A Practical Guide to Marketing Attribution

Attribution sounds like it should have a correct answer: which marketing touch caused a sale. It doesn’t. Every attribution model is a rule for dividing credit across a customer’s touchpoints, and different rules produce different answers from the exact same data. Understanding how each model works, mechanically, is the only way to know when to trust one and when it’s misleading you.

The rule-based models

The simplest models apply a fixed rule to the sequence of touchpoints a customer had before converting.

Last-click gives all credit to the final touchpoint before conversion. It’s the default in a lot of reporting because it’s the easiest to compute: find the last recorded interaction and attribute the whole conversion to it. The mechanical problem is that it treats the touch that happened to close the sale as the only one that mattered, so it systematically rewards channels like branded search and retargeting that tend to show up late in a journey, and erases channels like display or social that tend to show up early.

First-click flips this, giving all credit to the first recorded touch. It rewards discovery and awareness, and just as mechanically ignores whatever closed the deal.

Linear attribution splits credit evenly across every touchpoint in the recorded journey. It’s a fairer split mathematically, but it makes no distinction between a touch that was decisive and one that was incidental, so a channel that shows up often but contributes little gets the same weight as one that rarely appears but reliably closes.

Time-decay and position-based models are variations on the same idea: time-decay weights touches closer to conversion more heavily, and position-based gives extra weight to the first and last touch and splits the rest evenly among the middle. Both are still fixed rules applied regardless of what happened in a given customer’s journey.

Data-driven attribution

Rather than applying a fixed rule, data-driven attribution uses statistical modeling, often a variant of Shapley value calculation borrowed from game theory, to estimate each touchpoint’s actual marginal contribution based on patterns across all your converting and non-converting paths. It compares journeys that included a given touchpoint against similar journeys that didn’t, and credits that touchpoint based on the measured difference in conversion likelihood. This requires enough conversion volume to be statistically meaningful, which is why smaller advertisers historically couldn’t use it, though platforms like Google Ads have expanded access to it over time.

Where every touch-based model runs into the same wall

All of the models above depend on being able to observe a customer’s touchpoints across channels and tie them to the eventual conversion. That observation increasingly has gaps: browser restrictions on cross-site tracking, app-level permission prompts and cookie consent all mean a real share of the customer journey simply isn’t visible anymore. A model can only assign credit to what it can see, so as visibility shrinks, every touch-based model becomes less complete, not just less accurate.

The alternative: measuring without tracking individuals

Marketing mix modeling sidesteps the tracking problem by working with aggregate data instead of individual journeys: total spend by channel, total sales and external factors like seasonality or price changes, fed into a statistical model that estimates each channel’s contribution to overall results. It doesn’t need to see any single customer’s path, which makes it resistant to the same privacy changes that erode touch-based attribution.

Incrementality testing goes further and establishes causation directly, by withholding a channel from a randomly selected portion of the audience and comparing outcomes against a group that received it. Whatever difference shows up between the two groups is the channel’s true incremental effect, independent of any attribution model’s assumptions.

Choosing among them

No single model gives you the truth. A reasonable approach is to use a touch-based model like data-driven attribution for everyday channel comparisons, since it updates quickly and is directionally useful, while periodically validating your largest spending channels with an incrementality test, since that’s the only method that tells you what would have happened without the spend at all.