful.
What decision-grade data does not support: treating analytics as accounting. Revenue lives in your shop and your books. Analytics is the instrument panel, not the ledger, and the companies that get this right reconcile the two on purpose instead of discovering the gap in a panic.
"Close enough" has a number, and it is yours
Here is where I depart from most writing on this topic: you do not need an industry statistic about average data loss. Published figures are vendor content with unknowable sampling, and your real number is measurable in an afternoon. Take last month's orders in your shop backend, take GA4's purchase count for the same window, and divide. That ratio is your capture rate. Track it monthly.
A stable capture rate, whatever its level, means your instrument is calibrated: multiply through when you need absolute estimates, and read trends directly. A moving capture rate is a fire alarm with a date on it, and the verification procedure is how you find what changed.
The uncomfortable corollary
Once you accept the gap exists, "our data says X" becomes "our sample suggests X," and some decisions stop being data decisions at all. A 2% difference between two campaigns is noise wearing a percentage sign. The discipline is knowing your margin and refusing to read signals smaller than it. Fewer certainties, held more honestly, beat precise-looking numbers nobody verified.
That reframe, more than any tool, is what separates companies whose analytics earns trust from companies with the five data lies on rotation. And if you do not currently know your capture rate, that is the single highest-value afternoon available to your business this month. It requires no budget, no migration, and no consultant, though if the afternoon turns up a gap that moves around, finding out why is precisely the kind of thing I audit for a living.