Attribution model

The model quietly decides where the next budget goes. Under last touch, retargeting and branded search look brilliant while the post that created the interest gets nothing. Under first touch, awareness channels look strong while whatever closed the deal is ignored. Neither is false, and neither shows cause and effect: a model divides credit, it does not prove that a touch changed the buyer's mind.

• Start with last touch as a baseline, then add first touch and compare the two reports.
• A channel strong on first touch but weak on last touch is a discovery channel. Judge it on that job.
• Move to multi-touch attribution once every touch is logged against the deal, for example with UTM tags on each link.
• Set the attribution window at the same time, since it limits which touches count at all.
• Check any model against a test, such as pausing a channel in one region, before moving large budgets.

Most buyers meet a business several times before they pay: a post, a search result, a newsletter, a demo. An attribution model is the rule for dividing the credit for the sale across those touches. Only the rule changes, and with it the story about which channel works.

The common models are:

• Last touch: all credit goes to the final interaction before the sale. See last-touch attribution.
• First touch: all credit goes to the first interaction. See first-touch attribution.
• Linear: every touch gets an equal share.
• Time decay: touches closer to the sale get more credit.
• Position-based: the first and last touches get the most, the middle shares the rest.
• Data-driven: software estimates each touch's weight from past conversions, and needs plenty of them.

• Treating the model's output as proof of cause. It divides credit, it does not show that a touch changed anyone's mind.
• Relying on last touch alone and starving the channels that created the interest.
• Changing the model between reports, so that trends reflect the rule and not the market.
• Trusting each platform's own attribution, when every platform credits itself.
• Adding a data-driven model with too few conversions to learn from.
• Moving large budgets without a test, such as pausing a channel in one region.