Attribution

What attribution means

Attribution assigns a sale to the touchpoints that came before it. A customer sees an ad, reads a post, opens a newsletter and finally orders. Attribution decides which of those points the revenue is credited to. Because every advertising tool reports its own success, that decision is commercially relevant: it determines which channel receives more budget and which gets cut.

The common models

With last click the final touchpoint receives the entire revenue. It is simple, available in every tool and systematically disadvantages everything that happens early in the decision path, such as reach, social media and brand work. With first click it is the other way round. Linear models spread the revenue evenly across all touchpoints, position based models favour the first and the last. Data driven methods estimate the contributions from actual behaviour and need a very large volume of data for that.

There is no correct model. Each is an assumption about how people decide, and each favours the channel whose tool shares that assumption. Knowing this changes how the numbers are read.

Why the assignment has become difficult

• Consent: without permission, click paths are missing. What is not recorded cannot be credited.
• Several devices: the search happens on a phone, the purchase on a desktop. The path breaks exactly where the decision was made.
• Platforms count twice: Google and Meta see the same order and both report a success.
• Models change: platforms adjust their estimates without the shop changing anything. The year on year comparison limps.

What helps in practice

Three rules keep the effort small. First, add up the revenue all channels report and compare the sum with the actual revenue. The difference shows how strongly it is double counted. Second, judge channels additionally with an experiment, for instance a region without ads. Third, look at the contribution margin per channel, not only the revenue. A channel with high revenue and a weak margin is not a success.

Frequent questions

• Which model should I use? Last click for steering, data driven for judging, and both with caution.
• Are the platform numbers wrong? Not wrong, but partial. Every platform counts itself favourably.
• Does attribution work without consent? Only to a limited degree. Without data, estimates are what remain.