Platform lift tests
The platform splits people at random into a group that sees ads and one that doesn't. Google's version is Conversion Lift.
Best for: When your account is eligible. Clean, but run and reported by the platform being tested.
Draft: hidden from Google until Dan signs it off.
Cause
testing.
Cause testing, often called incrementality testing, shows what your ads actually added: the sales that wouldn't have happened without them. You hold ads back from a group of people or regions, keep everything else the same, and compare. Google offers Conversion Lift for this, and its researchers described the regional version back in 2011. I design and run these tests for businesses across Australia, from the Central Coast, NSW. It's how I decide where budget goes.
Written by Dan Picken, founder. Last checked 28 October 2026.From my audits
What tests
tend to find.
I can't share client results, but these are the patterns I see most often when the platform's numbers meet a proper test.
- 1
Brand search does less than claimed
It often looks like the best campaign in the account. Held back in a few regions, a lot of those sales still turn up through the free result underneath.
- 2
Retargeting takes credit for loyal customers
Ads shown to people already on their way back. The platform counts the sale. The test usually shows a much smaller real effect.
- 3
Prospecting does more than it looks
Ads to new people often look weak on last-click reports and stronger in a test, because the sale shows up later, somewhere else.
- 4
Calls go uncounted
In service businesses, people see the ad and ring rather than fill in a form. Without call tracking, the test is the only place that shows up.
- 5
The answer is spend less
Sometimes a test shows the last chunk of budget adds very little. Moving it, or saving it, is the most valuable result a test can give.
- 6
Tests too small to tell
Two weeks, one small region, a handful of sales. The result says 'no difference' and gets taken as fact. The planner below is there to stop that.
Why the dashboard isn't proof
The platform
marks its own homework.
Every ad platform reports the sales it can connect to an ad. That's attribution, and it's useful. It just can't tell you which of those sales the ad caused.
- 01
People who were coming anyway
Someone who searches your business name and clicks your ad was probably going to find you. The platform still counts the sale. eBay's experiment found its brand search ads had no measurable short-term benefit.
What I do: Test brand campaigns before calling them your best performers.
- 02
Seen, not clicked
Platforms can count sales from people who only saw an ad. Google's default windows include 1 day after an ad is seen.
What I do: Report view-based conversions separately, and test whether they hold up.
- 03
Estimates in the count
Google's Conversions column includes modelled conversions alongside observed ones.
What I do: Compare what the platform reports with what your CRM or till says.
- 04
Smart guesses still miss
In 15 large experiments at Facebook, common observational methods often failed to match the randomised results, usually overestimating the effect.
What I do: Where the decision matters, run a real test rather than a clever model.
None of this means ads don't work. Tests sometimes show ads doing more than the platform reports, especially when people see an ad and then phone. The point is knowing, not doubting.
Test planner
Can your test
actually tell?
A holdout test only works if the difference you're looking for is bigger than the everyday noise in your numbers. Put in your own conversions to see the smallest lift a test could reliably spot. If your ads add less than that, the test will say 'no difference' even when there is one.
Smallest lift this test could reliably spot
48%
Workable for a big question, like whether a campaign adds anything at all.
- Conversions in the ads group168
- Conversions in the held-back group72
A rough guide that treats conversions as steady and random. Real numbers swing more with seasons, promotions and uneven regions, so treat this as the best case. Ask a data person before betting a big budget on a close result.
The psychology of the dashboard
Why businesses
keep paying for ads that don't work.
The hard part of cause testing isn't the maths. It's that people, including marketers, are wired to believe the report that looks good.
01Reading the report
Confirmation bias
People tend to look for and favour evidence that supports what they already believe.
A flattering platform report gets accepted. An awkward one gets explained away.
What I do: Decide what result would change your mind before the test starts, and write it down.
02Deciding whether to stop
Sunk cost
Arkes and Blumer found people keep investing in something because of what they've already spent, not because of what it will return.
'We've spent three years building this campaign' is a reason people keep it, not a reason it works.
What I do: Judge every campaign on what the next dollar adds, not on its history.
03Taking the credit
Everyone counts the same sale
Each platform reports the sales it can connect to its own ads. Google, for example, counts conversions within its own attribution windows.
Add up what Google, Meta and the email tool each claim, and it often comes to more sales than you made.
What I do: Compare the platforms' total claims against your real sales. The gap is where testing starts.
Studies linked, so you can read them yourself. I'm a marketer: my job is turning them into changes you can test.
Types of test
Ways to test,
cleanest first.
The right test depends on your platform, your budget and how many sales you get. In rough order of how much you can trust the answer:
Geo holdout
Ads switched off in some regions, left on in similar ones, then compare total sales. The approach in Google's 2011 paper.
Best for: Most Australian businesses that sell across several regions. Measured with your own sales data.
Budget step test
Spend raised or lowered for a set period, then compared with the period before and similar regions.
Best for: Finding where more budget stops paying. See the Spend Curve.
On and off test
A campaign paused for a few weeks, then compared with before and after.
Best for: Small accounts with no better option. The weakest, because everything else changes too.
How most accounts are judged.
How I judge them.
Most accountsHow I do it
ProofThe platform's own report.A test that compares ads with no ads, using your sales data.Cause Testing
Brand searchCalled the best campaign.Tested before it gets the credit, or the budget.
Small accountsA quick test that proves nothing.A planned test, or an honest 'you can't test this yet'.
Bad newsExplained away.Reported plainly, with what to do about it.
How it works
Four steps.
Nothing hidden.
- 01
Free audit
I compare what your platforms claim with what your sales say, and check whether your account can support a test.
- 02
Pick one question
The one with the most money riding on it, like 'what do brand ads add?' or 'is Meta driving new sales?'.
- 03
Design and run the test
Groups or regions, length, what counts, written down before it starts. Run for weeks, not days.
- 04
Move the budget
A plain-English read-out, and budget moved to where the test showed it earns the most.
What you get
- A comparison of platform claims against your real sales
- A test plan: the question, groups, length and what counts, agreed before it starts
- Set-up in Google Ads, Meta or by region
- Monitoring while the test runs
- A plain-English read-out, including what the test can and can't tell you
- Budget recommendations based on the result
- Direct access to me, start to finish
Signals this works on
Out of the 8 signals every page and ad sends.
- Conversion trigger
Questions
people ask.
What is incrementality testing?
Measuring the sales your ads caused, not just the ones they're connected to. You compare a group that saw ads with a similar group that didn't. The difference is the increment.
Can every Google Ads account use Conversion Lift?
No. It isn't available for every account. Check eligibility and requirements in Google's Conversion Lift help. If your account isn't eligible, a geo holdout using your own sales data usually works instead.
How long should a holdout test run?
Long enough to cover your normal buying cycle and collect enough sales in both groups to tell a real difference. Usually several weeks. The planner on this page gives a rough guide from your own numbers.
Will turning ads off in some regions lose me sales?
It can, if the ads are working. That's the cost of finding out. A good test holds back as little as it needs to, for as short as it can.
Is my account big enough to test?
If you get only a handful of sales a week, a clean test is hard. It's often better to test a bigger change, like switching a channel off entirely, or run it longer. I'll tell you honestly if it's not worth it.
What is the difference between attribution and incrementality?
Attribution decides which ad gets credit for a sale. Incrementality asks whether the sale would have happened without the ad. Attribution is useful for running accounts. Incrementality is what budget decisions should rest on.
Find out what your ads
really caused.
Takes a minute. You'll hear back within 1 to 2 business days.
What happens next
- 1I'll ask for read-only access, or just your website for a search audit.
- 2I go through it and write up the 3 biggest fixes.
- 3A 20-minute call to walk you through it. No pitch unless you ask.
Keep going