How to Prove a Channel Caused Revenue Without Running an Experiment
Geo holdouts are not the only way to prove incrementality. Here is how causal inference on the GA4 export you already own answers the same question in minutes, and when you should still run the experiment.
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 4 min read
You can prove a channel caused revenue without switching anything off. The method is causal inference on the observational data you already have: reconstruct the counterfactual, meaning what sales would have looked like without that channel, and compare it to what actually happened.
Why does everyone tell you to run a geo holdout?
Ask any measurement vendor whether a channel is incremental and you will get the same answer: run an experiment. The pitch is not wrong. Switch spend off in a matched set of regions, keep it on elsewhere, and the sales gap is about as close to proof as marketing gets.
Then there is the politics. Someone has to tell the growth team to stop spending in regions that are working, in a quarter where targets are already tight. Most incrementality testing programs die in that meeting, not in the statistics.
What is a counterfactual, really?
Every sale in your dashboard happened in a world where your ads ran. The causal question lives in the world you never see: what would this customer have done without the ad? That unobserved world is the counterfactual, and every incrementality method, experiment or not, is a different strategy for estimating it.
An attribution model does not estimate the counterfactual at all. It looks at the touches that happened and redistributes credit among them. If a retargeting ad reaches someone who was already walking back to your site, attribution hands the ad the credit. Causal inference asks whether the sale would have happened anyway, and counts only the difference.
That difference is the only number a budget decision should care about. A channel can show a beautiful ROAS and still add nothing, because it harvests demand that already existed. Another can look weak on last click and quietly create most of your new customers. You cannot tell the two apart without a counterfactual.
How does causal inference work on data you already have?
Here is the part vendor blogs skip. Your account history already contains experiments; you just did not run them on purpose. Budget changes, a spend pause when a card failed, a creative that burned out, a channel you scaled for a promo week: each one moved spend independently of customer demand.
Now the honesty part: this works under assumptions, and you should know them. Spend has to vary over time, otherwise there is nothing to learn from. Tracking has to stay consistent across the period. And no large unmeasured demand shock may land in exactly the weeks spend changed, because no model can distinguish a PR mention from a budget increase if they always arrive together. When those conditions hold, the estimate is causal, not correlational. When they do not, no observational method can save you, whatever the sales deck says.
When do you still need an experiment?
First, a brand-new channel with no history: there is no variation to learn from, so a launch test is the only way to get signal. Second, a violated assumption: if a video went viral the same week you doubled Meta spend, that week is contaminated, and you should say so.
There is also a calibration argument. Run one clean geo holdout on your biggest channel, once, and compare it to the causal read on the same period. Our branded search and retargeting incrementality test walks through that comparison for the two channels where attribution most often overclaims.
Which method fits your question?
| Geo holdout experiment | Marketing mix modeling | Observational causal read | |
|---|---|---|---|
| What you can trust | Strong causal proof, one channel in one period | Channel-level budget allocation, wide intervals at small scale | Causal estimates on your real history, with stated assumptions you can audit |
Read the table as a ladder, not a fight. Experiments give the strongest proof for one cell at a time. MMM allocates big budgets across many channels once you have the history.
GA4 export in, causal read out, in 1 to 2 minutes, with no pixel and no annual lock-in.
Key takeaways
- Every incrementality question is a counterfactual question: what would sales have been without the channel? Attribution models never touch that question.
- Observational reads work under stated assumptions: spend variation, consistent tracking, no coincident demand shocks. When those break, run the experiment.
Further reading
Frequently asked questions
What is observational causal inference?
Observational causal inference estimates what would have happened without a marketing action by studying variation that already exists in your historical data, such as budget changes, spend pauses and promotions, instead of running a controlled experiment. It reconstructs baseline demand and measures the gap when spend moves. Results are causal when key assumptions hold: spend varies over time, tracking stays consistent, and no large unmeasured shock coincides with the spend changes.How is a causal read different from an attribution model?
An attribution model redistributes credit among the touches it observed, so a retargeting click gets credit even when the customer was already returning. It never asks whether the customer would have bought regardless. A causal read estimates that counterfactual, the world without the channel, and reports only the revenue the channel actually added. One answers "who touched it", the other answers "what did it cause".Do I ever still need a geo holdout experiment?
Yes, in three cases: launching a channel with no history to learn from, periods where an unmeasured shock like a viral moment coincides with spend changes, and decisions big enough that stakeholders demand experimental proof.What data do I need for a causal read without an experiment?
You do not need a new pixel, a consent banner change, or a data science hire. Consistent tracking across the period matters more than raw volume.Can a causal read replace marketing mix modeling?
MMM becomes the right tool later, when you have the history, the spend, and many channels including offline to allocate across.
Go deeper: Causal attribution, 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.
- Attribution ModelAn Attribution Model defines how credit for conversions is assigned to marketing touchpoints. It dictates how marketing channels receive credit for sales.
- Causal InferenceCausal inference is the set of methods for estimating how much an outcome changes because of a cause, rather than alongside it. It works by comparing what happened with an estimate of what would have happened without the cause.
- CounterfactualA counterfactual is what would have happened under conditions that did not occur, such as a customer's outcome had they not received the campaign they did receive. It is never observed, so every effect estimate has to stand in for it with a control group, comparable units or a forecast.
- IncrementalityIncrementality is the part of an outcome, such as orders or revenue, that a marketing activity caused: the extra results compared with what would have happened without it. It is a causal effect, not the number of conversions credited to a channel.
- Incrementality TestingIncrementality Testing measures the additional impact of a marketing campaign. It compares exposed and control groups to determine causal effect.
- Marketing MixThe marketing mix is the set of actions a company uses to promote its brand or product. It traditionally includes product, price, place, and promotion.
- Marketing Mix ModelingMarketing Mix Modeling (MMM) is a statistical analysis that estimates the impact of marketing and advertising campaigns on sales. It quantifies each channel's contribution to sales.