One budget decision, one credit model, and every assumption it rests on written next to the number.

Last-click gives credit to the channel that closed the conversion, even when other channels helped build demand. We create a model that distributes credit across the journey and state clearly what it cannot prove. You end up with a credit model for a specific budget decision, with its assumptions and limits documented beside the results.

Plan attribution modeling All Marketing Measurement & Attribution tasks

Some of the 500+ brands we've worked with

See all references

We design the model around the decision it must inform rather than applying a generic multi-touch template. Four steps connect the decision brief to a reproducible model and readout.

How we hold ourselves to it

Define which touchpoints count — We define the included touchpoints and the conversion receiving credit so the result applies to the stated decision.
Select a credit rule for the funnel — We choose a data-driven or rule-based approach and explain why it fits the funnel instead of defaulting to the ad platform's built-in model.
Compare more than last-click — We run the model beside last-click and at least one other approach to show how strongly the recommendation depends on the chosen model.
State what the model cannot prove — We state plainly that attribution allocates credit under assumptions. It does not prove that a channel caused the conversion.

Define the decision
We agree on the budget or channel decision the model must inform and the evidence that could change it. The budget owner confirms this is the decision at stake.
Decision brief
Build a reproducible model
We define touchpoint scope, identity handling, and the credit rule, then implement the model so another analyst can rerun it. The analyst confirms touchpoint scope matches how the funnel works.
Attribution model
Compare models and test sensitivity
We run alternative models to see how much the conclusion changes when the modeling choices change. The paid-media lead reviews where the model likely overstates confidence.
Sensitivity comparison
Present the decision readout
We present the result with its assumptions and limits, then explain what the evidence does and does not justify. The budget owner signs off before reallocating any spend.
Attribution readout

Choose a credit rule that fits the funnel

Automation does the arithmetic: it weights touchpoint paths under the chosen rule and reruns the model against last-click and one alternative. Every judgment call stays with a person — which decision is at stake, whether the touchpoint scope matches the funnel, and where the model reads more confident than it is.

The deliverables show how the result was produced, how sensitive it is, and how it applies to the decision.

Working document
Attribution model
The reproducible model, its scope, and its credit rule, documented well enough to rerun.
Comparison report
Model comparison
How the credit split changes across last-click and at least one alternative model, so you can see the sensitivity.
Decision memo
Readout and limitations
The recommendation for your specific decision, with its assumptions and what it can't prove stated plainly.

We call it done when: a second analyst reruns the model from the documented inputs, reaches the same split, and the budget owner can name what the model does not prove.

This work fits a specific allocation decision that needs a more careful view of channel contribution.

A good fit when

You suspect last-click is over-crediting one channel, often brand search or direct, while undervaluing upper-funnel work.
A specific budget reallocation decision needs better evidence than last-click can provide on its own.
Your budget decision can proceed with a range, but the team is treating one modeled credit split as a settled fact.

An analyst weighing touchpoint tokens on a multi-arm balance

Better handled as other work when

You need causal evidence that a channel creates incremental conversions. Incrementality & Lift Measurement answers that question with a controlled test.
You want us to manage or reallocate live campaign budgets. That stays with the paid-media team. We hand them the evidence for the decision they still have to make.

If one of these is closer to your situation, start here instead: All Marketing Measurement & Attribution tasks

We call it done when: the budget decision, the touchpoint scope, and the credit rule are written down and agreed before any modeling starts.

Zeo designs measurement systems that connect a business decision to governed collection and reporting you can check. The people shown here work on the part of that system this page covers.

Google Analytics
supplies the last-click baseline every proposed model gets measured against for improvement
Jupyter
reruns the model with inputs perturbed, so the sensitivity test is a shown result, not a claim
Northbeam
runs the multi-touch model itself, weighed against last-click and against simpler rule-based alternatives

Tell us the budget question you need to answer. We will build a model for that decision and explain how much confidence the result supports.

Plan attribution modeling

Can the model replace last-click reporting?

For the specific decision in scope, usually yes. We would not recommend replacing the entire reporting stack until the model has been tested on a real decision.

Why not always use a data-driven model?

Data-driven models need enough conversion volume to remain stable, and tracking gaps can still introduce misleading assumptions. We check whether the available data supports this approach before recommending it.

How does attribution differ from incrementality testing?

Attribution divides credit for observed conversions using assumptions about touchpoint influence. Incrementality testing uses a controlled experiment to measure whether a channel caused additional conversions. The methods answer different questions.

What data is required to build the model?

We need access to touchpoint and conversion data from GA4 and ad platforms, plus a clearly defined budget decision for the model to inform.

By team

By industry

By tool

Marketing Attribution Modeling

Stress-test the model before using the result

Define the decision

Build a reproducible model

Compare models and test sensitivity

Present the decision readout

A model your team can rerun and challenge

Attribution model

Model comparison

Readout and limitations