What does attribution do?
A conversion is rarely the product of a single advertisement. A buyer might watch a video, see a display impression, click a paid social post, and finally arrive through a branded search before purchasing. Attribution is the bookkeeping that decides how the credit for that one sale is divided across those touchpoints. It is the foundation of every channel report, and the rule it applies quietly determines which channels look like winners.
The crucial point is that attribution describes a credit allocation, not a measurement of cause. It works from observed conversion paths and a set of rules for splitting credit along them. Change the rules and the reported performance of every channel moves, even though the underlying behaviour of customers has not changed at all. That sensitivity is why the choice of model deserves far more scrutiny than it usually receives.
Common models and their biases
The most widely used model is last-click, which hands all the credit to the final touchpoint before conversion. First-click does the opposite, crediting the channel that opened the journey. Linear splits credit evenly across every touchpoint, while time-decay weights recent touchpoints more heavily. Data-driven attribution distributes credit according to a model of each touchpoint’s contribution rather than a fixed rule.
Each model carries a structural bias baked into its design. Single-touch models over-reward whichever position they favour and erase everything else on the path. Rules-based splits like linear or time-decay assume a credit pattern rather than measuring one, so they distribute credit confidently without ever testing whether that distribution reflects real influence. Data-driven models are more sophisticated, but they still operate inside the click path - they cannot see the demand a channel created among people who never clicked, and they remain blind to the delayed carryover captured by adstock.
Why last-click over-credits harvesting - a worked example
Suppose a brand runs £50,000 a month across upper-funnel video and lower-funnel brand search. Under last-click, almost every sale is credited to the brand search term the customer typed in the moment before buying, so brand search looks like the hero and posts a spectacular return. The obvious conclusion is to pour more budget into search and trim the video.
A data-driven or incrementality view tells a different story. It reveals that the upper-funnel video is what created the demand in the first place - the customers were searching the brand precisely because the video had reached them earlier. Brand search was simply harvesting intent that already existed. Cut the video on the strength of the last-click report and brand search volume quietly collapses a few weeks later, because the pipeline feeding it has been switched off. Last-click did not lie about which channel closed the sale; it lied about which channel caused it.
Why this matters for media planning
Media budgets are set on the basis of how channels appear to perform, so the gap between attributed value and incremental value is not academic - it directly misallocates spend. Attributed value is the credit a model assigns; incremental value is the revenue that genuinely would not have happened otherwise. Planning on attributed averages systematically over-invests in harvesting channels and starves the demand creation that feeds them, which is exactly why incrementality and iROAS belong at the centre of any serious budget decision.
Attribution and causal measurement are complements, not rivals. Media mix modelling estimates each channel’s contribution from aggregate spend and outcomes over time, and incrementality testing measures causal lift directly through holdouts and geo experiments. Together they correct the blind spots of click-path attribution. The decision that matters is always marginal - what the next pound will create, not what the last campaign was credited with - which is the core of forecast-led paid media forecasting for agencies. Pressure-test the efficiency of additional spend before you commit it with the marginal ROAS calculator.
Related terms
- incrementality - the causal lift advertising creates, measured against a baseline that excludes it.
- iROAS - the incremental revenue advertising actually creates, net of baseline demand.
- media mix modelling - estimating channel contribution from aggregate spend and outcomes over time.
- adstock - how advertising impact carries over after exposure.
Frequently asked questions
What is attribution in marketing?
Attribution is the method used to assign credit for a conversion to the marketing touchpoints that preceded it. Because a single sale is often touched by several channels, attribution decides how that credit is divided - and the model you choose, whether last-click, first-click, or data-driven, materially changes how each channel appears to perform.
What are the main attribution models?
The common models are last-click (all credit to the final touchpoint), first-click (all credit to the first), linear (credit split evenly), time-decay (more credit to recent touchpoints), and data-driven (credit distributed by modelled contribution). Each carries a structural bias: single-touch models over-reward whichever channel sits at their chosen position, while rules-based splits assume a credit pattern rather than measuring one.
Why does last-click attribution over-credit harvesting channels?
Last-click hands all the credit to the final touchpoint before conversion, which is usually a high-intent harvesting channel such as brand search or retargeting. It says nothing about the upper-funnel media that created the demand in the first place. As a result, last-click systematically over-rewards channels that capture existing intent and under-rewards the channels that generated it.
What is the difference between attributed and incremental value?
Attributed value is the credit a model hands a channel for conversions it touched; incremental value is the revenue that would not have happened without that channel. They are not the same. A channel can carry high attributed value while creating little genuine lift, because attribution counts conversions that would have arrived anyway.
How do MMM and incrementality complement attribution?
Attribution allocates credit within observed conversion paths but cannot prove causation. Media mix modelling estimates each channel’s contribution from aggregate spend and outcomes over time, and incrementality testing measures causal lift directly through holdouts and geo experiments. Used together, they correct the blind spots of click-path attribution and reveal where spend actually creates demand.