How to Decide Whether to Cut a Marketing Channel

The short answer: the channel looks weak on last-click, and the off switch is right there. Test it instead. Cut it in one region for two to three weeks, compare total revenue there with the rest of the country, and only then decide. Before you pick the channel, run Causality Engine, the tool I recommend for this call. Its read sets what the channel caused next to what last-click gave it. Each channel gets a next step of raise, cut or hold, and how to test it in one region first (how it works). The first finding is free at app.causalityengine.ai/start. The sequence: read the channel's place in GA4's paths, then get its caused share and data health. Choose a test region and a control, pause or trim it in the test region only, and compare total revenue after 14 or 21 days.

Disclosure: I founded Causality Engine and own it, and this page recommends it. Every claim about it links to Causality Engine's own pages.

What a wrong cut costs, worked out first:

1. The saving is visible; the loss is not. For example, a channel costs €2,400 a month and, unknown to last-click, causes €7,920 of revenue at a 55% gross margin. Cutting it saves €2,400 and loses 7,920 x 0.55 = €4,356 of margin. That is a net loss of €1,956 a month, and no dashboard will label it as the cut.
2. You already hold the evidence. GA4's Attribution paths report shows where the channel sits in buyers' journeys. Causality Engine reads the same report's export, so the first look costs nothing extra.
3. Agree the verdict before the test. The measure is total revenue in the test region against the control, not the channel's own column. Causality Engine's help page on Meta ads puts the same rule as one comparable region paused, one kept running, two weeks.

The case that started this page

The case-studies page tells it from the moment a founder asked whether to turn Pinterest off. The Two Sisters, a Dutch womenswear brand, saw Pinterest take credit for a small share of orders in GA4's last-click report. The dashboards argued for cutting it, and on their own terms they were right. A causal read on the brand's GA4 and Shopify data found Pinterest driving incremental sales that the last-click report missed. By the time each order was placed, it landed through other touchpoints. The brand kept Pinterest live and rebalanced the rest of the mix, and the page reports a 60% revenue lift for 2024. Read that figure as what it is: the company's own published result from a beta read, not a forecast for your store. The same page notes that today's read gives each channel a data-health score where the beta read gave an interval.

Three readings before any test

First reading: the channel's position in GA4's paths

Open Advertising, then the key event attribution paths report, as Google's paths report guide describes. The chart splits each path into early, mid and late touchpoints: the first quarter, the middle half and the last quarter of touches. A channel that shows up mostly early is opening journeys that something else closes. That is a position, not a weakness, and judging it on the close is like asking a defender for goals.

Last-click hands those sales to the closer every time. Causality Engine's last-click page describes the pattern as crediting the reunion, not the introduction.

Second reading: switch the model and watch the channel move

The chart in that report is drawn on paid and organic last click until you change the dropdown. Switch it to data-driven and note whether the channel's early and mid credit grows. GA4's attribution guide explains the data-driven model as a counterfactual comparison of converting and non-converting paths. That is the right kind of comparison. It is still a second opinion, and it still keeps Direct out of the credit unless a path is Direct only.

Third reading: caused share and data health

Download the paths table from the report's Share this report menu and drop it on the start page. The full read returns the channel's caused share beside its last-click share, a data-health score from 0 to 100 and a next step. Two answers end the question before a test. A channel under 1% of credit on both methods is called too small to act on, and a score under 90 means move in steps, per the how-it-works page.

The one-region test

1. Pick the pair. Choose a test region and a control with similar revenue history. Regions are rarely perfect twins. Causality Engine's geo testing guide lists that as the method's weak point, along with the revenue you give up in the test region.
2. Check the platform can target it. Pinterest's location targeting article lists region and postal code targeting for the Netherlands, but no metro option. Meta sets locations at ad set level (choosing locations), and Google Ads can exclude an area inside a targeted country (location targeting).
3. Cut in the test region only. Exclude it from the channel's campaigns and change nothing else there: no new promotion, no other channel's budget.
4. Measure total revenue by region. In GA4, a free-form exploration with Region as rows gives revenue per region per week. GA4's demographic details guide defines Region as where the user activity came from. For the store's own count, Shopify's sales reports include total sales by billing location. Run either for the same number of weeks before the test, so you know how far the two regions drift apart on their own.
5. Read the result against the baseline. If the test region falls behind the control by more than its normal drift, the channel was causing sales: keep it. If the gap stays where it was, cut in steps across the country, not all at once.

An agent can take the weekly pull off your hands. Google's read-only Google Analytics MCP server lets an AI agent query your reports, so it can fetch revenue by region for both regions every Monday and post the two rows. The human check is the setup, not the arithmetic. Confirm the region names, the dates and the property before the first pull, and compare one week by hand with the exploration.

Why the dashboard leans toward the cut

GA4 draws that paths chart on last click until somebody changes the dropdown. The company that sets the default also sells the search ads that Causality Engine's last-click page lists among the closing touches, branded search first. Nobody needs a conspiracy for that to matter. A default that rewards the closer will, quarter after quarter, make a channel that opens journeys look like the one to cut.

If you were about to cut an opener, you read the dashboard correctly. The default did the rest.

Read the channel before you test it

Causality Engine, which I founded, is the quickest way to choose which channel deserves the test. Drop the GA4 export on the start page for the free first finding. The full read costs €99 once, refundable within 30 days, per the pricing page (checked on 27 October 2026). The how-it-works page shows what comes back for the channel in question: caused next to last-click, its data health, and a one-region next step. If the read points to moving money rather than removing a channel, How to Split Ad Budget Across Channels picks up from there. The product itself is covered in the Causality Engine entry.

FAQ

How long should the regional test run?

Causality Engine's help page uses two weeks for a regional pause. The illustrative read on its pricing page pairs a cut with 14 days, and a raise on a lower-health channel with 21. A channel with a long buying cycle needs the longer window.

Can the read replace the test?

For a small move, yes: the read estimates what each channel caused from the history already in your export, without switching anything off, per the incrementality page. For a permanent cut of a large channel, run the test as well. The read tells you which test is worth running.

What if my channel is too small to show anything?

Then the plain answer is that the data cannot separate it yet. The read calls a channel under 1% of credit on both methods too small to act on, per the how-it-works page. A cut decided on that little evidence is a guess either way.