If you've just seen your Marketing Mix Model and Google Branded Search is showing up as one of your most efficient channels, or as a surprisingly large share of your conversions, you probably have one of two reactions. Either "great, we should double down on Branded Search," or "that can't be right."

Both reactions are fair. Branded Search is the channel that most often produces MMM results that don't match business intuition. This article explains why that happens, what we recommend you do about it, and how to decide which option fits your specific situation.

The short answer

Branded Search frequently gets inflated credit in MMMs because it's strongly correlated with the moment of conversion, but it usually isn't what caused the conversion to happen. Something else (a brand campaign, word of mouth, a competitor's ad, a news mention) caused someone to search for your brand. The Branded Search ad then made it easy for them to find you.

The MMM sees the correlation. It can't tell what caused the search.

A quick way to spot this in a readout: if Branded Search requires a meaningfully larger share of spend (spend outweighing the impact) than share of modeled effect, and it still looks very efficient on a last-click basis, that is often a sign that the channel is harvesting existing demand rather than creating new demand. In that situation, a geo holdout is usually the right next step because it answers the causal question directly: what actually happens if we reduce or remove Branded Search in a set of markets?

The cleanest way to find out what Branded Search is actually contributing is to run a holdout test. Short of that, we can include it in the model with a biased prior that reflects the belief that it's less incremental than raw spend alone would suggest or we can consider removing it from your MMM altogether.

The question to ask yourself first

Before looking at the model, ask yourself: "Do I believe Branded Search is actually driving new customers, or is it making it easier for people who were already going to find us?"

This single question resolves most of the debate.

If your honest answer is "it's making it easier for people already looking for us," then Branded Search is a cost of doing business. If your honest answer is "it's defending against competitors who are bidding on our terms" or "it's genuinely introducing us to new customers searching the category," Branded Search is a real growth channel.

Either answer is valid. What matters is committing to the answer and building your measurement approach around it.

Why branded search is different from other paid channels

Imagine you sell coffee beans and advertise on TV. TV puts your brand in front of people who weren't thinking about you. The more you spend, the more people think of your brand when they crave coffee.

You also sell your beans in a grocery store. The store charges you a premium to stock at eye level; otherwise your beans go on the top shelf, where they're harder to reach.

Branded Search has more in common with the grocery-store placement than the TV ad. When you spend on Branded Search, you aren't reaching anyone new. You're making it easier for people who already intend to find you. When someone buys your coffee, the store placement probably wasn't what caused the purchase. It was the TV ad, a friend's recommendation, or something else.

Some argue Branded Search isn't advertising at all. It's infrastructure for converting existing demand.

That framing is important because it explains why an MMM can simultaneously show (1) a high correlation between Branded Search and conversions, and (2) very little actual incremental impact if you turned it off.

When branded search actually is incremental

The "cost of doing business" framing isn't universal. Three situations flip the picture.

• Competitors are aggressively bidding on your brand terms. If another company is buying your brand name and driving your click share down, Branded Search becomes a defensive investment. Pausing it hands demand directly to the competitor. This is common in categories with aggressive challengers and in markets where price competition is fierce.

• Your brand name is also the generic category term. Kleenex, Q-Tips, Uber. If customers search your brand name as a product noun, advertising on brand terms captures genuine generic demand, not just existing intent.

• The model may be under-attributing, not over-attributing. In some cases, the model's estimate for Branded Search is lower than the true contribution. This is less common but it happens. If your team's intuition is that Branded Search is genuinely driving customers and the model disagrees, a holdout test is the way to find out who's right.

Your Two Options

Option 1 (preferred): Exclude from the MMM and run a holdout

We drop Branded Search from the MMM entirely, which prevents the model from attributing credit it may not deserve. Then we run a holdout test in a set of geographies or an audience segment to measure what Branded Search is actually contributing.

• What's good about this. You get a clean, causal answer. The MMM is more trustworthy for your other channels because Branded Search isn't absorbing credit. And we aren't setting arbitrary priors that would be hard to defend later.

• What's uncomfortable. Your Branded Search spend won't appear in the model's output. Stakeholders who expect every channel to be represented may push back. We keep the spend visible in reports even though it's not in the model, but the experience of "Paramark is ignoring my Branded Search spend" needs to be expected and explained upfront.

Option 2 (fallback): Include in the MMM with a biased prior

We keep Branded Search in the model, but we set a prior that reflects the belief that it's less incremental than raw spend would suggest. In practice this usually means constraining the model to assume Branded Search efficiency is something like one-fifth to one-third of what a naive read of the data would imply.

• What's good about this. Your Branded Search spend appears in the model. The resulting estimates are more believable than an unconstrained model.

• What's uncomfortable. The specific prior we pick is a judgment call. It isn't as rigorous as a causal test. You may still end up with results you don't fully love.

We generally lead with Option 1, but Option 2 is a legitimate fallback, and for customers with tight timelines or very small Branded Search spend, it's sometimes the right first step.

How We Design the Branded Search Holdout

Branded Search holdouts are usually simpler than upper-funnel tests. A few design points worth knowing.

• The test can be shorter than a typical geo-lift. Branded Search is bottom-of-funnel, so a 3-week test often gives directional signal where an upper-funnel channel would need 4 to 6.

• We design against your target CPA, not the MMM's estimate. If your target cost per conversion is $X, we design the test to detect whether Branded Search is more or less efficient than $X. Using the MMM's estimate as the anchor would be circular.

• We map adjacent channels before launch. Affiliate activity, retargeting campaigns, promotions, or creative launches happening during the test window can contaminate the result. We check for these upfront and either pause them or explicitly lower the confidence bar for the test.

• Whole-country holdouts work for smaller markets. If you operate in multiple small markets, we can use one market as a 100% Branded Search holdout and use the others as controls. This is often cleaner than a sub-national split.

• Audience-based holdouts are an option. If you have the operational setup, we can split your addressable audience 50/50 at the cookie level, expose one half, hold out the other, and measure conversion rate difference. This works well when geo-testing is constrained.

What to Say to Your CMO, CFO, or Board

If you're explaining why Branded Search isn't in the model

"Branded Search correlates strongly with conversions, but it isn't typically what causes them. The MMM would overstate its impact if we left it in. We're running a holdout to measure what it's actually contributing, and we'll incorporate that result into the next model refresh. In the meantime the spend is still visible in our reports. We're treating it as a cost of maintaining our presence on our own brand terms."

If you're explaining a decision to exclude it or pull back

"Our measurement partner's recommendation, which I agree with, is that Branded Search is primarily capturing demand rather than creating it. We're treating it as a fixed cost of doing business and focusing our MMM on the channels that actually drive new customer acquisition. If we're wrong, the holdout test will show us."

If someone cites Google Conversion Lift numbers showing Branded Search as highly incremental

"Google's Conversion Lift Study measures incrementality inside Google's own ecosystem. For Branded Search specifically, it tends to capture demand that already exists, rather than demand the ad created. That's a known methodology constraint, not a signal that our MMM is wrong. A geo-based holdout is the cleaner read for this channel."

How We've Handled This at Other Customers

Every customer's situation is a little different. A few patterns we've seen:

• A B2B software customer ran a Branded Search holdout after the MMM showed suspiciously low cost per lead. The holdout returned no detectable lift at medium confidence. The team used that result to calibrate the next model and continues to pressure-test Branded Search through future experiments.

• A real estate marketplace designed a holdout against a specific cost-per-incremental hypothesis. Their belief was that Branded Search was under-credited in the model, not over-credited. The holdout covered a small share of total channel spend, so it was designed as a directional read rather than a high-confidence test. The learning was valuable regardless of direction.

• A consumer health business in a competitive market treats Branded Search as an active defensive investment. Competitors bid aggressively on their brand terms, which drove click share volatility from 60% down to 40%. Their question isn't "is it incremental?" but "what's the optimal spend level to hold share?"

• A B2B SaaS customer worked through three different output metrics before landing on one where the model behaved sensibly, then ultimately decided to exclude Branded Search entirely. The exclusion wasn't a data problem; it was a business decision that matched how they thought about the channel.

None of these customers is "right" or "wrong" about Branded Search. They're each making the measurement choice that matches their business model and strategic situation.

What This Means for Your Next Decision

Two actions worth taking.

• First, ask yourself the question at the top of this article. Your answer determines whether Option 1 or Option 2 is the starting point.

• Second, if Branded Search is more than a small share of your paid spend, bring it up with your Growth Advisor as a candidate for a holdout test. A three-to-six-week test gives you a defensible causal answer, and the result improves the MMM for every channel after it, not just Branded Search.

A practical signal to watch in your MMM readout

One pattern that often triggers this conversation is a large gap between Branded Search's share of spend and its share of modeled effect. For example, if Branded Search takes a meaningful slice of paid spend but contributes a much smaller share of incremental sign-ups, leads, or subscribers in the MMM, that's usually a sign to pause before scaling the channel based on platform-reported efficiency alone.

That gap does not automatically mean Branded Search is wasted spend. It usually means the channel is acting more like demand capture than demand creation. In that situation, a geo holdout is often the right next step because it answers the operational question the MMM cannot: what happens to conversions if we reduce or remove Branded Search in a defined set of markets?

A helpful rule of thumb: use the MMM to identify the mismatch, then use the holdout to decide whether the channel should be treated as a growth lever, a defensive cost, or something in between.

What to bring to a branded search incrementality test kickoff

If you're planning a kickoff with your Growth Advisor, the fastest way to get to a concrete test plan is to bring a few decisions and inputs into the first conversation.

• Your primary business question. Are you trying to learn whether Branded Search is incremental at all, whether it is efficient at your target CPA, or whether some portion of budget can be redeployed into other channels without hurting conversions? The test design depends on which of those questions matters most.

• The outcome metric that should move. Pick the KPI you actually care about before launch — for example leads, qualified leads, subscribers, purchases, or revenue. If you have several candidate metrics, identify the one you would use to make a budget decision and treat the others as secondary reads.

• Recent geo- or audience-level history. Share daily performance and spend data at the geo, market, or audience level for the metric you want to measure. This is what lets us estimate variance, assess power, and determine whether a geo holdout, audience split, or market-level pause is realistic.

• Your practical testing constraints. Come prepared to discuss whether you can pause Branded Search in selected markets, whether competitor bidding is active on your brand terms, whether there are legal or operational limitations, and whether there are adjacent channels or promotions that could contaminate the read during the test window.

• A budget or efficiency threshold. Directional tests are possible, but the planning gets better when you can say what level of incremental efficiency would count as a pass or fail. If you have a target CPA, CAC, cost per lead, or cost per qualified lead, bring that threshold into the kickoff.

• A realistic timeline. Short branded-search tests can work, but you still need enough time to observe signal. The kickoff should cover when the test could launch, what else is changing in the business during that window, and how long the team is comfortable holding the treatment in market before reading results.

Most kickoff calls end with a recommendation on the right test structure, the data needed to size it, and the success criteria that will be used to interpret the result. Getting those inputs aligned early makes it much easier to move from "we should test this" to an actual launch plan.

When a geo holdout is possible in theory but risky in practice

Sometimes the right long-term answer is still a geo holdout, but the first practical step is a user-level holdout inside the ad platform. This usually comes up when your brand campaign is managed globally, when splitting markets would require disruptive campaign restructuring, or when the account is in an unusual state after a major change and you want to avoid introducing extra instability during the test.

In that situation, a platform-run Conversion Lift Study can be a useful first read for Branded Search if your conversion volume is high enough and your tracked event matches cleanly in the platform. It reduces operational complexity and lets you learn without immediately reworking campaign structure. The tradeoff is that it remains a narrower measurement: it depends on the platform's observable users and usually works best for faster, tag-trackable outcomes rather than long-lag business metrics.

A good rule of thumb is: use the user-level holdout to get an early directional answer when geo execution would be unusually disruptive, then return to a geo test once the account is stable enough to support a cleaner market-level read. That sequence is often the best way to balance learning speed with methodological rigor.

When tracking or cookie coverage is in flux

Sometimes the blocker is not the test idea — it is the measurement setup. If your cookie coverage, conversion tagging, or platform match quality is actively changing, a user-level holdout inside the ad platform can be a reasonable temporary step for Branded Search, because it lets you learn without immediately restructuring markets. The tradeoff is that the result only reflects the users and conversion events the platform can actually observe.

A practical rule of thumb: if your tracked event is stable and you have enough observable conversion volume, a platform Conversion Lift Study can give you an early directional read while the broader measurement setup settles. But if cookie coverage is inconsistent, tags are still being fixed, or the event you really care about is mostly offline or delayed, the cleaner long-term answer is still a geo-based holdout once the measurement environment is stable again.

If you'd like to walk through your specific situation, your Growth Advisor can run through the decision with you and flag what would change depending on whether your brand faces active competitor bidding, operates in multiple markets, or has constraints that limit your test options.

When a branded search holdout should not be the first test in the quarter

A branded search holdout is often high-value, but it is not always the best first experiment in a given quarter. In practice, timing matters as much as methodology.

It can make sense to postpone the holdout when all three of these are true:

• You already know how you would act on the answer. If the business plans to keep buying as much branded demand as it can in the near term regardless of the test outcome, the learning may be real but not immediately decision-changing.

• The upcoming window is commercially sensitive. Peak seasonal periods can make a holdout harder to interpret and harder to operationalize if the downside risk feels too high.

• There are higher-priority questions elsewhere in the mix. If another large channel is more uncertain, less defended, or more likely to change near-term budget allocation, that test may deserve to go first.

Postponing a branded search test does not mean abandoning it. It usually means sequencing it for a cleaner window when the result can change a real budget decision.

What to test instead if brand is not the top Q1 priority

If the immediate goal is to learn where budget should move next, a different test may produce a more useful answer than a branded search holdout.

• Test channels where you suspect over-investment. If non-brand search or another performance channel looks expensive, a pulse-down or holdout can answer whether current spend is truly incremental.

• Test channels with scale-up potential. If paid social, Bing, or another channel appears underfunded relative to its likely headroom, a pulse-up can show whether more spend creates efficient incremental lift.

• Keep brand stable if that makes the rest of the quarter easier to read. A stable branded-search baseline can reduce interpretation noise while you test reallocation decisions elsewhere in the portfolio.

A practical prioritization rule: lead with the experiment that is most likely to change a real budget or planning decision in the next 90 days.

Questions to use when sequencing the roadmap

When deciding whether branded search belongs now or later, ask:

• If the result lands either way, would we actually change spend this quarter?

• Is the planned test window unusually seasonal or operationally risky?

• Which channel is most likely to improve our next budget allocation decision if we learn its true incremental impact now?

• Would holding brand steady make the rest of our experiment roadmap easier to interpret?

Those questions often make the sequencing choice clearer than debating branded search in the abstract.

When another channel test should come before a branded-search holdout

Sometimes the best next step is not to test Branded Search first. If the bigger planning question is whether an upper-funnel or demand-generation channel is creating more future branded demand, it can make sense to keep Branded Search stable first and test that other channel before you run the brand holdout.

A common example is deciding between a paid social pulse-up and a branded-search holdout. If the team wants to learn whether additional paid social spend increases branded search activity, holding Branded Search steady during the first test usually makes the later brand read easier to interpret. In practical terms, the sequence is often:

1. keep branded search stable while you test the upstream channel

2. review whether that channel appears to change branded demand or branded-search volume

3. run the branded-search holdout afterward if the business still needs a direct causal read on branded search itself

This is usually the better sequence when the brand result would not change an immediate budget decision, when brand is concentrated in a commercially sensitive period, or when a stable brand baseline makes other tests cleaner. The exact spacing between tests depends on the channel, the target metric, and how quickly effects show up in the business, so the safest next step is to size that timing with your Growth Advisor rather than assume a fixed gap.

Want to learn more about marketing mix modeling? Check out other Paramark marketing mix modeling articles here