Every advertiser eventually runs the same exercise. Someone adds up what each platform says it drove, compares the total to what the business actually recorded, and finds a gap large enough to end the meeting.
The instinct is that somebody is wrong. Usually nobody is. The platforms are answering different questions, each internally consistent, none comparable to the others, and none of them the question you actually asked.
Every number carries a definition with it
A reported conversion is not an observation. It is the output of a rule about which conversions a platform is entitled to claim. Change the rule and the number moves without anything in the world having changed. Three of those rules do most of the damage.
The attribution window
Each platform counts conversions that happen within some period after an interaction, and the periods differ. A purchase two weeks after a click lands inside one platform's window and outside another's. The same purchase, the same customer, counted once here and not at all there. Nothing about the customer's behavior explains the difference. The difference is administrative.
View-through
Some platforms count conversions from people who saw an ad and never clicked it. Whether that impression influenced the purchase or merely preceded it is an assumption, and it is being made by the party whose performance depends on the answer. Platforms that count view-through will report larger numbers than platforms that do not, and the gap says nothing about which media worked harder.
Identity
Each platform resolves who a person is inside its own graph. A logged-in ecosystem can follow someone across devices. A tool relying on browser state often cannot. Your analytics platform is usually counting sessions, the ad platform is counting people, and your finance system is counting accounts or orders. Three different units of analysis, all reported as though they were the same unit.
Before asking why two numbers disagree, ask what each one is counting, over what period, and about whom. Most reporting disputes end there, without anyone needing to be wrong.
Why the total is always too big
The arithmetic problem is worse than the definitional one, and it follows from something simple: audiences overlap.
A customer who saw a video ad, later clicked a search ad, and bought that evening can be claimed in full by both platforms. Neither is lying. Each is reporting a conversion that occurred within its window following an interaction it delivered. But the customer bought once, and the revenue exists once.
Sum the platforms and you are counting that person twice, or three times, and you have described a business that does not exist. The clearest symptom is when the total claimed across channels exceeds total company revenue, which happens more often than people admit and is usually explained away rather than investigated.
The bias is not random either. It is largest exactly where you are spending the most, because heavy spend produces more overlap. So the channels most likely to be overstated are the ones already receiving the most budget, and the reporting quietly argues for more of the same.
Your neutral tool is also a convention
The usual response is to trust the analytics platform instead, on the grounds that it is not selling media. That is an improvement in incentive and not in method. A tool that assigns credit to the last non-direct click is applying a rule of its own, one that systematically favors whatever the customer touched most recently. It is neutral about who benefits and still arbitrary about how credit is assigned.
There is no reporting tool anywhere in the stack whose number is the truth. Every one of them is an accounting convention, and the useful question is which convention, applied to whom, over what window. The most common of those conventions has a bias that runs in one direction.
What reconciliation can and cannot do
You can align these. Standardize the windows, decide a single position on view-through, agree what a conversion is and which system is authoritative for it, and the numbers become comparable to each other. That is worth doing, and most companies have never done it.
What it does not do is make any of them causal. Aligning conventions gets everyone counting the same things the same way. It does not establish that the spend produced the outcome, because none of these systems observes what would have happened otherwise. Consistency is an accounting achievement. It is not a measurement one, and the two get confused constantly.
The only way out is an external check. Something that creates a comparison the platforms cannot supply, because it comes from a group that did not receive the media. A geo experiment is the usual instrument, and where geography is not available there are other designs with different tradeoffs. Any of them gives you one number that does not come from a party with a position on the answer, and that number is what everything else gets checked against.
Platform reporting is useful for running campaigns day to day. It was never built to tell you whether the spend caused the sale, and asking it to is the source of most of the disagreement.