How to Tell If Your Paid Media Spend Is Truly Incremental

Your Google Ads dashboard says last month produced 214 conversions at a 5.2x ROAS. Your finance lead looks at the P&L and cannot find the money. Both views can be correct. That is the problem.

Platform reporting and financial reality can disagree for years because they answer different questions. The dashboard asks: which conversions touched our ads? The business asks: which conversions happened because of our ads? The gap between those two questions is where wasted budget lives.

This article is about closing that gap. Not with a new tool, but with a definition and a sequence of tests you can run in order of cost, starting free.

Every ad platform reports attributed conversions: sales and leads where the platform can show its ad was seen or clicked somewhere along the way. Attribution is a record of contact. It is not a record of cause.

The distinction matters because some of the people your ads touch were already on their way to you:

• The customer who searched your brand name, saw your ad above your own organic listing, and clicked it. One paid conversion, zero new customers.
• The basket abandoner who was retargeted for a product already sitting in their basket, on a site they already intended to return to.
• The repeat buyer inside a broad remarketing audience who buys on schedule every eight weeks, ad or no ad.

In every case the platform records a conversion, charges for the click, and takes the credit. In every case the honest answer to "did the ad cause this sale" ranges from "partly" to "no". Platform-reported conversions are usually the ceiling of your ads' contribution, not the floor.

This is not about bad intent from the platforms. It is a structural feature. The system that spends your budget also writes the report card, and it grades on attribution because attribution is what it can see. We covered how this bind tightened over the last decade in PPC in 2026 vs 2016. The short version: automation made accounts more efficient at finding conversions, and also made it harder to see which conversions were real wins.

Three places account for most of it.

Brand absorption. Campaigns that serve on your own brand terms often convert well because the customer had already chosen you. The strongest published warning sign is eBay's paid search experiment: researchers found no measurable short-term benefit from eBay's brand-keyword ads, and much of the traffic moved to free channels when the paid ads stopped. eBay is an extreme case, a very large brand with strong organic visibility, so the result does not generalise to every business. Your brand terms can still carry incremental value, especially where competitors bid on your name. But "some value" and "full platform credit" are not the same claim. If Performance Max, broad match, or AI Max absorbs brand traffic and blends it into the results, reported ROAS can look stronger than the business result. If your best-performing campaign contains your brand name in its search terms, read how to audit what Performance Max is actually doing.

Retargeting harvest. Retargeting audiences are, by design, people already engaged with you. Some genuinely need the nudge. Others were coming back anyway. Platform attribution cannot reliably tell these groups apart. That is why retargeting can report the highest ROAS in the account while adding less new revenue than the report implies. The tell is retargeting "performance" that scales with your site traffic, not with the retargeting budget.

Already-converting demand in automated campaigns. Smart Bidding optimises towards the conversion actions you give it. If those actions over-credit brand, retargeting, or low-quality leads, automation can scale the wrong signal. SMEC's analysis of more than 250 Search campaigns using AI Max found a median 13% increase in conversion value alongside a median 16% increase in CPA, with a wide ROAS spread across accounts. That is not an argument against automation. It is an argument for controlled testing: extra attributed conversion value is not the same as extra profit.

None of this means the spend is worthless. It means the reporting cannot tell you which part is worth anything. That requires a different definition.

Incremental revenue is revenue that would not have happened without the spend. Nothing else counts.

Written as a question you can put in a board pack: if we had not spent this budget, how much of this revenue would we have received anyway? Platform ROAS assumes the answer is "none of it". Reality is somewhere between "some" and "most", and the entire discipline of incrementality measurement is about locating your account on that line.

Two useful terms fall out of the definition:

• Incremental ROAS (iROAS): incremental revenue divided by spend. It is usually lower than platform ROAS. This is the number that should govern budget decisions.
• Incrementality rate: the share of attributed conversions that were genuinely caused. Brand and retargeting often test lower than their platform reports imply. Cold prospecting can look worse in-platform than it really is, because some buyers convert later or through another route. Blended accounts sit between those extremes, and the platform will not tell you where.

For a worked example of how far the two numbers can diverge, see our short explainer: platform ROAS vs incremental ROAS.

You do not need a data science team to start. You need a ladder: each rung costs more and answers more precisely. Climb only as high as your spend justifies.

Rung 1: Reconciliation (free, start this week). Take 90 days of platform-reported conversion value across all channels and add it up. Put blended revenue from your CRM or finance system next to it. In some accounts, the platforms collectively claim more revenue than the business recorded. That single comparison, platform claims versus money that exists, is the cheapest incrementality signal available, and it reframes every conversation that follows. While you are there, run the tracking audit checklist: double-counting and broken deduplication create fake incrementality problems that are cheaper to fix than to test.

Rung 2: Structural separation (free, analysis only). Split every report into brand and non-brand. Separate retargeting from prospecting. Now look at where reported performance concentrates. If most of your blended ROAS is coming from brand and retargeting, the account's headline number is mostly harvest, and the signs of waste will usually be visible from here.

Rung 3: Pause tests (cheap, some nerve required). Pick a candidate segment, such as brand search with no competitor pressure, or a retargeting layer. Pause it for two to four weeks. Watch blended revenue, not platform metrics. If revenue holds while spend drops, you have found likely non-incremental budget. If revenue dips beyond normal variance, you have evidence of real contribution and can restart with more confidence. Illustrative example, not a client figure: an account spending £4,000 a month on brand search pauses it for three weeks. Organic brand clicks rise. Total orders stay within normal variance. That £4,000 was probably buying traffic the SEO listing was already winning.

Rung 4: Geo and holdout experiments (stronger evidence). Run the spend in some matched regions and not others, or hold out a slice of the audience, and compare; the full design guide is in what is a geo holdout test. This is one of the closest paid media methods to a controlled test. It needs enough volume to read a signal and a few weeks of patience. Design each test with a pre-agreed pass mark and end date, or the result can be argued into whatever the loudest stakeholder wanted.

Most businesses under £100k a month of spend can get the main practical value from rungs 1-3. This is the core of how we run paid media efficiency engagements: baseline against revenue first, separate the harvest from the hunt, then test the biggest suspect segments before touching budget allocation.

Three conservative rules keep the answers honest.

Respect seasonality and noise. A two-week pause during a demand spike proves little in either direction. Compare test periods to matched prior periods. Treat any effect smaller than your normal week-to-week variance as "not proven" rather than "proven zero".

One change at a time. A brand pause run in the same month as a site migration and a price change measures nothing. Boring test hygiene is the whole game.

Expect uncomfortable asymmetry. The segments that look best in platform reporting, usually brand and retargeting, are often the first to shrink under incrementality testing. The segments that look expensive, such as cold prospecting, can be undervalued because attribution loses sight of buyers who convert later or on another device. Incrementality testing usually argues for spending braver, not just spending less.

That last point matters. This is not an exercise in cutting budgets. It is an exercise in moving budget from spend that flatters the dashboard to spend that grows the business. When teams run this discipline well, they can spend with more confidence because finance can see the logic. That is also the foundation for a measurement framework both teams can trust.

Run this against your own account this week:

• Sum 90 days of platform-claimed revenue across all channels. Put blended revenue next to it. Note the ratio.
• Split brand from non-brand in every report. No exceptions.
• Separate retargeting from prospecting and note where reported ROAS concentrates.
• List your three highest-ROAS segments. Ask of each: would this customer have converted anyway? If the honest answer is "probably", it is a pause-test candidate.
• Check whether Performance Max is serving on brand terms.
• Verify tracking before testing: deduplication, consent, offline conversions.
• Design one pause test with a pre-agreed pass mark, start date, and end date.

If the ratio in step one is uncomfortable, or steps four and five exposed a weak spot, that is normal. Most accounts have never been asked these questions. And if a budget increase is on the table, work through how to audit paid media waste before you increase budget first.