State of Paid Acquisition: Q2 2025 and the Incrementality Turn

Four numbers from Meta's Q2 2025 results tell most of the story, and none of them is the one the headlines led with.

Revenue grew 22%. Ad impressions grew 11%. The average price per ad grew 9%. And in the same week, a marketing agency published the first big test of Meta's new incremental attribution setting. Put those side by side and you get this quarter's real theme: volume is growing, prices are still rising, and the question advertisers are starting to ask is no longer "how many conversions did Meta report?" but "how many did Meta cause?"

This is the Q2 issue of State of Paid Acquisition, a quarterly read on what is changing in paid media and what to do about it. I write these for operators who spend their own money and want the signal without the spin.

What did Meta report for Q2 2025?

A strong quarter. Revenue was $47,516 million, up 22% year over year. Advertising revenue was $46,563 million, up 21%. Net income was $18,337 million, up 36%. Daily active people averaged 3.48 billion in June, up 6%.¹

The two numbers that matter most to advertisers are in the middle of the release. Ad impressions rose 11%, and the average price per ad rose 9%.¹

How does that compare with Q1?

In Q1 2025 (see the Q1 issue), impressions rose 5% and the average price per ad rose 10%.² So the mix shifted. Impression growth more than doubled while price growth cooled by a point.

My read: Meta is serving a lot more ads, and demand for them is strong enough that the average price still went up. That is a healthy auction for Meta and a competitive one for you.

What does that mean for the prices you pay?

Less than you might think, and the reason is worth understanding.

"Average price per ad" is a company-wide average across every app, placement and format. It is not a CPM benchmark, and it does not tell you what your account will pay. A shift toward cheaper placements can lower the average while your costs rise, or the reverse.

The practical effect on budgets

When impressions and prices both rise, the same budget buys a different mix of reach and cost than it did a year ago. That is not a reason to spend less. It is a reason to know your break-even cost per customer to the dollar, so that a price swing in the auction shows up as a decision you made, not a surprise on a statement.

Read your own account instead

Three numbers answer the question better than any earnings release:

1. Your CPM trend over the last four quarters, by placement.
2. Your cost per result at a stable offer and audience.
3. Your frequency on the audiences that matter.

If CPM is up and cost per result is flat, your creative and offer are absorbing the price. If CPM is flat and cost per result is up, the problem is downstream: the page, the offer or the quality of the conversion event.

Where is automation actually paying off?

Meta says it is, in a lot of places. But the evidence comes from Meta, so I weigh it accordingly.

In February, Meta said early tests showed a 10% lower cost per qualified lead when Advantage+ was on for leads campaigns, and that advertisers who followed Opportunity Score recommendations saw a 5% median decrease in cost per result.⁴ When it rolled Opportunity Score out to all advertisers in June, Meta reported a 12% median decrease in cost per result for advertisers who adopted recommendations during testing.⁵ It also said Advantage+ shopping campaigns grew 70% year over year in Q4.⁴

What I take from that

Automation clearly works when the inputs are good. It is the part of the system that scales creative testing and finds pockets of demand that humans miss. But every one of those gains is a median across advertisers whose accounts, offers and tracking vary enormously. Automation multiplies whatever you feed it. Feed it clean conversions on a good offer and it compounds. Feed it noisy leads and it finds more of them, faster.

That is why I keep telling people to fix signal quality before they chase the next toggle.

What is the incrementality turn?

It is the shift from counting conversions the platform touched to estimating conversions the platform caused.

Standard attribution gives Meta credit when someone sees or clicks an ad and later converts. That is generous: some of those people would have bought anyway. Meta's incremental attribution setting tries to count only the conversions that would not have happened without the ads, by comparing the behavior of exposed and unexposed users.³

What Seer Interactive found

On July 30, the same day as Meta's earnings, Seer Interactive published results from testing the setting across six or more Meta accounts and about $1.05 million in ad spend in April 2025, comparing standard and incremental optimization and cross-checking against GA4.³

• Meta's view: 87% of conversions were incremental on average, so 13% would have happened regardless.
• GA4's view: 67% were incremental, so 33% would have happened anyway.
• By targeting: Broad targeting and hyper-narrow segments such as retargeting showed the lowest incremental performance. Refined mid-funnel audiences showed the clearest incremental impact.³

How to read the gap

Two tools looked at the same spend and disagreed by 20 points. Seer's own advice is the sensible one: treat Meta's self-reported incrementality as a directional signal, cross-reference it with independent sources, and expect the truth to be somewhere in between.³

I would add one more point. The finding that retargeting looks less incremental than people assume is the part to take seriously. Retargeting reports well because it reaches people who were already close to buying. That does not mean it caused the purchase.

How do you test incrementality on a normal budget?

You do not need a seven-figure budget to ask whether an audience is worth what you pay for it. You need one clean comparison.

Three simple designs

• Audience holdout. Split an audience in two, show ads to one half, withhold from the other, and compare conversions from both. Meta's lift tools do a version of this for you.
• Pause test. Pause one campaign for a defined period while everything else stays stable, then compare. It is noisy, but a dramatic drop or no change at all tells you something.
• Geography split. Run in some regions, hold out comparable ones, and compare results. This works best when the regions behave similarly before the test.

The rules that keep the result honest

Decide the success metric before you start, so you cannot move the goalposts. Run it long enough for your sales cycle: a consideration period of three weeks cannot be judged in three days. Change nothing else during the test. And write down what you expected, because the gap between expectation and result is where the learning is.

None of these designs is perfect. All of them are better than trusting a dashboard that was built by the party being measured.

What changed in the Meta toolbox this year?

It is worth listing, because the pace has been fast and the changes point in one direction.

• February: Meta tested a streamlined Advantage+ setup (see Advantage+ Sales Campaigns), renamed shopping campaigns as Advantage+ sales campaigns, extended Advantage+ to leads campaigns and began testing Opportunity Score.⁴
• June: Opportunity Score went to all advertisers globally.⁵
• By midyear: The incremental attribution setting was showing up in accounts, and on July 30 Seer Interactive published its test of it.³

The direction is clear: more automation, more recommendations and more measurement that Meta itself provides. Each piece is useful. Together they raise the stakes on one thing you cannot hand to the platform, which is knowing your own numbers.

What should you do in Q3?

Five moves, in order of importance.

What are the risks to watch?

Meta flagged a few in its own release. It has appealed the European Commission's Digital Markets Act decision, introduced a "less personalized ads" option in Europe in November 2024, and warned that changes could have a significant negative impact on its European revenue, possibly as early as later in Q3.¹ It guided Q3 revenue to $47.5 billion to $50.5 billion and narrowed its 2025 capital expenditure outlook to $66 billion to $72 billion.¹

If you advertise into Europe, plan for volatility. If you do not, the lesson is more general: platforms change the rules on their own timetable, and the advertisers who handle it best own their data.

The takeaway

Q2 2025 was a good quarter for Meta and a mixed one for advertisers: more inventory, higher average prices, more automation and, for the first time, a serious conversation about how much of the credit Meta deserves.

The advertisers who will do well are not the ones with the cleverest tricks. They are the ones who know their real cost per customer, send the platform honest conversion data and test before they trust. That is not a Q3 trend. It is the job, and it is what the system I build for clients is designed to support.

Next quarter, I will look at what changed and whether the incrementality conversation moved from agencies to everyday accounts.

Keep building.Ekai