Do you need attribution software? A sizing test

Not until the rules you can run for free disagree about a decision that is bigger than a test. A spreadsheet fed by GA4's attribution paths and your Shopify orders shows which channels appear on converting paths, and it is enough while every rule gives the same call. Software starts to earn its cost when the rules disagree and your volume can carry the method: Google's documentation recommends at least 200 conversions and 2,000 ad interactions in 30 days for data-driven attribution, and also says every conversion action is eligible regardless of volume.

What can a spreadsheet and GA4 already answer?

GA4's Key events attribution paths report lists the paths buyers took to a purchase, with purchase revenue, days to key event and touchpoints to key event for each path, and you can download it. Put your Shopify orders export next to it and you can answer four questions with no tool: which channels start, assist and close paths, how many touches buyers have, how long they take, and how much revenue sits in Direct or Unassigned.

What do the tools' own documents say they need?

Documented data needs, in each tool's own words. Some are recommendations, not limits:

• Data-driven attribution in Google Ads. Every conversion action is eligible regardless of volume, but Google recommends at least 200 conversions and 2,000 ad interactions in supported networks within a 30-day period, and says performance improves with more data.
• Marketing mix modelling in Google's Meridian. Its documentation says historical data should be a minimum of two years' worth of weekly data for geo-level models and three years' of data for national-level models.
• Incrementality experiments in Google Ads. Google says a single experiment that once might have cost upwards of $100,000 can now be done for $5,000 (Google Ads announcement of 11 November 2025, published by Google and not independently audited).

All three come from Google, which sells the ads these tools measure, so read them as guidance for Google's own tools. Below a documented number the tool still produces output. The Google Ads page says performance improves with more data, so treat thin-volume output as weaker evidence, not a finished answer.

How much of your data is noise?

Some limits come from the data, not the tool. A peer-reviewed paper, Lewis and Rao (Quarterly Journal of Economics, 2015), reported on 25 large field experiments with major U.S. retailers and brokerages. Individual-level sales were very volatile, with a coefficient of variation of 10 common, and informative experiments could easily require more than 10 million person-weeks. The lesson is general: a test has to beat the normal swing in sales before it says anything.

Measure your own swing. Total your last twelve weeks of revenue by week, then divide the standard deviation of the weekly totals by their average. For illustration: weekly revenue averages 50,000 euros and swings by 7,500 euros, which is 15%, so a channel that adds 5% of revenue, 2,500 euros a week, sits well inside the swing and one short test can't confirm it.

How do I run the sizing test?

Three gates, in order. Stop at the first one that says stay.

1. Do the rules disagree on a call? List the channels that get spend; each one is a call to make. Rebuild last click, first click and linear from your paths export (the models post has the steps) and write raise, hold or cut beside each channel under each rule. Stay with the spreadsheet if every channel gets the same call under all three. Go on if a call flips and the budget behind it is bigger than what a two-week holdout would cost you in revenue.
2. Can your data carry a test or a model? Put your 30-day orders, your conversions per conversion action, next to the documented numbers above, and the lift you expect next to your weekly swing. Stay with the spreadsheet if the lift is inside the swing: a short test can't resolve it, so run it longer or accept the uncertainty.
3. Is the tool cheaper than the hand work? Time your first rebuild, multiply it by how many times a year you would redo it and by your hourly cost, and compare that with a year of the price on the vendor's own pricing page. The platform comparison lists prices as read on stated dates. If the hand work costs less than the price, keep the spreadsheet.

Software is worth pricing only when the first gate says go, the second says you can read the answer and the third says it is cheaper than doing it by hand. Otherwise the next step is a holdout on your own numbers, not a subscription.

Sources, 30 September 2026: About data-driven attribution (Google Ads Help, 2026); Collect and organize your data (Meridian documentation, Google for Developers, 2026); Strengthen media measurement and ROI clarity with incrementality testing improvements (Google Ads Help, 11 November 2025); The Unfavorable Economics of Measuring the Returns to Advertising (Lewis and Rao, Quarterly Journal of Economics, 2015); Key events attribution paths report (Google Analytics Help, 2026).

Related answers

• Attribution platforms compared on published facts
• How to choose an attribution tool: a decision guide
• Multi-touch attribution models explained with one order
• Cut the channel, but hold it out first
• Incrementality testing