There are only two attribution models left

Google removed first click, linear, time decay and position based attribution. New conversion actions lost access in June 2023, and the forced migration finished in September 2023. In Google Ads and GA4 today you get last click or data-driven, and that is the whole menu.

Plenty of marketing plans still contain a line about choosing an attribution model. There is nothing left to choose. The decisions that actually move your reported numbers are somewhere else, and they are almost never on the agenda.

The threshold nobody quotes

Data-driven attribution in GA4 is not automatic. It needs at least 400 conversions for that specific conversion action and at least 20,000 total conversions across all actions inside the lookback window. Below that, you are on last click and the interface will not shout about it.

Go and check yours. Most mid-sized Australian businesses with a considered purchase sit under the 400 threshold on their primary action, which means the sophisticated model they think they are reading is the simple one.

If you fail the threshold, you have two honest options. Count a higher-volume upstream event as a conversion action so the model has something to learn from, or accept last click and interpret it as what it is: a report on the final touch, useful for closing, blind to everything that created the demand.

Lookback windows do more than the model

The click lookback window decides which touches are even eligible for credit. Set it to 30 days on a business with a 70-day sales cycle and you have deleted most of your marketing from the record before any model runs.

Three things to set deliberately:

1. 1.Click window. Match it to the real gap between first touch and purchase, measured from your CRM, not guessed. If your median deal takes nine weeks, 30 days is wrong.
2. 2.Engaged-view window. Video that is watched but not clicked still gets counted here, and it changes what YouTube and Demand Gen appear to be worth. Know whether yours is on, and at what length.
3. 3.View-through window. Generous view-through windows make display and video look extraordinary. Shorten them, or hold them constant, and never compare two periods with different settings.

Changing a window changes history in the interface. Write down the date you changed it and put it on the same chart as the numbers. Half the mysterious performance shifts in a client account are a settings change nobody logged.

Primary and secondary conversions just became a bidding decision

There is a change here most teams have not caught up with. Google's journey-aware bidding, in beta on Search Target CPA campaigns, lets the bidder learn from non-biddable conversion goals: phone calls, form fills, newsletter signups. Previously those were reporting only.

That flips the meaning of a secondary conversion action. The stuff you marked as observe-only because it was noisy is now an input to how the machine spends your money. If your newsletter signup is counted and your newsletter attracts people who never buy, you have just taught the bidder to buy those people.

Do the audit. List every conversion action in the account, decide for each whether you would happily pay to get more of it, and remove or reclassify the ones where the answer is no.

Modelled conversions are already in your numbers

When consent is missing, Google reports modelled conversions rather than leaving a hole. That is defensible statistics, and it is also a number you cannot audit. The share that is modelled rises as consent rates fall, so two periods with different consent rates are not directly comparable even with identical settings.

Google Analytics and Google Ads also split on consent differently since 15 June 2026, when ad_storage became the sole control over whether GA4 shares cookies and identifiers with linked Ads accounts. The Google Signals toggle now only governs signed-in behavioural reporting. Accounts that never noticed can be over-collecting or quietly missing Ads data, depending on which direction they were set.

The audit is quick. Open DevTools, Network tab, filter for google-analytics.com, then compare the gcs parameter and the gcd parameter on a reject and an accept. Do not use GA4 Tag Diagnostics for this, it lags 48 to 72 hours.

What replaces the model argument

Three practices, in order.

Clean signal. Server-side tagging, enhanced conversions with hashed first-party data, a conversion action list that reflects money rather than activity. This is the whole game now, because both platforms bid off what you feed them.

One number that crosses channels. Pick a single blended figure, revenue divided by total marketing cost, tracked weekly. It is crude and it is honest, and it does not care which platform claims the sale.

Incrementality on the big lines. A geo holdout twice a year on your two largest channels gives you a correction factor to apply to platform reporting. That is the only measurement that survives the next reporting change.

Last click is not a model of reality. It is a filing system. Use it to file, and use a holdout to decide.