Which channel should I test first?

The one where a wrong call costs most. Look for big spend, credit that shrinks under a stricter model and a seat late in the path, like retargeting or brand search. Test one channel at a time, so the result points at one cause.

Can I cut part of a channel instead of all of it?

Yes, and often you should. Trim the campaigns that score worst under the stricter view, keep the rest, and compare total sales with regions where nothing changed. A go-dark measures the whole channel, so a trim needs its own test.

Does Meta's incremental attribution tell me which campaign to cut?

It helps inside Meta. The Incremental column counts only conversions Meta considers incremental, so you can rank Meta campaigns against each other. It can't compare Meta with Google, and Meta advises comparing within one attribution model. Confirm a big cut with a lift test.

How to pick the channel to cut, step by step

Set a break-even bar from your Shopify margin, line up each platform's conversion window and compare credit under two models. Find the channels that open journeys, rank the suspects, then switch the top one off in some regions and cut only if total sales hold.

By Joris van Huët, Founder & CEOUpdated 2026-10-018 min read

Run the numbers for your store: the free attribution window calculator.

Usually in seven steps: set a break-even bar from your margin, line up each platform's conversion window, and compare credit under two attribution models. Then find the channels that open journeys and rank the suspects. Switch the top one off in some regions, and cut it only if total sales hold.

Credit tells you where to look. Only a test tells you what to cut. So this is a shortlist, then a test. The first six steps take an afternoon and decide what to test. The last one takes a few weeks and decides what to cut.

Step by step

Set the bar a channel must clear. Break-even ROAS is 1 divided by your gross margin. Hold each channel to it on the sales it causes, not the sales it claims. Shopify's profit reports only count products with a cost recorded when they sold, so fill in costs first. Path: Shopify admin > Analytics > Reports > Category filter > Profit margin > Gross profit by product.
Collect spend and claimed revenue for one period. Use the same full quarter for every channel. Take spend from each ad platform, because Shopify's Growth page doesn't show cost or ROAS for Facebook and Google campaigns. Path: Google Ads > Campaigns menu > Campaigns, and Meta Ads Manager > Columns: Performance.
Write each platform's window next to its claim. Google Ads defaults to 30 days after a click for a new conversion action. Meta's standard setting counts 1 or 7 days after a link click and 1 day after a view. GA4 looks back 90 days by default for purchases. Path: Google Ads > Goals icon > Conversions > Summary > your action > Edit settings > Click-through conversion window.
Compare each channel's credit under two models. Read the channel report under last non-direct click, Shopify's default, then under first click. A channel that shines only at the end is a closer; one that gains under first click opens journeys. Path: Shopify admin > Growth > View channel report > Attribution model.
See where each channel sits in the journey. In GA4's paths report, switch the chart to data-driven and read the Early, Mid and Late bars. Early touchpoints are the first 25% of each path. Path: GA4 > Advertising > Key events dropdown > Key event attribution paths > attribution model drop-down.
Rank the suspects. A suspect has big spend, a claim that shrinks under the stricter model, and a seat late in the path. Download GA4's model comparison and set its % Change column next to your spend. Path: GA4 > Advertising > Attribution > Attribution models > Share this report.
Test the top suspect, then decide. Switch it off in some regions, keep it on in the rest, and compare total sales by region. Lift reports give a range, not just one number: Google's geo report shows one, and Meta's lift results carry lower and upper bounds. If even the top of that range sits below your bar, cut. If the range straddles the bar, trim and test again. Path: Google Ads > Campaigns menu > Campaigns > Settings icon > Locations > Enter another location > Location options > Presence.

On Meta, the same test runs as a Conversion Lift study in Experiments, and its result is the number to judge the cut on.

A worked example

For illustration, with round invented numbers for one quarter. Say your gross margin is 40%, so the bar is 2.5x: 1 divided by 0.40, the sum on one store's Break-even sheet. Shopify's own profit example lands on the same margin, with $50 of net sales and $30 of cost.

For illustration	Spend	Revenue the platform claims	Claimed ROAS	Seat in GA4 paths
Google brand search	€6,000	€48,000	8.0x	Late
Meta retargeting	€9,000	€45,000	5.0x	Late
YouTube	€9,000	€9,000	1.0x	Early

A ROAS league table cuts YouTube. The steps point elsewhere. Meta retargeting holds the most spend among the closers, so it becomes the top suspect. Say its Shopify sales fall by half when you switch the channel report to first click.

Say the go-dark test finds retargeting caused €16,200 of sales on its €9,000: an iROAS of 1.8x. If the test's range runs from 1.2x to 2.4x, even its top sits under the 2.5x bar. Cut it.

For illustration, the cut gives up €16,200 of sales, which carried €6,480 of margin. For illustration, it also saves €9,000 of spend, so the quarter ends €2,520 better off. Sales fall and profit rises, both at once.

For illustration, brand search still looks best on paper at 8.0x. It sits late in the path, so it is next in line for a test, not a free pass.

Next quarter, say you test YouTube, the channel the league table wanted gone. Say the test finds €27,000 of caused sales on €9,000: 3.0x, above the bar. The lowest-ROAS rule would have cut a channel that pays its way.

What to check when the report looks wrong

Meta's lift result and Ads Manager disagree. Expected. Meta says lift results are not meant to be compared with Ads Manager, which credits conversions inside attribution windows. Decide the cut on the lift number.
Google says No significant lift detected. The groups didn't differ enough to tell apart, which is not the same as a channel that did nothing. Google suggests checking for lift in specific conversion slices. A flat result whose range sits below your bar still supports a trim.
Sales dipped everywhere after the cut. Look at the regions that kept the channel. If they dipped too, the season, a promotion or the weather did it, not the cut.
Brand search and Direct sag a few weeks later. That can be an opener's work arriving late. Compare those rows across test and control regions before you call the cut a win.
The test regions bought more, not less. A go-dark that lifts sales usually points to noise or a regional event, not a channel that hurts. Rerun it with more regions before you read anything into it.
A paid channel shows no revenue in GA4. GA4's Paid Social rule needs a source on its social list and a medium that looks paid. Anything else lands in another row, or in Unassigned when no rule matches.

What to do this week

Run the best-case check. In Google Ads, open Campaigns in the Campaigns menu and divide each campaign's conversion value by its cost. Do the same in Meta Ads Manager. Pass: every channel clears your break-even bar on its own claim. Fail: one misses even on its own count, so trim it now, unless its window is far shorter than the others.
Check Meta's entry ticket. Meta's guide asks for a campaign from the past year with $5,000 USD or more of spend and 500 conversions. Pass: your suspect qualifies, so set up a Conversion Lift test in Experiments. Fail: it doesn't, so test a Google channel first, or plan a regional holdout by hand.
Set the suspect's location option now. In Google Ads, select the Settings icon next to the campaign and expand Location options. Pass: it is set to Presence, which Google's geo guide uses to prevent location leakage. Fail: it is set to anything else, so switch to Presence before the test starts, not halfway through.

Check the homework. Your GA4 Attribution paths export already holds the evidence. Causality Engine reads that one file and shows what each channel caused next to what last-click gave it, in 1 to 2 minutes, for €99 once (excluding VAT), refundable within 30 days. Check the homework

Sources, 1 October 2026: Profit reports (Shopify Help Center); Measuring marketing performance (Shopify Help Center); About conversion windows (Google Ads Help); About attribution models and attribution settings (Meta Business Help Center); Select attribution settings (Google Analytics Help); Key events attribution paths report (Google Analytics Help); Key event attribution models report (Google Analytics Help); Implement campaigns for geo experiments (Google Ads Help); Understand your Conversion Lift based on geography measurement data (Google Ads Help); Differences between Conversion Lift test results and other reporting tools (Meta Business Help Center); About Conversion Lift (Meta Business Help Center); How to view results for incremental attribution in Meta Ads Manager (Meta Business Help Center); Facebook Lift metrics glossary (Meta Business Help Center); Default channel group (Google Analytics Help).

Frequently asked questions

Which channel should I test first?
The one where a wrong call costs most. Look for big spend, credit that shrinks under a stricter model and a seat late in the path, like retargeting or brand search. Test one channel at a time, so the result points at one cause.
Can I cut part of a channel instead of all of it?
Yes, and often you should. Trim the campaigns that score worst under the stricter view, keep the rest, and compare total sales with regions where nothing changed. A go-dark measures the whole channel, so a trim needs its own test.
Does Meta's incremental attribution tell me which campaign to cut?
It helps inside Meta. The Incremental column counts only conversions Meta considers incremental, so you can rank Meta campaigns against each other. It can't compare Meta with Google, and Meta advises comparing within one attribution model. Confirm a big cut with a lift test.

Go deeper: Causal attribution, explained.

Sixty-second versions of these ideas: Causality Engine on YouTube Shorts.

Keep reading

Terms in this article

Browse the full glossary