Is Your Branded Search Actually Incremental? The Retargeting Cannibalization Test

Probably not as incremental as the dashboard says. Across anonymized GA4 exports we read for brands in the €1M to €10M range, the same journey keeps appearing: a shopper clicks a retargeting ad, later googles your brand name, clicks the brand ad, and buys. Google claims the conversion. The retargeting platform claims the same conversion. Both book it as performance, and you pay twice for one order. The test below takes about 30 minutes on a GA4 export, needs no geo holdout, and shows how big your overlap is, plus where pattern detection ends and a causal read begins.

Why do branded search and retargeting both claim the same sale?

Because the platforms never compare notes, and each one grades its own homework.

Here is the journey, mechanically. A prospect visits your store and gets tagged. Your retargeting follows them for days across social feeds and display placements. When they decide to buy, they do the cheapest navigational thing available: they type your brand name into Google, click the first result, which is your own brand ad, and check out.

On the Google side, the brand click takes the credit. On the retargeting side, the same order falls inside the platform's attribution window and gets claimed there too. One order enters your books. Two platforms each report a conversion, each computes a healthy ROAS, and each asks for more budget next month.

The question neither platform is paid to ask is the counterfactual one: would this shopper have bought anyway? That is the incrementality question. If the answer is yes for a large share of these orders, part of both line items is a toll you pay on demand you already had, not a driver of new revenue.

To be fair, some branded spend is defensive. If competitors or resellers bid on your name, disappearing from that auction costs real sales. The goal of the test is not to zero the line. It is to separate the defensive share from the toll-collecting share.

Why is this the first line item your CFO audits in 2026?

The 2026 tariff and margin squeeze changed the budget conversation. When COGS inflation eats gross margin, finance stops accepting attributed revenue at face value and starts auditing the categories where double counting is structural. Branded search and retargeting are exactly those categories, and they are meaningful money: for one anonymized €4M skincare brand we read, the two lines together were 28% of paid media spend in Q2 2026.

Until recently, the only credible answer came from enterprise incrementality vendors, and they gate it behind five-figure experiments: geo holdouts, matched markets, weeks of runway. For a brand spending €20K to €200K a month, that price kills the question before it gets asked. So the double claim compounds quarter after quarter, and both platforms keep reporting these as their best-performing lines. If the evidence is headed for a budget review, the CFO budget-defense kit shows how to package it.

There is also a reporting trap that hides the problem. If you add the two platform reports together, the combined performance looks better than your actual revenue from these customers, and the blended ROAS you present upward inherits the double count. Finance eventually notices when platform-reported revenue grows faster than the bank account, and that is when the audit lands on your desk instead of the other way around.

You do not need the enterprise experiment to find out whether you have the problem. You need an afternoon and your own data.

How do you run the cannibalization test on a GA4 export?

You need user or session-level data with timestamps, source and campaign fields, and transaction IDs. A GA4 BigQuery export is ideal; a well-built exploration with clean UTM parameters works for most brands. Then run five checks.

1. Isolate branded-search orders. Flag every conversion where the last non-direct touch was paid search on a brand term. Keep the brand list strict: exact brand name, common misspellings, brand plus words like discount or review.
2. Look back 24 hours. For each branded-search order, check whether the same user carried a retargeting touch inside the prior 24 hours. That share is your headline overlap rate.
3. Time the retargeting click. If the median gap between the retargeting click and the order is minutes rather than days, the ad caught someone already on the way back to buy.
4. Compare brand demand with brand spend. If branded impressions and click prices climbed for two quarters while brand search volume stayed flat, you are paying more to harvest the same demand.
5. Check new-customer rates. Retargeting that mostly converts returning visitors with full baskets is not creating customers; it is accompanying them to checkout.

A parallel interception pattern sits one step later in the journey, at the checkout itself, and the coupon and affiliate incrementality audit walks through it. Optional confirmation pulse: pause your brand-term ads for one week and watch total brand-name revenue. This is not an enterprise geo holdout; it is a cheap, slightly noisy self-test. If revenue barely moves, your branded ads were mostly a toll booth.

What does the overlap look like in real numbers?

Here is an illustrative, anonymized example from a €4M home-and-living brand, Q2 2026:

Two things stand out. First, the combined platform-claimed revenue of €159,700 exceeds what the shop actually booked from these customers, because both platforms counted the same orders. Second, the channels do not go to zero under scrutiny: in this example a third to two fifths of claimed value held up in the causal read. The actionable finding is narrower and more valuable. A specific slice of spend, on the order of €20,000 per quarter here, was subsidizing orders that would have happened anyway.

What happens next is usually a reallocation, not an execution. In this example the brand kept a defensive branded core, capped retargeting frequency, and moved the freed budget into prospecting that a causal read showed was creating genuine first touches. Reported platform ROAS on the two lines dropped afterwards, which is the point: the pretty number was the expensive one.

Where does pattern detection end and a causal read begin?

Be honest about what the five checks give you: strong suspicion, not proof. An overlap rate is a correlation between touches. Some doubly-touched buyers genuinely needed the retargeting nudge; some branded-search buyers would have found you through a reseller had the ad been gone. Patterns tell you where to dig. They do not tell you how much to cut.

The counterfactual, what would have happened without the spend, is a causal question. Incrementality testing with a geo holdout is the most direct evidence, and it remains the right tool when you have the scale and the budget. The SMB-feasible path is causal inference on the observational data you already have: using natural variation in your time series to estimate how conversions would have behaved without the impressions. That is the approach behind proving a channel caused revenue without running an experiment.

How often should you re-run this? Quarterly is the right cadence for most brands, plus once before budget season and once after any major change in retargeting spend or brand bidding strategy. The overlap rate drifts as creatives, audiences, and competitors change, so a one-time read goes stale within two quarters.

We built Causality Engine for exactly this read: GA4 export in, causal read out, €99 per read, 5 to 10 minutes, no pixel, no annual lock-in, with Pro at €299 per month if you want it always on. Check the per-read pricing and run it on last quarter's export tonight.

Key takeaways

• Branded search and retargeting each claim full credit for the same orders, and both platforms book the overlap as performance; the dashboard ROAS on these two lines is structurally inflated.
• The headline diagnostic is free: the share of branded-search converters with a retargeting touch inside the prior 24 hours, computed from a GA4 export in about 30 minutes.
• Flat brand demand against rising brand spend, and minute-level click-to-purchase gaps on retargeting, are the supporting tells that you are paying a toll on existing demand.
• Pattern detection ranks suspicion but cannot produce the counterfactual; geo holdouts remain the gold standard if you can afford five figures, and causal inference on observational data is the SMB-feasible read.
• Cut the toll, keep the defense: reduce the non-incremental slice, keep a defensive branded core, and re-read quarterly through budget season.

Further reading

• The Coupon and Affiliate Incrementality Audit: Stop Paying for Sales You Already Had
• The CFO Budget-Defense Kit: Proving Marketing Caused Revenue in a Margin-Crunch Year
• How to Prove a Channel Caused Revenue Without Running an Experiment
• Causality Engine pricing: €99 per causal read, €299 per month Pro