How much of your results do Meta and Google ever see?

The formula, and the assumption in it

We model the lag from click to outcome as exponential, with the median you enter. The share of outcomes landing inside a window of W days is 1 − 0.5W / median. That is the baseline with the fewest assumptions, and it is still an assumption. Real lags have a fatter tail. So the true share inside a short window runs a little lower than this page says. Treat the number as the optimistic case.

The windows, as published

• Meta: click-through attribution is 1 or 7 days. Nothing longer exists for clicks.
• Google: you set the conversion window per conversion action. Default 30 days, up to 90 for most actions. Offline imports count the click date, not the upload date. A late import still lands inside the window if the click did.

Why it matters

The platform optimises on what it can see. A sold event that lands outside the window never trains the account. So the account learns from a fraction of its own results. That fraction is not random. It leans toward the buyers who decided fastest. That is a different population from your buyers.

Worked examples, checked on the page

What it leaves out

• View-through attribution, which has its own window and is off by default in most setups.
• Consent and modelling: Meta fills part of the gap with modelled conversions; how much is not published.
• A sale that closes on a second click belongs to that click, with its own clock.