Attribution is lying to you, and that is fine

Run the same sale through Google Ads, Meta, and your email platform's own attribution, and you will get three different, confident claims about which channel caused it — and the three numbers, added together, will usually exceed the number of sales you actually had. Every platform is built to claim credit for outcomes it merely touched, and none of them are lying exactly — they are all measuring something real, just not the same thing, and not the whole picture.

Why every platform claims the same sale

A single customer might see a paid social ad, later search your brand name directly, then click an email link before buying. Meta counts an assisted conversion. Google counts the branded search as its own conversion. Your email platform counts the final click as the cause. All three are technically correct about the touchpoint they own, and all three are answering "did we appear somewhere in this journey," not "would this sale have happened without us" — which is the question that actually matters for a budget decision and the one no platform can honestly answer about itself.

Why this is fine, not a crisis

The instinct is to chase a single unified number that resolves the contradiction — a "true" attribution model that finally tells you the real answer. In practice, no model does this reliably, because the underlying question (what would have happened in the counterfactual world where this ad never ran) is not something any tracking pixel can observe. Treating attribution data as directionally useful rather than precisely true is not a compromise — it is the accurate way to describe what the data actually is.

How we actually make budget decisions with it

Instead of trusting any single platform's claimed conversions, we look at incrementality: pause a channel briefly, or run it in a subset of regions, and measure whether total sales actually move. A channel that shows strong "attributed" conversions but barely moves total sales when paused is mostly claiming credit for sales that would have happened anyway. A channel that looks smaller in its own dashboard but causes a visible drop in total sales when paused is doing real work the attribution model was undercounting. This test costs a week of slightly awkward data and settles arguments no amount of dashboard-staring ever will.

The practical rule

Use platform attribution to catch obvious problems — a channel with genuinely no conversions, a campaign spending against the wrong audience — and use incrementality tests, run periodically, to decide where the next real dollar of budget actually goes. Trusting the dashboards completely and trusting them not at all are both mistakes. The useful position is in between, and it is less comfortable than either extreme, which is probably why most businesses default to one or the other instead.