Attribution Model

Before buying, someone might watch a video ad, search a few days later and read a blog post, then come back through an email link and make a purchase. How much each touchpoint contributed to the sale determines where the budget goes. An attribution model is the rule-based answer to that question.

Rule-based models distribute credit with a fixed logic: all of it to the last click, all of it to the first click, or equally across every touchpoint. Data-driven models compare converting and non-converting paths to estimate each touchpoint's contribution statistically. Since 2023, Google Analytics 4 and Google Ads have mainly offered the data-driven model and last click. In every case, the model's reliability depends on the quality of the data layer.

• Conversion definition: What counts as a valuable outcome and how the value of that outcome is calculated.
• Touchpoints: The channels and interaction types taken into account: clicks, views, email opens and so on.
• Lookback window: How far before the conversion touchpoints are considered.
• Credit distribution rule: Last click, first click, linear, position-based or data-driven distribution.
• Identity and consent coverage: How far the same user can be matched across devices and sessions.

Consider an e-commerce brand advertising on both search and social media. In the last-click report most sales are credited to the branded search campaign, while social campaigns look weak. The team cuts the social budget and, a few weeks later, sees branded searches falling too.

The reason is that social media was creating demand and search was only capturing it. Comparing different attribution models and validating them with an incrementality test makes the real role of each channel visible. The same data also changes how ROAS should be interpreted.

The measurement foundation comes first: consistent UTM parameters, a single conversion definition, consent management that works correctly and, where possible, revenue data from the CRM. The model and lookback window in the tool being used are then chosen deliberately and stated clearly in reports.

Comparing how different models distribute the same period shows which channels create demand and which capture it. Before major budget decisions, these readings are validated with incrementality experiments such as geographic tests or control groups.

An attribution model shows how credit for conversions that happened is shared among touchpoints, but it does not tell you whether a conversion would have happened anyway without an ad. Incrementality testing asks exactly that question: it compares similar groups that did and did not see the ads to measure the additional outcome the advertising created. Attribution is used for day-to-day optimisation, incrementality for major budget decisions. Reading both together is the common ground of our performance marketing and advanced analytics work.