Why Multi-Touch Attribution Fails Without Complete Data

Multi-touch attribution is supposed to answer the most important question in marketing: which channels and campaigns actually drive revenue? The models — linear, time-decay, position-based, data-driven — are sophisticated. The math works. But the data feeding the models is fatally incomplete.

Attribution requires complete journeys

For any attribution model to work correctly, it needs to see the complete customer journey — every touchpoint from first awareness to final conversion. A typical eCommerce purchase might involve:

A multi-touch attribution model would distribute the €120 across all four touchpoints based on the model logic. But here is the problem: if the visitor rejected cookies on Day 1, the first three touchpoints are invisible. The attribution model sees only the direct visit on Day 7 and assigns 100% of credit to “direct.”

The data gap

In the EU, traditional analytics lose part of actual traffic to consent banner rejection, ad blockers, browser cookie restrictions, and data sampling. In the compounded worst case, only about 13% of visits survive. Measured on a real Shopify store over 48 days, GA4 did not record 29% of visits, and the loss was uneven: Sealmetrics saw 11% more direct traffic than GA4, but 62% more organic search and 133% more organic social. This means your attribution model is seeing a biased subset of touchpoints and making conclusions about budget allocation.

The consequences are predictable:

• —Direct traffic is inflated — it absorbs all untracked touchpoints
• —Top-of-funnel channels (organic, social, display) are systematically undervalued because first touches are most likely to be lost
• —Email and retargeting are over-credited — they tend to be later in the journey when cookies are more likely to be active
• —Budget allocation follows the bias, reinforcing spending on channels that appear to perform better simply because they are more visible to cookies

Google’s data-driven attribution is not the answer

GA4’s data-driven attribution (DDA) uses machine learning to distribute credit across touchpoints. It is technically advanced, but it has a fundamental limitation: it can only learn from the data it has.

If a share of touchpoints is missing, the ML model learns patterns from a biased sample — the visitors who accepted cookies, did not use ad blockers, and had persistent cookie storage. The model then extrapolates these patterns to the entire population. It is a sophisticated answer to the wrong question.

What complete data changes, and what it does not

Measuring traffic without depending on consent with cookieless analytics does not rescue multi-touch attribution. A user-level model needs to connect a person’s visits across days, and that requires a persistent identifier. Without one there is no journey to split credit across, and Sealmetrics deliberately does not create one.

What complete data changes is the base the simpler model runs on. Sealmetrics credits each conversion to the source of the session in which it happens, by last click, whether or not the visitor accepted cookies, not only on the fraction that did. The channel totals are measured rather than extrapolated, and they can be checked against real orders.

It does not give first touches the credit a multi-touch model would. Organic search, social and display still receive credit only for the sessions in which they close a conversion. When the question is how much those channels contribute earlier on, the honest tools are incrementality tests and marketing mix models, compared in last-click vs modelled attribution.