Last-click attribution gives search engine advertising too much credit. First-click ignores the close. Data-driven attribution with AI attempts to model the actual contribution of each channel. Here is how it works and where it falls short.
If a customer discovers a product on Instagram, compares it on Google, and makes the purchase through email, which channel gets the credit? How you answer that question determines how you allocate your marketing budget. And how you answer it also determines how wrong that can go.
Traditional attribution models are rules, not data analyses. The most commonly used:
All of these models are deterministic rules based on assumptions. They are not based on actual causal relationships.
Data-driven attribution uses machine learning to estimate the actual contribution of each channel based on conversion data. Instead of rules, the model empirically determines which channel combinations more often lead to conversion versus which paths more often end without a purchase.
Google has data-driven attribution available in Google Ads and Analytics 4. Facebook (Meta) has its own variant. But the quality of these models depends on the amount of data, the completeness of the measurement chain, and the model's assumptions.
A mathematically well-grounded approach to attribution is the Shapley value method, originating from game theory. The idea: the contribution of a channel is measured by looking at how many conversions occur across all possible combinations of channels, with and without that channel.
In practice this means: if paid search is always present in conversions but also in non-conversions, its incremental contribution is lower than you might think. If email is always present in conversions but rarely in non-conversions, its contribution is higher.
This model is computationally intensive but statistically fairer than rule-based alternatives. It is available in Google Analytics 4 and in custom Python implementations.
Attribution modelling is becoming harder as tracking restrictions increase. Safari has been blocking third-party cookies for years. Firefox too. Google Chrome limited third-party cookies. iOS 14+ and ATT require opt-in for ad tracking on Apple devices.
This means touchpoints are increasingly invisible in the data. A click on an Instagram ad is measured, but the subsequent website visit is not always captured. That makes attribution inherently incomplete.
AI models can partially correct for this via probabilistic matching and modelling, but they do not solve the fundamental problem of incomplete tracking data.
Marketing Mix Modelling (MMM) is a statistical approach that uses aggregate data rather than individual customer paths. You analyse the correlation between media spend per channel and sales results over time.
MMM is less sensitive to tracking restrictions, but requires years of historical data and provides no customer-level insight. AI makes MMM more accessible and faster to compute, but the model assumptions remain a source of uncertainty.
Mach8 often advises clients to combine digital attribution (for direct channel data) and MMM (for broader media effects) as the basis for budget decisions.
If you really want to know whether a channel contributes to conversion, incrementality testing is the most reliable method. You divide your target audience into two groups: one that sees the channel, one that does not. The difference in conversion is the incremental contribution.
This is more expensive and complex than modelling, but it gives causation rather than correlation. AI helps with the design and analysis of such experiments, but the test setup requires human judgement.
Attribution modelling with AI gives better insights than rule-based alternatives, but is not a perfect representation of reality. Incomplete tracking data, the causation problem, and model assumptions remain challenges.
Want to improve your attribution approach? Get in touch with Mach8 for an analysis of your measurement setup and next steps.
We help you go from strategy to implementation. Schedule a no-obligation call.
Schedule a call