Web Analytics · Marketing Measurement

Marketing Attribution Modeling

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.

A credit model for a specific budget decision, with its assumptions and limits documented beside the results.

Plan attribution modelingAll Marketing Measurement & Attribution tasks
A Zeo analyst weighing touchpoint tokens on a multi-arm balance

Some of the 500+ brands we've worked with

See all references
  • Bayer
  • Arabam.com
  • Little Caesars
  • İstikbal
  • QNB Finansfaktoring
  • Jack Martin Menswear
  • Ajansspor

How we run it

Stress-test the model before using the result

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.

  1. Define the decision

    We agree on the budget or channel decision the model must inform and the evidence that could change it.

    Decision brief

    AI assist
    Drafts candidate decision statements from the stakeholder brief.
    Human gate
    Budget owner confirms this is the decision at stake.
    Owners
    Measurement Analyst, Budget Owner
    Illustrated figure holding up a signed agreement page

  2. Build a reproducible model

    We define touchpoint scope, identity handling, and the credit rule, then implement the model so another analyst can rerun it.

    Attribution model

    AI assist
    Weights each touchpoint path under the chosen credit rule.
    Human gate
    Analyst confirms touchpoint scope matches how the funnel works.
    Owners
    Measurement Analyst
    Illustrated figure sketching plans at a drafting table

  3. Compare models and test sensitivity

    We run alternative models to see how much the conclusion changes when the modeling choices change.

    Sensitivity comparison

    AI assist
    Reruns the model under last-click and one alternative rule.
    Human gate
    Paid-media lead reviews where the model likely overstates confidence.
    Owners
    Measurement Analyst, Paid-Media Lead
    Illustrated figure reading an oversized measurement dial

  4. Present the decision readout

    We present the result with its assumptions and limits, then explain what the evidence does and does not justify.

    Attribution readout

    AI assist
    Drafts the readout narrative from the model's stated limits.
    Human gate
    Budget owner signs off before reallocating any spend.
    Owners
    Measurement Analyst, Budget Owner
    Illustrated figure presenting a bar chart on an easel

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.

What you get

A model your team can rerun and challenge

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.

    Accepted when

    Rerunning it from the documented inputs reproduces the same credit split.

    Cadence: Rerun each reporting cycle

  • Comparison report

    Model comparison

    How the credit split changes across last-click and at least one alternative model, so you can see the sensitivity.

    Accepted when

    Every model in the comparison runs on the same window and the same touchpoint scope.

    Cadence: Once per modeling round

  • Decision memo

    Readout and limitations

    The recommendation for your specific decision, with its assumptions and what it can't prove stated plainly.

    Accepted when

    The memo names at least one claim the model cannot support.

    Cadence: At the readout

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.

Fit and readiness

When last-click is not enough for the budget decision

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.

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.

People who build your measurement system

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.

Tools we use

Tools behind this work

  • 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

Next step

Build the credit model around a real budget decision

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

Before we start

Questions teams ask before booking

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.