Marketing attribution models: choose the rule for your question

Choose the question before the model

A marketing attribution model is a rule for distributing conversion credit across eligible interactions. Choosing one changes the interpretation of a recorded journey; it does not change how much the customer paid.

Start with the decision you need to inform. Are you examining the first recorded acquisition touch, the interaction closest to payment, or the sequence of touches involved in a longer buying journey? Those questions justify different views.

Keep the conversion set fixed while comparing models. If one view includes renewals and another includes only first purchases, the difference is not merely attribution. Define revenue, customers, timing and eligible channels before discussing weights.

• Revenue attribution: define the amount first

Use single-touch views for a clearly bounded question

First-touch gives credit to the earliest eligible recorded interaction. It can help you investigate where observed journeys begin. Last-touch gives credit to the final eligible interaction and helps inspect what appears near conversion.

Both compress a journey into one credited position. Neither describes every influence on the buyer. A missing first visit can change first-touch attribution; excluding direct visits can change the last eligible touch.

For a hypothetical $180 purchase after social, email and search interactions, first-touch assigns $180 to social and last-touch assigns $180 to search. Total revenue remains $180 in each view. Do not add the two views together.

• Detailed first-touch versus last-touch example

Use fractional rules when the intermediate path matters

Linear attribution divides credit equally across the eligible units in the model. Those units might be interactions or another defined grouping, so inspect the implementation rather than relying on the label alone. With three eligible touches and $180, equal per-touch credit is $60 each.

Time-decay attribution gives greater weight to more recent eligible touches according to a specified decay rule. Its result depends on the chosen timing parameters and available timestamps. Position-based rules use explicit weights for selected journey positions.

These are interpretable conventions, not discoveries of causal contribution. Choose them because their assumptions match the question you are asking, then test how sensitive the result is to reasonable alternatives.

• Work through a fractional multi-touch example

Treat data-driven attribution as a specific implementation

Data-driven attribution learns an allocation from data rather than applying a fixed equal-credit or last-touch rule. The phrase does not identify one universal algorithm, training dataset or set of eligible interactions.

Google Ads currently documents last-click and data-driven choices; its first-click, linear, time-decay and position-based options have been retired. That product-specific availability does not mean the conceptual models ceased to exist elsewhere.

Do not infer that a product offers a learned model simply because it records multiple touches. Check its current model documentation, data requirements and supported reporting scope before recommending it.

• Google Ads: current attribution model availability

Compare models with one controlled worksheet

Select the same transaction sample, revenue basis and observation period. Record the included touches and then apply each model. Reconcile allocated plus unallocated revenue to the source total before interpreting channel shifts.

Next, change one assumption at a time. Shorten the lookback window while holding the model constant, or compare models while preserving the same eligible path. Changing both together obscures the reason for the result.

If a channel’s apparent value depends heavily on one rule, turn that sensitivity into a question for further investigation. It may reveal a long buying cycle, incomplete identity capture or a need for an incrementality experiment.

Separate report allocation from conversion delivery

A reporting model distributes credit for analysis. A conversion-delivery lookup finds the identifiers needed to send a supported event to an ad destination. They may use different settings and should not be treated as interchangeable.

DATALYR’s attribution documentation explains capture and identity resolution for conversion delivery. Use the reporting surface’s actual model settings when comparing channel credit; do not assume a delivery preference governs every report.

Record the chosen default and why it exists, keep comparison views available, and review it after a material change in the buying journey. The goal is a reproducible measurement decision, not finding the model that produces the highest ROAS.

• DATALYR capture and conversion lookup