Incrementality

Incrementality asks the question attribution cannot answer: what would have happened without the channel? Attribution distributes the credit for the sales that happened – incrementality measures how many of them only happened because of the activity. An affiliate sale is incremental if, without the affiliate’s link, the customer would not have bought, would have bought less or would have bought later. The two measures can point in opposite directions: a channel can win many last clicks and still create few new sales.

The classic example is the discount code search. A customer is at checkout, opens a new tab, searches for the store name plus discount code, clicks on a coupon site and completes the purchase. Last-click attribution credits the coupon site, but the purchase was under way before the click happened – incrementality is low. Conversely, a thorough review on a niche site may have created the entire purchase decision weeks earlier; that effort was highly incremental, even if a later click walked off with the attribution.

Incrementality is measured with control groups. Part of the audience is exposed to the activity, a comparable part is not, and the difference in conversions between the groups is the incremental effect. In practice, geo-split experiments, hold-out groups or pause tests – where a channel is switched off for a limited period – are used. The method needs volume: with a few hundred orders a month the effect drowns in random variation, and the result is noise.

A worked example: if a store sells 1,000 orders in a month in the regions where a campaign runs, and 940 in comparable regions without it, the 60 extra orders – 6 % – are the campaign’s incremental effect. Even if last click perhaps credited the campaign with 300 of the orders, it only created 60.