How to test data-driven against last click before switching

Usually neither wins everywhere, so test them on your own numbers before you switch. Compare the two on purchases only in GA4, then price both in Google Ads' Model comparison report. If you switch, move one conversion action, reset its bid targets and wait out the lag before you judge.

Step by step

You need GA4 with purchase marked as a key event, and a Google Ads account with a purchase conversion. Changing GA4's attribution settings needs the Marketer role or above. Give the whole test a month, not an afternoon.

1. Read both settings before you touch them. In GA4, note Reporting attribution model and Channels that can receive credit on the Attribution settings page. In Google Ads, open your purchase conversion and note its Attribution model. A new GA4 reporting model rewrites past reports too, so record the old one first. Menu path: GA4 Admin > Data display > Events > Attribution settings; Google Ads Goals > Conversions > Summary > your conversion > Edit settings.
2. Compare purchases only, at event time. Open the Attribution models report and pick purchase in the key events drop-down, because all key events are added together by default. Leave Reporting time on Event time, the default, so both models judge the same purchases. Ad interaction time would credit touchpoints in your date range, even when the purchase came later. Then set one Attribution model (non-direct) column to Data-driven and the other to Paid and organic last click. Menu path: Advertising > Attribution > Attribution models.
3. Price both models in Google Ads. Open Model comparison and set Last click against Data-driven with the Compare and With drop-downs. Use the Cost / conv. and Conv. value / cost columns to read CPA and ROAS under each model. Campaigns that look better under data-driven are the ones last click undervalues. Menu path: Google Ads > Goals > Attribution > Model comparison.
4. Check what the comparison leaves out. Model comparison excludes conversions from the Search Partner Network, Gmail and App campaigns. It also filters out any network or campaign without enough data, with a note about partial history. If a campaign you care about is missing, pick a more recent date range. Menu path: Google Ads > Goals > Attribution > Model comparison.
5. Switch one conversion action, not all of them. Use the Switch to DDA tab in the attribution reports, or open the conversion, choose Data-driven under Attribution model and click Save, then Done. One action at a time keeps cause and effect readable. Menu path: Google Ads > Goals > Conversions > Summary > your conversion > Edit settings > Attribution model.
6. Add the current model columns. On the Campaigns page, open the Columns menu and add Conversions (current model) and Conv. value / cost (current model) from its Attribution section. They show your recent history as the new model would have counted it. Set them beside the regular columns. Menu path: Google Ads > Campaigns > Columns > Attribution.
7. Move each Target ROAS by the same percentage. Work out the percentage change from Conv. value / cost to its current model column, campaign by campaign. Google's help says to move the old target by that same percentage. Leave the most recent few weeks out of the date range, so the lag does not skew it. Menu path: Google Ads > Campaigns, regular columns beside the current model columns.
8. Wait out the lag before you judge. Open Path metrics, set Measure from to the first ad interaction and read Avg. days to conversion. The Campaigns page reports conversions by the date of the ad interaction, so a switch can bring a temporary dip in recent days. Google recommends waiting until the average days to conversion have passed. Menu path: Google Ads > Goals > Attribution > Path metrics.

A worked example

For illustration, say you run two Google Ads campaigns on Target ROAS: Brand and YouTube. In this worked example, Model comparison shows Brand's Conv. value / cost at 8.0 under last click and 6.0 under data-driven. YouTube shows 1.0 under last click and 2.0 under data-driven.

Credit moved from the campaign that closes to the one that starts journeys. Nothing about your sales changed, only who gets the credit.

Say you switch the purchase conversion to data-driven. In this worked example, Brand's ratio falls by a quarter, so its target falls by a quarter too. If the Brand target was 700%, it becomes 525%.

YouTube's ratio doubles, so its target doubles. If the YouTube target was 150%, it becomes 300%. Leave the old targets in place, and bidding would underbid on Brand and overbid on YouTube.

Then the lag. Say the Path metrics report shows 9 days on average from first interaction to conversion. In this worked example, you judge the switch no sooner than 9 days later, on a date range that skips the latest weeks.

On one store's Journeys sheet, journeys of 2 to 3 touches took 12.5 days to buy. Journeys of 4 to 9 touches took 16.9 days on the same Journeys sheet. If your buyers are that slow, a switch judged after one week is judged on half-finished journeys.

And YouTube? Its ratio doubled on paper, but credit is not cause. Before you scale it, pause it in some regions and compare total sales with the regions where it kept running.

What should you check when the numbers look wrong?

• Conversions dipped right after the switch. The Campaigns page dates conversions by the ad interaction, and data-driven spreads credit back along the path. A temporary dip in recent days is expected, so judge after the lag.
• Model comparison totals differ from the Campaigns page. The attribution reports leave out Search Partners, Gmail and App campaigns, and they report by the time of the conversion. Add the by conv. time columns to the Campaigns page to line the two up.
• Offline sales are missing from Model comparison. Its modelling only processes conversions that occurred within the last 7 days. An offline conversion uploaded more than 7 days after the event is bypassed, so keep upload delays under a week.
• Tiny channels swing wildly in GA4. A channel with a handful of purchases can double on one sale. Judge the % Change columns on channels that carry real revenue.
• Spend jumped between campaigns overnight. The model you set changes how automated bid strategies optimise, so a switch moves bids as well as reports. Check whether you reset the targets in step 7.
• Your targets feel wrong a week later. They may be fine, with the lag still running. Check the date of the switch against Avg. days to conversion before you touch them again.

What to do this week

1. Price your top campaign under both models. In Model comparison, read Conv. value / cost for your largest campaign under Last click and Data-driven. Pass: both sit on the same side of its Target ROAS, so the model will barely change your bidding. Fail: they straddle the target, so plan the switch and the target reset together.
2. Find actions still on last click. Open the Switch to DDA tab in the attribution reports. Pass: every primary purchase action runs on a model you chose on purpose. Fail: one runs on a model nobody remembers choosing, so decide and write the choice down.
3. Book the lag in your calendar. Read Avg. days to conversion in Path metrics and set a reminder for that many days after any model change. Pass: nobody judges the switch before the reminder fires. Fail: someone reads next morning's dip as a verdict.

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: Select attribution settings (Google); Key event attribution models report (Google); About attribution models (Google); About attribution reports (Google); About data-driven attribution (Google); Best practices for managing attribution model changes (Google)

Related answers

• All Attribution models questions
• Last click vs data-driven attribution, which is better?
• Last click or data-driven for a food and beverage store?
• Data-driven attribution, defined
• Incrementality testing, explained