The two halves of attribution
Attribution is usually discussed as one thing. It is two, and they fail for different reasons. The first half is capture: at the moment a stranger becomes a contact, something must record where that contact came from — a click identifier, a set of campaign parameters, a tracked phone number, a form name, or a person typing 'referral from the Hollands'.
The second half is inference: taking that record and deciding what deserves credit for the eventual outcome. Capture can be nearly perfect and inference can still be wrong, because a person who clicked an ad may have already been recommended by a neighbour and merely used the ad as a convenient door.
Source capture versus causal attribution
Keeping these two words apart prevents most attribution arguments.
- Source capture — an observed, recordable fact: this contact arrived carrying these parameters, on this page, at this time.
- Causal attribution — a modelled claim: this campaign is why the contact arrived, and therefore why the revenue exists.
- Capture can be audited. Causation can only be reasoned about, and honest reporting says which one it is showing.
- When a report cannot distinguish the two, it is presenting an opinion with the confidence of a receipt.
Why platform numbers and CRM numbers disagree
An ad platform counts what happens inside its own visibility: an impression, a click, a form submitted on its network. It cannot see whether the lead was reachable, whether it was in the service area, whether the appointment held, or whether the invoice was ever paid.
A CRM sees all of that and none of the impressions. So a platform will happily report thirty conversions on a week the business booked four appointments and sold one job. Neither number is lying; they are counting different events. The disagreement is only a problem when someone treats one as a correction of the other.
What attribution is actually for
The point is budget decisions. Attribution earns its keep when it changes where the next dollar goes — pausing a campaign that produces contacts who never qualify, defending one that produces fewer but larger jobs, or noticing that the cheapest leads cost the most to convert.
That means the resolution only needs to be good enough to rank channels reliably. Chasing a single-lead-level certainty that no measurement system can deliver wastes the time that should be spent acting on the ranking you already have.
The limits worth stating out loud
- Cross-device journeys break identity unless the person identifies themselves; a phone-then-laptop path is usually two strangers to the system.
- Phone calls are matched by number and timing, which is strong evidence, not proof.
- Word of mouth is systematically undercounted because it leaves no parameter behind.
- Long sales cycles outlive attribution windows; a 30-day window silently discards a 60-day decision.