Transcript
Welcome to Hard Questions, a SpendSignal podcast. I'm Sachin Uppal. Every episode, we take one question that no ad dashboard can answer, and work through it end to end. Today's question: How much of my Google 'best performer' is just collecting customers who were already coming — and what would I really lose if I cut brand search? My guest is Kabir Shah, Performance Lead at a skincare brand.
Kabir, what's your best-performing campaign?
On paper? Brand search. "Brand Exact." ROAS of eleven. CPA of ninety rupees. Every month it's the top line on the report.
You said "on paper."
Because I know what it is. It's people typing our brand name into Google. Somebody saw an influencer reel, or a Meta ad, or their friend's bathroom shelf, and then they Googled us. Google shows our ad, they click it, and Google takes the credit for a customer it didn't create.
So turn it off.
(laughs) That's the conversation I have with my CFO every quarter. Three problems. One: competitors bid on our brand name. If I go dark, some of those people click a competitor. Two: some of those clicks would come through the organic result for free — but I don't know how many. Three: if I turn it off and revenue dips that week, I'm the guy who broke the business.
What do your tools tell you?
Google Ads tells me brand search is a hero. GA4's data-driven attribution redistributes credit between touchpoints, but it's still only distributing credit among paths it can see — and it's Google's own model. Northbeam is better at showing me Meta's contribution to those journeys, but it's still attribution. All of them answer "who touched this order?" None of them answer "would this order have happened without the ad?"
And Semrush?
Tells me who's bidding on my brand. Useful for the competitor argument. Says nothing about incrementality.
So the honest answer today is?
A brand holdout. Turn brand search off in some regions for a few weeks, compare. We did it once with a partner. It was the most useful thing we did all year — and also expensive, slow, and nobody wanted to repeat it. And the answer was for that month. Since then our Meta spend doubled and we launched two products.
OK. So here's how SpendSignal approaches it. You connect Google Ads, Meta, GA4 and Shopify. On the Google side it pulls campaign spend, keyword performance — spend, clicks, conversions, quality score, impression share — and the search terms people actually typed. You also tell it your brand terms once.
Just our name?
Name, product names, common misspellings. That's how it splits brand from non-brand keywords. Now there are two lenses. The first is the channel level. The mix model estimates the incremental ROAS of Google Search from your history — how revenue moves with search spend after accounting for everything else, including Meta and the seasonality of your launches. Then it compares that with what Google reports. That ratio is called over-credit.
Give me a number.
In this example: Google reports a 6.2x ROAS on Search. The model says incremental is about 2.1x. Over-credit of roughly three. Google is claiming three rupees of revenue for every one it actually causes.
That matches my gut. But it doesn't tell me which part.
Right, so the second lens is keyword intelligence. It shows brand keywords as a share of search spend — say 41% — and their impression share — say 92%. That's the tell: you're paying for nine out of ten searches for your own name. Every keyword also gets tested against target CPA and against its own previous 28 days, with a verdict. And the search-term report flags queries that spend and never convert — "free sample," "dupe," "how to make at home."
So what does it actually recommend?
Something like: brand is over-credited, so cap brand spend and stop chasing impression share above what defends against competitors. Move that money into non-brand terms that pass their tests — say "serum for pigmentation," which beat target CPA with room to grow. Add the wasted search terms as negatives. And because the brand number matters so much, I'd confirm it with a geo holdout before cutting deep — SpendSignal flags that test automatically whenever Search is the channel with the widest range.
Wait — so it still tells me to run the test I didn't want to run?
For a decision this size, yes, and I think that's the honest answer. The difference is you're no longer guessing which test to run or why. And between tests, the model keeps updating as your mix changes, instead of relying on a number from last spring.
Let me be the CFO for a second. "Your model says search is over-credited. Google's says it isn't. Why should I believe yours?"
Because one of them is grading its own homework. Google's number is attribution inside Google's world. The mix model doesn't care who the click belonged to; it looks at whether total revenue actually moves when search spend moves. And it shows its uncertainty. If Search's range were 1.3 to 4.5, it would say "uncertain" and refuse to call it.
And what can't it tell me?
Good question, and I want to be straight. The model doesn't separately estimate how much brand search demand Meta or influencers created. It tells you how over-credited search is, not exactly who deserves that credit instead. If you need that, the holdout is how you get it.
Fair. So the CFO conversation becomes...
"Google Search is over-credited about three times. Brand is 41% of search budget at 92% impression share, which we mostly own anyway. We cap it at a defensive level, move the savings into non-brand terms that are proven below target CPA, and add twelve negatives. We confirm with a four-week holdout in three states before going further."
That's the first time brand search has been a decision instead of a religion.
That's Hard Questions. Thanks, Kabir. If you want to ask this of your own data, connect your accounts at spendsignal.co. See you next episode.