What an agent can actually do with attribution data
Four tasks a model genuinely handles well once the data carries its uncertainty, and two it should not be given no matter how good the data gets.
Three platforms, three answers, one budget to defend. Your GA4 export already holds the answer, and the first finding is free and never leaves your browser. Get your first finding free
By Joris van Huët, Founder & CEOUpdated 2 min read
Four tasks a model handles well once the data carries its uncertainty, and two it should not be given regardless.
The four it does well
| Task | Why it suits a model |
|---|---|
| Summarising a read into prose | Mechanical, and the fields constrain it |
| Comparing this period to the last | Arithmetic plus a written comparison |
| Drafting the decision memo | Structure is repetitive; the judgement is yours to edit |
| Answering "what does this field mean" | Explanatory, and checkable against published method |
The third is the sleeper. Most of the work in a monthly report is assembling a document that says the same things in the same order with different numbers. A model does that well, and the human contribution becomes the paragraph about what to do, which is the part that needed a person.
The two it should not be given
Deciding to move budget. An observational estimate needs someone weighing what the model cannot see: creative fatigue, auction dynamics, a supply constraint, a competitor's promotion. None of that appears in a channel-level dataset.
Choosing the window. If a model can select the period, it will select the one that best supports whatever it is being asked to demonstrate, without any intent to mislead. Fix the window in the request. The same discipline applies to human analysts, as covered in a checklist for defending an attribution finding.
The honest boundary
None of this makes an agent an analyst. The longer version is in can an AI agent do your attribution.
Related answers
Frequently asked questions
What can an AI agent do well with attribution data?
Summarise a read into prose, compare this period against the last, draft the recurring decision memo, and explain what a field means. All four are assembly tasks that the field structure constrains tightly.What should an agent not be allowed to do?
Decide to move budget, because an observational estimate needs a human weighing context the model cannot see; and choose the analysis window, because it will pick the one that best supports whatever it was asked to show.What two rules keep a model honest about attribution?
Treat an interval spanning break-even as unresolved, and never recommend action on a channel labelled unmeasurable. Both belong in the system prompt rather than the data, and together they prevent most overstatement.
Go deeper: Incrementality testing, explained.
Sixty-second versions of these ideas: Causality Engine on YouTube Shorts.
Have an audience of ecommerce operators? Earn 20% recurring commission by recommending the €99 read.
Keep reading
Terms in this article
- AttributionAttribution identifies user actions that contribute to a desired outcome and assigns value to each. It reveals which marketing touchpoints drive conversions.
- Lift TestA lift test is an experiment that measures lift, the extra conversions or sales a campaign causes, by comparing a group that could see it with a control group that could not. Lift is reported as a difference or as a ratio to the control group's result.
- ShopifyShopify is an ecommerce platform for creating online stores and selling products. Attribution modeling shows which marketing channels drive traffic and conversions within Shopify.