Marketing analytics: attribution explained honestly, and a measurement plan you can keep

Marketing analytics works when you pick one source of truth per question instead of expecting one tool to answer all of them. GA4 is the right place to see which channels and pages bring people who complete a key event, and its Attribution reports offer three models: data-driven, paid and organic last click, and Google paid channels last click. Channel spend reconciles in a spreadsheet or warehouse, offline revenue lives in your CRM, and the question of whether a channel caused the sale is settled by holding it out and watching what happens. Write that down as a one page measurement plan and the monthly reporting argument ends.

This page covers attribution, what to ask GA4, and a measurement plan a small team can keep running. For building the GA4 tags themselves, follow Google's setup documentation.

Attribution, honestly

Attribution is the act of assigning credit for an important action to the touchpoints along the path to it. That is all it is: a rule for splitting credit. It is a reporting convention, not a measurement of cause.

Google's attribution documentation lists three models available in the Attribution reports in Google Analytics properties:

Model	How credit is assigned	What it is good for
Data-driven	A model built from your own account's data, comparing converting and non converting paths. It uses factors such as time from the key event, device type, number of ad interactions, the order of ad exposure and the type of creative	Accounts with enough volume, when you want credit spread across the path rather than parked on the last click
Paid and organic last click	The last paid or organic channel before the key event takes all the credit	A simple, stable number that everybody understands and nobody can game
Google paid channels last click	The last Google paid click takes the credit	Reconciling GA4 against Google Ads reporting

One detail changes how you read every one of them: Google notes that all of these models exclude direct visits from receiving credit, unless the whole path was direct. So "direct" in an attribution report is quieter than direct traffic really is.

The useful habit is to pick one model as your reported number, keep it for the year, and look at a second model only to ask a specific question. Switching models mid year produces a change in the report that nobody can explain to a finance team. Set the model and the lookback window deliberately in attribution settings, and note the date you changed it.

What to ask GA4, and what to ask somewhere else

The question	Where it gets answered
Which channels bring people who complete a key event?	GA4, acquisition and attribution reports
Which landing pages earn the next click?	GA4, landing page report plus Search Console for query data
Where do people drop out of a funnel?	GA4 funnel exploration
What did each channel cost, and what did it return?	A spreadsheet or warehouse joining platform spend to GA4 or CRM outcomes
Did the deal actually close, and for how much?	Your CRM, with the lead source stamped at creation
Did this channel cause the sales, or just sit near them?	A holdout: pause it in some regions, keep it in others, compare
Why did this page lose traffic?	Search Console, impressions and average position by query

That last row is the one most teams skip. A geographic holdout or a paused campaign for two weeks tells you more about a channel's real contribution than any model, because you changed one thing on purpose.

Setup that prevents most reporting arguments

Nine tenths of mistrust in analytics traces back to six setup decisions. Make them once, write them down, and tell whoever joins next.

• Define key events as money moments. Key events should be the actions you would pay for: a form submitted, a call started, a purchase. Scroll depth and video plays are useful events, not key events.
• Name events once, in a convention. Lower case, underscores, verb then noun: form_submit, quote_request. Renaming later splits your history.
• Filter internal and developer traffic. Use GA4 data filters so your own team's sessions are not in the numbers you report.
• Standardise UTMs. One sheet, lower case, fixed values for utm_medium. Every inconsistent medium becomes a new channel row in your report.
• Turn on User-ID if people log in. User-ID stitches activity across devices for signed in users, which makes the user counts mean more.
• Record the dates you changed anything. A change log next to the dashboard answers the "why did this jump" question in seconds.

One more thing to know when somebody compares GA4 against an old Universal Analytics number: Google's own comparison notes that Universal Analytics highlighted Total Users while GA4 reports Active Users, so a column labelled Users in each tool is calculated differently. Compare like with like or compare nothing.

A measurement plan on one page

This is the whole deliverable. Five rows, one page, agreed by marketing and finance before the quarter starts.

• The objective. One sentence, with a number and a date.
• Two or three reported metrics. For each one: the definition, the tool it comes from, who pulls it, and how often.
• The attribution model and lookback window you will report on all year.
• The diagnostic metrics you look at when a reported metric moves, listed in the order you check them.
• One experiment per quarter, named, with the holdout written down before it starts.

Is Google Analytics 360 worth it?

Analytics 360 is the paid tier for GA4 properties, and the difference is headroom rather than new reports. Google's 360 documentation sets out higher limits: event parameters rise from 25 to 100 per event, event scoped custom dimensions from 50 to 125, key events from 30 to 50, audiences from 100 to 400, and exploration sampling from 10 million to 1 billion events per query, with unsampled explorations available only on 360. It is also covered by a service level agreement.

So the honest test is whether you are hitting those ceilings. If your explorations are sampled, you have run out of custom dimensions, or you need the SLA for contractual reasons, 360 solves a real problem. If not, the money buys more in tagging and warehouse work. Pricing is contractual, so ask Google or a reseller for a quote rather than working from a figure you read somewhere.

Where people get stuck on marketing analytics

Which attribution model should I use?

Pick paid and organic last click if you want a stable number everybody understands, or data-driven if your volume supports it and you want credit spread along the path. Then keep it. The model you keep is worth more than the model that is theoretically best, because consistency is what makes a trend readable.

Why do GA4 and Google Ads report different conversion numbers?

They count against different rules. Google Ads credits the click that led to the conversion and reports it on the date of the click, while GA4 reports the key event on the date it happened and applies its own model and lookback window. Use Google Ads numbers to manage bids and GA4 numbers to compare channels, and reconcile once a month rather than chasing a match.

How do I measure channels that do not produce a click?

With holdouts and with asking. Run the channel in some regions and not others, and compare. Add a one line "how did you hear about us" question to the form and read it alongside the report. Both are cheap, and together they cover what click data leaves out.

Do I need a data warehouse?

You need one when you are joining spend, GA4 and CRM data by hand every month and the join takes longer than the thinking. Before that point, a disciplined spreadsheet with a fixed layout does the same job. GA4's BigQuery export is the usual first step when you cross that line.

What is the first thing to fix in a messy GA4 property?

Key events. Make sure the ones marked as key events are the actions worth money, that each fires once per real action, and that internal traffic is filtered out. Everything downstream improves when the denominator and the numerator are both trustworthy.

One question your dashboard does not have a row for yet

A measurement plan lists questions and the tool that answers each. Add this one: when a buyer asks an assistant about our category, are we named, and which sources does it use?

No attribution model in GA4 will answer it, because there is no click to attribute. It is closer to brand tracking, and it belongs on the same page of the plan with an owner and a cadence. Monthly is enough for most teams. AI Knows Us supplies the data for that row.