How to check if multi-touch attribution still works for you

To check whether multi-touch attribution still works for you, measure how much revenue sits on paths with more than one touch. Then see how far data-driven moves each channel against last click. If little revenue is multi-touch, or the models barely disagree, the model is not your problem. Test channels with a holdout instead.

Step by step

You need a GA4 property that records purchase as a key event. The Google Ads, Meta and Shopify steps are optional: skip any platform you do not use. Run every step on the same full quarter.

1. Confirm which multi-touch model you still have. Read Reporting attribution model on the attribution settings page. Data-driven is GA4's only multi-touch choice now, because first click, linear, time decay and position-based went in November 2023. If it reads Paid and organic last click, your key event reports give each sale to one touch. Menu path: Admin > Data display > Events > Attribution settings.
2. Measure the revenue a split can reach. Open Attribution paths, pick purchase and note the purchase revenue in the table's top row. In Path length, choose greater than, enter 1 next to touchpoint(s) and click Apply, then note it again. The second figure divided by the first is your multi-touch share. Menu path: Advertising > Attribution > Attribution paths > Path length.
3. Measure how far the models disagree. Open Attribution models and pick purchase. Set the two Attribution model (non-direct) columns to Paid and organic last click and Data-driven, then read the % Change columns for each channel. Big moves mark the channels whose value depends on the model. Menu path: Advertising > Attribution > Attribution models.
4. Run the same comparison inside Google Ads. Go to Attribution in the Goals menu and open Model comparison, then use the Compare and With menus to pick Last click and Data-driven. This covers only your Google ads, so it shows which campaigns gain when credit spreads across them. The Overview page also counts conversions whose ad interactions spanned more than one device. Menu path: Google Ads > Goals > Attribution > Model comparison.
5. Measure channel overlap with Shopify's Any click. Filter Reports to the Marketing category, open Performance by referring channel and select Any click and Last non-direct click in the Attribution menu. Any click gives every clicked channel full credit, so its total runs past your sales. The further past it runs, the more your journeys mix channels. Menu path: Shopify admin > Analytics > Reports > Category: Marketing > Performance by referring channel > Attribution.
6. Ask Meta for its incremental view. In Ads Manager, open the Columns: Performance menu, choose Compare attribution models, select Incremental and click Apply. That column counts only the conversions Meta considers incremental, so read it next to your standard results. Menu path: Ads Manager > Columns: Performance > Compare attribution models > Incremental.
7. Send the channel that moved most in steps 3 to 6 to a holdout. Google's user-based Conversion Lift needs at least 1,000 observed conversions and a USD 5,000 budget. Meta's guide asks for a campaign with USD 5,000 of spend and 500 conversions. Below that, pause the channel in some regions and compare total sales. Menu path: Google Ads > Campaigns > Experiments > Lift studies, or Meta's Experiments tool.

A worked example

Start with one store's real shape. On its Journeys sheet, journeys with 1 touch hold 79.5% of revenue, and journeys with 2 or more touches hold 20.6% between them. Step 2 run on that store would leave about a fifth of revenue for any model to argue over.

The same store's Channels sheet adds the overlap check. Its touched view, which credits every channel a journey touched, adds up to 110.4% on the Channels sheet. It runs past the whole by design, the same way step 5's Any click total does.

Now your own quarter. For illustration, say step 2 shows €200,000 of purchase revenue, with €60,000 left after the path filter. In this worked example, your multi-touch share is 30%, so the model debate covers under a third of your sales.

For illustration, say step 3 shows Paid Search at €40,000 under last click and €34,000 under data-driven, a change of minus 15%. For illustration, Paid Social climbs from €6,000 to €11,000, and Email barely moves. In this worked example, Paid Social is the channel the models argue about: its credit nearly doubles.

Step 5 shows the overlap in money. For illustration, say Any click totals €95,000 against €80,000 of Shopify sales. In this worked example, the extra €15,000 is credit that a second or third channel claimed for the same orders.

Step 6 then gives a second opinion. For illustration, say Meta shows 150 purchases under standard attribution and 60 under Incremental. In this worked example, Meta's own model counts fewer than half of its credited purchases as caused by its ads.

So Paid Social gets the holdout. In this worked example, it falls short of Meta's guide on purchases, so you pause it in a third of your regions for four weeks. Then you compare total Shopify sales there with the regions where it kept running.

If those regions hold their sales, Paid Social was collecting credit more than creating sales. If they drop, data-driven was right to lift it. Either way, you stop arguing about models and start reading a result.

What should you check when the numbers look wrong?

• Direct does not move between models. That is by design. GA4 credits Direct only on paths that were direct all the way, and those have nothing to share.
• Google Ads' comparison leaves campaigns out. Google's help says the Model comparison report filters out any network or campaign without enough data. Pick a more recent date range for complete data.
• Meta's incremental column is blank. It has no results for date ranges before 1 April 2025, so check your dates first.
• Data-driven matches last click to the euro. Google's help says the two can produce the same results, depending on how much data you have. Small accounts give the model little to learn from.
• Any click barely exceeds your sales. Then few journeys in Shopify's view mixed channels, and multi-touch has little to split there.
• Shopify and GA4 give different multi-touch shares. They keep separate cookies and separate memories, and Shopify resets a journey after every order. Compare the direction of each tool's answer, not its decimals.

What to do this week

1. Write down your multi-touch share. Run step 2 for the last full quarter. Pass: a real share of revenue sits on multi-touch paths, so keep data-driven and read it monthly. Fail: only a sliver does, so model choice barely matters; spend the time on tags and user IDs instead.
2. Name the channel the models fight over. From steps 3 to 6, pick the one whose credit moves most. Pass: one channel stands out, so you know what to test. Fail: nothing moves, so the models agree and your remaining question is what each channel caused.
3. Put the holdout in the calendar. Pass: the channel clears a platform's lift minimums, so book the study this week. Fail: it does not, so write a regional plan: which regions pause, for how many weeks, and which Shopify report you will compare.

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: Select attribution settings (Google); Key event attribution paths report (Google); Get started with attribution (Google); Key event attribution models report (Google); About attribution models (Google); About attribution reports (Google); Marketing reports (Shopify); About data-driven attribution (Google); How to view results for incremental attribution in Meta Ads Manager (Meta); Set up Conversion Lift based on users (Google); About Conversion Lift (Meta).

Related answers

• All Attribution models questions
• Is multi-touch attribution dead?
• Is multi-touch attribution dead for health and beauty?
• Incrementality testing, explained