What do early, mid and late touchpoints mean in GA4?

Early touchpoints are the first 25% of a path, rounded to the nearest whole touch. Mid is the middle 50%, and late is the last 25%. A one-touch path has no early or mid segment, so late takes all its credit.

Why are there decimals in the Key events column?

Because the data-driven model shares one key event between several touches. Google calls it fractional credit: in its example, two keywords each get part of a key event, and the parts add up to one. It is expected, not a tracking error.

Does changing the attribution model change past data in GA4?

Yes for the model, no for the lookback window. Google says a new reporting attribution model applies to historical and future data. A new lookback window applies going forward only, so note the date you change it.

How to read the GA4 conversion paths report, step by step

In GA4, open Advertising, then Attribution paths, and keep only the purchase key event. Read the early, mid and late chart and switch its model from last click to data-driven. Then sort the table by days and touchpoints, and filter out one-touch paths.

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

In GA4, open Advertising, then Attribution paths, and keep only the purchase key event. Read the early, mid and late chart, then switch its model from last click to data-driven and see what moves. If one-touch paths dominate, filter them out: no model can split them, so the argument lives in the longer paths.

The report answers two questions at once: who touched your buyers, and how one model shares out the credit. Read them in that order and it makes sense. Read them together and it reads like a horoscope.

Step by step

Open the report. Click Advertising on the left, then Attribution paths under Attribution. Google's page for the report calls it Key event attribution paths, under a Key events drop-down, so look for either label. Menu path: Advertising > Attribution > Attribution paths.
Keep purchase and untick the rest. GA4 selects every key event by default and adds them together. With ads personalization on and Google Ads linked, it also marks steps like add_to_cart and session_start as key events. Leave session_start ticked and ordinary visits start counting as paths. Menu path: key events drop-down at the top left of the report.
Read the chart's three segments. Early touchpoints are the first 25% of a path, mid the middle 50% and late the last 25%. A one-touch path gives late all the credit, and paths under three touches leave mid empty. Menu path: the bar charts under Early, Mid and Late touchpoints at the top of the report.
Check which model drew the chart. It starts on Paid and organic last click, which ignores direct traffic and gives the whole sale to the last channel clicked. So the early and mid bars only fill when direct visits close a path. Menu path: attribution model drop-down above the chart.
Switch the chart to data-driven. Data-driven shares each sale across the touches before it, so decimals can appear in the Key events column. Write down the channels whose early credit grows the most. Menu path: attribution model drop-down > Data-driven.
Change the dimension when one channel needs a closer look. The chart and the table both start on Primary channel group. The drop-downs at the top switch them to Source, Medium or Campaign. Menu path: dimension drop-downs above the chart and the table.
Sort the table by time, then by length. The table starts sorted by the paths with the most conversions. Click the arrow next to Days to key event, then Touchpoints to key event, to bring slow and long paths to the top. Menu path: column headers in the data table.
Split short paths from long ones. Use the path length filter: pick an operator, enter a number next to touchpoint(s) and click Apply. Equal to 1 shows the paths no model can split; greater than 1 shows the rest. Menu path: Customize report > Path length.

Two habits keep these steps honest. Use the same date range every time you switch models, and write the model next to any number you share.

A worked example

For illustration, here are four invented paths with round revenue, split the way Google's segment rules split them.

Path (invented)	Touches	Revenue	Early	Mid	Late
Direct	1	€500	empty	empty	Direct
Paid Social > Organic Search	2	€200	Paid Social	empty	Organic Search
Paid Social > Email > Paid Search	3	€200	Paid Social	Email	Paid Search
Organic Social > Paid Search > Email > Organic Search	4	€100	Organic Social	Paid Search, Email	Organic Search

Under the default model, each sale goes to its last non-direct click. In this worked example, that puts all €1,000 of credit in late: Direct keeps its €500, and the other three paths credit their final touch. The early and mid bars stay empty.

In the worked example, data-driven can move some of the €500 from the three longer paths into early and mid. In this worked example, only the 3-touch and 4-touch paths have a mid segment, worth €300 of the €1,000 between them. In the worked example, the €500 Direct path stays in late under every model.

Notice what the chart cannot say here. In this worked example, Paid Social opens both the €200 two-touch path and the €200 three-touch path. Whether those sales needed Paid Social is a different question, and no segment answers it.

So half the invented revenue sits where no model can reach it. In one store's export, the Journeys sheet puts 79.5% of revenue in journeys with 1 touch. On the same Journeys sheet, the journeys a model can argue over hold the other 20.6%.

That is the habit worth building: before you compare models, find out how much revenue they are actually fighting over.

What should I check when the report looks wrong?

Every bar sits in late. You are probably on the default last click model, or most of your paths are one touch long. Switch the model first. If little moves, filter path length to equal to 1 and see how much revenue the short paths hold.

The report is empty. Without a key event, there is no path to show. Check that purchase is starred as a key event under Admin > Data display > Events, then reload the report.

The top row's revenue looks too high. Several key events are ticked and added together. Untick all but purchase and check the total again.

Rows show decimals. That is fractional credit from the data-driven model, not a bug. In Google's example, two keywords share one key event, and their parts add up to one.

A row's credit looks too neat. The credit in each row is an average across the individual paths in it. Hover over a touchpoint in a path to see its own credit and key events.

Old months look thin. GA4's data-driven model starts on November 1, 2021, and the rules-based models on June 14, 2021. A date range that reaches back before those dates shows partial data.

Paths look shorter than your buyers' real journeys. The key event lookback window decides how far back a touch can earn credit. For purchases it is 90 days unless someone chose 30 or 60, and changes apply going forward only.

Google Ads shows different paths. Google Ads keeps its own Conversion paths report, under Attribution in the Goals menu, built from ad interactions. It will not match GA4's channel paths, and it is not meant to.

What to do this week

Read your slowest paths against your lookback window. Sort the table by Days to key event, then check the window under Admin > Data display > Events > Attribution settings. Pass: your slow, common paths fit well inside the window. Fail: the window is 30 days and your long paths come close to it.
Trace one campaign through the segments. Switch the dimension to Campaign and the chart's model to Data-driven, then find your largest prospecting campaign. Pass: it earns credit in early touchpoints. Fail: it only appears in late, so it closes sales that other channels opened.
Put Google Ads' paths beside GA4's. In Google Ads, go to Attribution in the Goals menu and open Conversion paths for the same month. Pass: you can say which report each number came from. Fail: someone quotes the Google Ads paths as if they covered every channel.

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: Key events attribution paths report (Google Analytics Help); Get started with attribution (Google Analytics Help); Select attribution settings (Google Analytics Help); Mark events as key events (Google Analytics Help); About attribution reports (Google Ads Help).

Frequently asked questions

What do early, mid and late touchpoints mean in GA4?
Early touchpoints are the first 25% of a path, rounded to the nearest whole touch. Mid is the middle 50%, and late is the last 25%. A one-touch path has no early or mid segment, so late takes all its credit.
Why are there decimals in the Key events column?
Because the data-driven model shares one key event between several touches. Google calls it fractional credit: in its example, two keywords each get part of a key event, and the parts add up to one. It is expected, not a tracking error.
Does changing the attribution model change past data in GA4?
Yes for the model, no for the lookback window. Google says a new reporting attribution model applies to historical and future data. A new lookback window applies going forward only, so note the date you change it.

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