Every attribution conversation eventually arrives at the same place: the model you're using is wrong. The question is not which model is right, none of them are, but which model is wrong in ways that don't mislead your decisions.
Attribution for direct marketing in 2026 is harder than it's ever been. Third-party cookie deprecation in Chrome has removed a significant chunk of cross-site tracking data. MPP has inflated email open data. iOS privacy changes have reduced mobile app tracking fidelity. The tools that marketers relied on to attribute revenue to channel are producing increasingly unreliable data, and the vendors selling those tools have a financial interest in not telling you how unreliable they've become.
What's broken and why
Last-click attribution: Attributes 100% of revenue to the final touchpoint before purchase. The problem: it systematically undervalues awareness and nurture touchpoints and overvalues paid search (which captures intent that was built by other channels). If you're using last-click, you're probably underfunding email and overfunding retargeting, because retargeting appears at the end of the funnel by design.
First-click attribution: The inverse problem. Overvalues acquisition channels; undervalues the channels that converted the customer. Most brands abandon first-click attribution quickly for this reason.
Linear attribution: Divides revenue credit equally across all touchpoints. Conceptually fairer than last-click, but practically useless for decisions, it doesn't differentiate between a touchpoint that built genuine purchase intent and one that happened to be in the path.
Data-driven attribution (DDA): Uses machine learning to assign fractional credit based on historical conversion patterns. More accurate than rule-based models, when it has sufficient data. The catch: it requires large volumes of conversion data to be statistically meaningful. Most SMB-scale direct marketing programmes don't generate enough data for DDA to be reliable.
The model that works for most direct marketing programmes
A pragmatic approach that most direct marketing teams can implement without advanced tooling: positional attribution with custom weights, combined with incrementality measurement for key channels.
Positional attribution assigns credit based on position in the customer journey: 40% to the first touchpoint (acquisition credit), 40% to the last touchpoint before conversion (decision credit), and 20% spread across intermediate touchpoints. This is a rule-based model, which means it has known biases, but it has the virtue of being transparent about those biases.
Incrementality measurement is the complement: periodically run holdout tests on your key direct marketing channels to measure whether they're actually driving incremental revenue. Send email campaigns to 90% of your list, hold out 10% as a control group, and compare purchase rates between the two groups. The difference is the incremental revenue attributable to the campaign, not modelled, not inferred, measured directly.
Email-specific attribution rules that reduce error
Email attribution introduces unique distortions that don't exist in other channels. Three rules that reduce the error:
Use 5-day click attribution windows, not 30-day. Email's influence on a purchase decision typically concentrates in the 24-72 hours after a click. A 30-day window picks up a lot of purchases that would have happened without the email. Test your own window by running holdout groups, but as a starting point, 5 days is more accurate for most programmes than 30.
Don't use open-based attribution. With MPP inflating open counts, revenue attributed to opens is unreliable. Click-based attribution only.
Deduplicate across channels in the same window. If a customer clicks an email link on Monday and then converts after clicking a Facebook retargeting ad on Wednesday, don't give both channels full credit. Define a clear deduplication rule, last click wins, or split credit, and apply it consistently. The worst attribution error is double-counting revenue across channels in a way that makes your total attributed revenue exceed actual revenue.
The honest position on attribution
Attribution is a framework for making decisions under uncertainty, not a measurement of truth. The goal isn't to find the model that perfectly captures how your customers buy, no such model exists. The goal is to find a model that produces decisions you'd be willing to defend with the underlying data.
That means: documenting the model you're using, documenting its known biases, and being consistent. A wrong model that's applied consistently over time produces trends that are meaningful, even if the absolute numbers are not. A team that changes attribution models every time a new vendor presents a more compelling dashboard produces trend data that's not comparable period-to-period, which is useless for programme management.
Pick a model. Document it. Apply it consistently. Supplement it with incrementality measurement where the stakes are high enough to justify the holdout cost. Review the model annually.
That's attribution in 2026, imperfect, pragmatic, and better than any alternative currently available at scale.