What perfect tracking would require
For a report to capture every action, every visitor would have to load and keep your tracking script, accept every cookie, stay on one device and one browser from first click to purchase, and complete the action inside whatever window the platform counts. Not one of those conditions holds in the real world.
Blockers strip tracking requests before they leave the browser. Refused consent stops tags firing at all. Browser privacy features shorten the identifiers that link one visit to the next. People research on a phone, ask a colleague, and buy on a laptop. Plenty of sales finish on a phone call, in a shop or over WhatsApp, where no script is watching. Platforms fill some of the resulting holes with modelling, which is an estimate honestly labelled as one.
Why the gap matters
The direction of the error is predictable: measurement undercounts. The exception is the opposite problem, where two advertising platforms each claim the same sale and their totals overstate what the business earned. Both distortions are structural, not faults in your setup.
Where it really bites is expectation. A client promised complete accuracy will read every discrepancy as incompetence, and every reporting conversation becomes a defence. A client told at the start that GA4, the ad platforms and the accounts will never agree, and why, treats the numbers as instruments for making decisions. Say it before the first report, not after someone questions a figure.
Where the promise of total accuracy goes wrong
The obvious harm is an agency that sells it. The subtler harm is the time lost chasing it: rebuilding a working tag setup every time a discrepancy appears, or trying to make GA4 and Google Ads agree when they count on different dates, over different windows, with different attribution. If those two ever matched exactly, that would be the surprising result worth investigating.
A quieter cost is neglect. Teams that decide tracking is unreliable stop using it altogether and go back to guessing, which is worse than a directionally correct estimate. Others ignore the offline half of the journey because it cannot be captured automatically, when a simple question at the point of enquiry would have captured most of it.
What to do about it
Agree one source of truth for money — the order or accounting system — and let platform figures explain how that money arrived rather than compete with it. Write the known limitations into the first report as a short standing note, and repeat it whenever a new stakeholder joins.
Then narrow the gap deliberately: send conversions from your own server where the volume justifies it, use consent mode so modelled conversions have a signal to work from, import qualified outcomes back from the CRM, and ask new enquiries how they found you. Each of those recovers a slice. None of them, alone or together, produces a complete picture, and a sound measurement setup is one that is honest about the remainder.