Single-touch vs multi-touch attribution on the same orders

For illustration: the same 100 orders give Paid Social 0 orders under last click, 55 under first click and 25 under linear multi-touch credit, and give Paid Search 51, 12 and 29. No order changed; the rule did. A single-touch rule answers one question, who closed the sale or who started the journey, and a multi-touch rule spreads credit over the touches it can see. None of them shows what would have happened without the channel, which takes an experiment.

What do three rules say about the same 100 orders?

For illustration: take 100 orders and the paths buyers took to them. The paths below are invented.

For illustration (invented paths): 100 orders

Orders  Path
    18  Organic Search
    12  Paid Search
    24  Paid Social > Paid Search
    16  Paid Social > Email
    15  Paid Social > Organic Search > Paid Search
    15  Email

Credit in orders    Last click  First click  Linear
Paid Social                  0           55      25
Paid Search                 51           12      29
Organic Search              18           18      23
Email                       31           15      23
Total                      100          100     100

For illustration: 45 of the 100 orders have a single touch, so every rule credits them the same, and the other 55 are where the rules part ways.

Read the table by channel. Paid Social never closes an order in this data, so last click gives it nothing, while it starts every multi-touch order in this data, so first click gives it a great deal. Paid Search runs the other way. Linear credit sits between them, and it hands Paid Social a share of every order it touched whether or not its ads changed anyone's mind.

Which decisions can each rule support?

What each rule can and can't carry:

• Last click shows which touch closes sales. It can guide what you show at the last step. It can't justify cutting a channel that mostly starts paths, like Paid Social above.
• First click shows which channel starts journeys. It can guide where new buyers are first reached. It can't credit the channel that closed.
• Linear multi-touch shows which channels appear together on a path. It is a neutral split when you have no reason to weight touches. It gives every touch a share whether or not the touch mattered.

None of the three decides a budget alone. Google's GA4 help page defines attribution as assigning credit to ads, clicks and other factors along the path to an action, so every rule above divides credit among touches that were observed. A budget call asks something else: what would have happened without the spend.

What does the published evidence say about that gap?

The published test compares estimates made from observed data with randomized experiments. Gordon et al. (Marketing Science, 2019, peer-reviewed) used data from 15 U.S. advertising experiments at Facebook comprising 500 million user-experiment observations and 1.6 billion ad impressions. The authors report that the observational methods often fail to produce the same effects as the randomized experiments, even after conditioning on extensive demographic and behavioral variables. That paper tested estimation methods, not the three rules above. The rules make no attempt to estimate an effect, which is the point.

Meta, which also sells the ads, took the same position on 3 March 2026, in the announcement that narrowed its click-through definition to link clicks. It said advertisers should ask what an ad or campaign caused that would not have happened otherwise, and that incrementality experiments are the best way to answer it and "the gold standard in measurement". That is a platform's own view, published by Meta and not independently audited.

Experiments are noisy too. Lewis and Rao (Quarterly Journal of Economics, 2015, peer-reviewed) reported on 25 large field experiments with major U.S. retailers and brokerages. The median confidence interval on return on investment was over 100 percentage points wide. Read any test result with that margin in mind.

How do I check which rule is deciding on my own data?

1. Build last-click, first-click and linear columns from your GA4 paths export. The models post has the steps.
2. Next to each channel, write the call you would make under each column: raise, hold or cut.
3. Pass: the call is the same under all three columns, so the rule doesn't matter for that channel and you can stop. Fail: the call flips. Carry that channel to the next step.
4. For each channel that fails, change its spend a little in one region for two weeks and compare revenue with a region you left alone. Read the gap against your normal week-to-week swing in revenue: a difference inside the usual swing tells you nothing either way.

Later, once a test has told you which channels to question, a causal attribution read like Causality Engine's can set what each channel caused next to last-click, from the same GA4 export.

Sources, 30 September 2026: Get started with attribution (Google Analytics Help, 2026); A Comparison of Approaches to Advertising Measurement: Evidence from Big Field Experiments at Facebook (Gordon, Zettelmeyer, Bhargava and Chapsky, Marketing Science, 2019); Simplifying Ad Measurement for a Social-First World (Meta for Business, 2026); The Unfavorable Economics of Measuring the Returns to Advertising (Lewis and Rao, Quarterly Journal of Economics, 2015).

Related answers

• Multi-touch attribution models explained with one order
• Why multi-touch and last-click numbers disagree
• Do you need attribution software? A sizing test
• Cut the channel, but hold it out first
• Incrementality testing