Catalog

What each term means here, and how it usually gets misread.

Measurement

What counts as a conversion, and who gets the credit for it.

Attribution

marketing attribution · conversion attribution · attribution model · атрибуція

In marketing analytics, attribution is the rule that assigns a conversion to one or more touchpoints, so what it produces is credit under a chosen rule, not the causal effect of the ad.

The word does at least three different jobs, and most arguments about attribution numbers are really arguments about which job is meant. In marketing analytics it means splitting credit for a conversion between channels. In mobile measurement it means the narrower mechanical step of tying an install to a click or an impression inside a stated window. In publishing it means naming who wrote a thing.

Every attribution figure carries a rule and a window with it. Move the window from seven days to one, or move from last click to a multi touch split, and the same raw events produce a different table while nothing about the campaign has changed. A figure without its rule and its window written next to it cannot be compared with anything, including its own value from last quarter.

Common misreadingAttribution gets read as a measure of what the advertising caused. It is bookkeeping applied to events that were already recorded, so a channel that appears late in the path collects credit for demand it did not create. The stable gap between a platform count and a warehouse count is the same misreading in another form: two systems answering slightly different questions, not a broken pipeline.

Incrementality

incremental lift · true lift · causal lift · інкрементальність

Incrementality is the share of conversions an ad actually caused, established by withholding the ad from a randomised control group rather than by reading an attribution report.

Attribution answers who to thank. Incrementality answers what would have happened anyway. The two questions give different answers whenever a channel is good at reaching people who were going to convert regardless: retargeting and branded search are the usual pair, strongest in an attribution report and weakest in a holdout.

The measurement always needs something to compare against: a randomised group withheld from the campaign, a set of regions where it never ran, or a period when it was switched off. Without a comparison group there is no incrementality figure at all, only a correlation with a confidence interval around it.

Common misreadingThe word gets used as a compliment for a campaign instead of as a measured quantity, so a channel is called incremental with nobody naming the control group, the period or the interval. The second misreading is the short test: an interval that still crosses zero after two weeks does not mean the effect is zero, it means the test cannot tell yet, and stopping there produces a confident number with nothing behind it.

Plumbing

How the event gets from the browser to the warehouse without being lost or counted twice.

Conversions API

CAPI · Meta Conversions API · Facebook Conversions API · серверна відправка подій

The Conversions API is Meta's server-to-server endpoint that sends the same conversion events as the browser pixel, deduplicated against it by a shared event ID.

The pixel and the API carry the same events. The difference is who sends them. The pixel fires from the reader's browser and dies with an ad blocker, a tracking preference or a closed tab. The server sends from infrastructure you control, so an event that never left the browser still reaches the platform.

Both channels normally run at once, which means every event exists twice. Meta collapses the pair by a shared event ID together with the event name. If the browser and the server generate their identifiers independently, nothing collapses and reported conversions quietly inflate. The deduplication key is the integration: the rest of it is plumbing.

Common misreadingThe API is sold as a way around iOS restrictions and lost conversions. It bypasses nothing. Consent still governs what may be sent, and an event the server never learned about stays lost. What it buys is delivery of the events your own systems already hold. Teams that install it expecting recovery read the first week of extra conversions as recovered demand, when part of that lift is the same event counted twice through a broken deduplication key.

Meta Conversions API, official documentation