MMM or multi-touch attribution: how to choose, step by step

To choose between MMM and multi-touch attribution, first sort your decision: a channel budget split points to MMM, campaign tweaks point to attribution. Then check the data. In GA4, see how much revenue sits in multi-touch paths. In Shopify and Google Ads, count your weeks of sales and spend. Usually the data decides.

Nine checks follow, in the order you can run them. Most take minutes in tools you already pay for. The plain answer covers why each one matters.

Step by step

1. Sort the decision by level. Splitting next quarter's budget between channels is a mix-model question. Pausing a campaign, ad set or creative is a job for attribution or a platform test. Meridian's guide calls MMM a macro tool that works well at channel level. Where: one line at the top of your plan.
2. List the channels a path can see. Paid search, social ads and email send visits that GA4 records. TV, podcasts, print and unclicked video views send none, so attribution cannot credit them. Where: Reports > Acquisition > Traffic acquisition, next to your list of paid channels.
3. Measure revenue in multi-touch paths. Click Advertising, then Attribution paths under Attribution. Set Path length to greater than or equal to 2 touchpoints, click Apply and compare Purchase revenue with the unfiltered total. Path: Advertising > Attribution > Attribution paths.
4. Compare data-driven with last click in GA4. Click Advertising, then Attribution models under Attribution. Use the drop-downs in the Attribution model (non-direct) columns to put Data-driven next to Paid and organic last click, then read % Change. Path: Advertising > Attribution > Attribution models.
5. Run the same comparison in Google Ads. Go to Attribution in the Goals menu, select Model comparison, then use the Compare and With drop-downs. Google says the two models can give the same results, depending on data. It recommends at least 200 conversions and 2,000 ad interactions in 30 days for data-driven attribution.
6. Count your weeks of sales. In Shopify admin, go to Analytics > Reports and open Total sales over time. In the configuration panel's Dimensions menu, set the time unit to week. Robyn's guide asks for two years of weekly data at minimum.
7. Check that spend moved. In Google Ads, open the Campaigns table, select the segment icon, then Time and Week. Meridian's guide warns that too little variation in media spend hurts national models. Path: Campaigns > segment icon > Time > Week.
8. Count data points per parameter. Divide your weekly rows by the channels, controls and other inputs you would model. Meridian's guide treats this ratio as a key check for national models. Its example of 156 points for 10 parameters, roughly 15 each, might give directional results. Where: a spreadsheet, before anyone sends you a quote.
9. Pick, then plan the test that checks it. If steps 3 to 5 barely move and steps 6 to 8 fall short, test first. In Google Ads, Lift measurement in the Goals menu offers Conversion Lift based on geography. Robyn's guide wants experiments like that as the calibration for any mix model.

A worked example

Take Store A, one store's anonymised GA4 export from 1 January 2024 to 21 August 2026. It holds shares of revenue by path and by channel, and no spend.

Step 3 on that file: journeys with 2 or more touches hold 12.2% + 5.4% + 3.0% = 20.6% of revenue on the Journeys sheet. Journeys with 1 touch hold 79.5% and took 0.5 days to buy, on the same Journeys sheet. A multi-touch model could re-split about a fifth of revenue, and no more.

The export even gives step 4 a rough answer. On the Channels sheet, Direct holds 57.7% of revenue in last click, first click and touched alike. So no change of model would reach that share.

Steps 6 to 8 need files the export does not hold. The export spans about 137 weeks, which clears Robyn's two-year bar if the sales and spend records go back as far.

For illustration, say the store would model four paid channels, two controls and two knots, which makes eight parameters. If the store had 120 weekly rows, that would be 15 per parameter, the level Meridian's example calls directional.

So the choice for a store with this export is fairly clear. A better attribution model moves about a fifth of revenue at most, and leaves Direct where it is. A mix model could work on the weeks, but needs spend and sales pulled from Shopify and the ad accounts first. Step 9 comes first: a test on the biggest paid channel.

Which result points where?

Line up what steps 3 to 8 told you, then read across. Most stores land in one row.

The last row is not a consolation prize. A test answers the cause question that both methods only estimate, and it gives a later mix model the calibration Robyn's guide asks for.

What to check when the comparison looks wrong

• Data-driven and last click match exactly. Google's help says that can happen, depending on data availability. Check your volume against step 5 before reading anything into it.
• Last week's numbers keep shifting. GA4's data-driven model can reattribute conversions for up to 7 days after they happen. Leave the most recent week out of any comparison.
• Direct hardly moves between models. That is by design. GA4's models give Direct credit only when a whole path is Direct.
• Google Ads and GA4 disagree. Google Ads' data-driven model looks at interactions with your Search, Shopping, YouTube, Display and Demand Gen ads. GA4's paths also hold your other channels, so the same campaign can earn different credit.
• A new lookback window changed nothing in old reports. GA4 applies lookback changes going forward only. A new reporting attribution model, by contrast, applies to historical data as well.
• The paths table is empty. GA4 shows no path without a key event, so check that purchase is marked as one.
• A mix-model quote models more inputs than your rows can carry. At roughly 15 rows per input, Meridian's example only reaches directional reads. Ask which channels the plan would combine or drop, as Meridian's guide suggests.

What to do this week

1. Download your paths as a baseline. In Attribution paths, set the date range to the last year, click Share this report and download the table. Pass: a file with paths, key events and purchase revenue. Fail: an empty table, which means purchase is not marked as a key event.
2. Run the Google Ads comparison by campaign. In Goals > Attribution > Model comparison, choose campaign as the Dimension and compare Last click with Data-driven. Pass: some campaigns gain under data-driven, so path credit adds something for search. Fail: the columns match, so treat model choice as noise for now.
3. Write your rows-per-input sum. Divide your weeks of sales by the inputs a mix model would need. Pass: around 15 or more per input, where Meridian's example says directional reads become possible. Fail: fewer, so combine channels, wait for more weeks, or test first.

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: Get started with attribution (Google); Key events attribution paths report (Google); Key event attribution models report (Google); Select attribution settings (Google); Traffic acquisition report (Google); About attribution models (Google); About data-driven attribution (Google); Use segments in your tables (Google); Set up Conversion Lift based on geography (Google); Time ranges for reports (Shopify); Amount of data needed (Google); An Analyst's Guide to MMM (Meta)

Related answers

• All MMM & measurement questions
• MMM vs multi-touch attribution, which is better?
• MMM, MTA or incrementality: a guide by data volume
• How do I run a geo holdout test?
• Incrementality testing, explained