How attribution works

What a touchpoint is, how Clakta walks back from an order to the ads that caused it, and why its numbers never match what the ad platforms report.

Attribution is the step that turns "we spent this and we sold that" into "this ad caused that sale". Clakta does it from your own data, not from what each ad platform claims.

Touchpoints

A touchpoint is one moment where a visitor met your marketing. Every touchpoint carries a timestamp, a platform, campaign / ad set / ad ids where known, and a flag for whether it was paid.

There are four kinds:

Type	Meaning	Paid?
`click`	A visit carrying a `clkt` ad parameter.	Yes
`view`	A verified ad impression the visitor did not click.	Yes
`pageview`	An ordinary visit with no ad parameter.	No
`event`	A tracked interaction on your site.	No

The paid flag is what separates an ad from an organic visit. Paid-only models ignore everything unpaid; Linear counts it all.

From order to credit

When an order arrives, Clakta:

Collects the journey. Every touchpoint belonging to that visitor, ordered by time.
Applies the window. Touchpoints must have happened before the order and within the attribution window — 7 days for clicks, 1 day for views.
Runs the model. The project's attribution model decides how to split credit between the surviving touchpoints.
Distributes revenue. Each touchpoint's credit multiplied by the order value becomes its attributed revenue.

For every model except Triple Attribution, the credits for one order sum to exactly 1.0. An order with no eligible touchpoint gets no credit at all and is reported as unattributed.

Fractional conversions

Multi-touch models produce fractions, not whole numbers. An order split between three ads contributes 0.33 conversions to each.

This is why a campaign row can read "4.7 conversions". It is not a rounding error — it is four whole conversions plus shares of others. Revenue splits the same way.

Last Click and First Click always assign 1.0 to a single touchpoint, so their conversion counts are whole numbers. If fractional values are confusing your reporting, those two models avoid them.

Why Clakta and the platforms disagree

They will never match, and that is the point.

Each platform only sees itself. Meta cannot know the visitor also clicked a Google ad. Both claim the sale, so summing platforms double-counts.
Platforms count view-throughs generously. A sale that followed an unclicked impression is often reported as a conversion.
Platforms model missing data. Where signal is lost they estimate. Clakta reports only orders your backend actually confirmed.
Attribution windows differ. A platform may use a 7-day click plus 1-day view window by default; other settings shift the totals again.

Clakta deduplicates against your real orders, so its total across all channels can never exceed the orders you actually took.

If you want a number comparable to what the platforms report, switch to Triple Attribution — it deliberately gives every channel that touched the order full credit, mirroring how each platform reports independently.

Changing the model is safe

The attribution model is a reporting choice, not a collection choice. All touchpoints are stored regardless; the model decides how they are read.

Switching it re-reports your entire history immediately. Nothing is lost and nothing is recomputed destructively — you can compare models on the same data and switch back.

Change it from the dashboard toolbar or from any Ads Manager page.

It only affects Clakta metrics

Switching the model changes Clakta-calculated columns. Platform-reported metrics come from the ad platforms and are never modified.

Attribution models

The six models Clakta offers, exactly how each splits credit, and which one to use for the decision you are actually making.

Attribution windows

How long a click or a view stays eligible to receive credit, and what that means when your sales cycle is longer than the window.

Unattributed orders

Why some orders receive no ad credit, how to tell a real organic sale from a tracking gap, and what to fix in each case.