Blog: Which Dollars Are Actually Working Across Your Marketing Mix? Shifting Marketing Budget Allocation From Gut Feel to Data-Backed

Most marketing budgets aren’t built. They’re inherited. Last year’s numbers, adjusted at the margins—a little more here, a little less there, based on which channels reported the best efficiency.

The problem: “working” and “working according to ad platforms” aren’t the same thing. Jordan Crawford has seen this play out across enough accounts to know that a channel can look efficient on every metric a team is tracking and still be the wrong place for the next dollar. Money keeps flowing where it’s always flowed because nobody’s stopped to ask.

Jordan has also seen what it takes to ask that question and why most teams don’t, even when they suspect the answer might not be what they expect. 

“A lot of [the CMOs we work with] have had 10 plus years of running budgets a certain way, with real results to point to.” The challenge is showing them a better way to see something they’ve been looking at for a decade before the gap between “reported as working” and “actually working” gets expensive.

The Real Obstacle to Better Budget Decisions Isn’t the Math

When Jordan thinks about what stands between a marketing team and a data-backed budget, the technical pieces (incrementality benchmarks, testing frameworks, marketing mix models) aren’t where she starts.

“I think there’s an education element to it,” Jordan says. “It’s a different way of looking at things.”

She’s watched media managers leave agencies entirely rather than accept that the platform reporting they’ve relied on for their entire career might be misleading. “A lot of marketers have come up in this digital marketing environment, and if you’re told, hey, no, the platform’s lying to you, it’s a big shift.”

That education gap runs both ways. The marketing team needs to understand why the old framework misses the full picture. And the CEO and CFO need to be brought into that same understanding because the budget conversation that follows won’t land otherwise.

The second obstacle is more concrete, but Jordan calls it “a close number two”: even with agreement on the new approach, the data infrastructure often isn’t ready. 

“Your CRM has to be set up in a certain way to be able to pull the data that you need,” Jordan says, and she’s seen this hit hard in industries like senior living or higher education, where CRM systems “are not built to easily pull data out and share with the marketing team.”

Education comes first, not because the technical work doesn’t matter, but because without it, the technical work has nowhere to land.

How to Audit Where Your Marketing Budget Is Actually Going

Once a team is ready to look at their budget differently, start with an audit. It breaks into three layers.

Settings

What are the channels actually configured to do? 

Performance Max is the clearest example. 

“Out of the box, if you turn on PMax, it will automatically serve both branded and non-branded terms,” Jordan says, and because PMax is built to find the cheapest path to its goal, it gravitates toward branded terms fast, since they’re usually the cheapest to win. 

Without a brand exclusion, a campaign meant to reach new prospects can quietly become “a branded search campaign that’s hiding as a new prospecting campaign.” 

Jordan’s number on how often this goes unaddressed: “I would say, like, 90% of the sales audits we do, people don’t have that on.”

Signals

What data is being fed back to the platform to define success? 

Jordan estimates that 80-90% of the time, accounts optimize toward leads and phone calls without distinguishing real leads from noise. 

“If you’re not also feeding back information about how many of those leads were spam or how many turned into appointments or applications, the platform has no way to learn what good looks like.” And bot traffic, which has “gotten even worse with AI,” gets treated the same as a real prospect.

Tracking Infrastructure

Is the platform seeing all of its conversions?

With up to 50% of Americans using ad blockers, a huge share of activity never reaches the ad platform through normal tracking. 

Server-side tracking — “a direct line between the ad platform and your server,” bypassing the browser — recovers some of that lost data, giving platforms a fuller picture to optimize against.

How to Calculate What Your Business Can Actually Afford to Pay for a Customer

Underneath all three audit layers sits one question that’s almost always missing: what can this business actually afford to pay for a customer?

“I find that often we either don’t have a clear goal going in, or there’s a goal that nobody can explain where it came from,” Jordan says.

Here’s a real example from a retail entertainment brand offering public events and private parties. They didn’t have a target CAC. When asked, the honest answer was “I have no idea.” 

So Jordan’s team worked from the business’s own economics. They pulled the cost of goods and services for each event type, calculated it per seat (since that’s how customers buy), and looked at revenue per seat, including the food and drinks customers buy once they’re there.

That gave a break-even number: the most the business could ever pay to acquire a customer before losing money was $48. Apply the business’s target margin, roughly 50%, and the target CAC drops to $24.

This number becomes the anchor for everything that follows. Every channel, every campaign, every reallocation gets measured against it.

Why Ad Platform Numbers and Real Performance Numbers Almost Never Match And How to Find the Difference

Once you have a real target CAC, the next step is understanding what each channel is actually contributing.

This is where platform reporting breaks down most visibly. Ad platforms measure success by tracking clicks and attributing conversions to them. But not every conversion a platform takes credit for was actually caused by that channel. Some of those customers would have converted anyway. When a platform counts those conversions, it’s inflating its own results.

To find the true number, Jordan’s team uses incrementality testing: a structured way to measure what would have happened without a given channel, and compare it to what actually happened with it. The gap between those two outcomes is the channel’s real contribution.

The result of that testing is an incrementality multiplier—a ratio that adjusts what the platform reports down to what the channel actually drove. A multiplier of 20% means only 20% of the conversions a platform reported were genuinely caused by it; the other 80% would have happened regardless.

Here’s what that looks like in practice

A contribution report showed Google branded search drove 1,280 leads. Testing revealed the real number was 256—about 20% of what was reported. The other 1,024 leads, in Jordan’s words, “would have come in organically and converted anyway.”

Channel	Platform-Reported CPL	Incrementality-Adjusted CPL	Target CPL
Branded Search	$14	$72	$120
Performance Max	$18	$89	$120
Meta Prospecting	$38	$44	$120
YouTube	$52	$57	$120

Once CPL is recalculated against the true incremental number, the budget math often flips. A channel that looked efficient—and like an obvious place to add spend—can become the most expensive channel in the mix when measured against what it actually drove.

Incrementality multipliers come from two sources

Silverback maintains a database of thousands of incrementality tests—its own and from published industry research—and uses category averages as a starting baseline. Branded search, on average, comes in around 20% incremental.

But averages can mislead just as easily as platform numbers. Jordan has often seen branded search incrementality closer to 5%. For one client in a competitive DC-area market, however, testing showed branded search was 100% incremental—when they pulled back spend, they lost revenue that organic didn’t recover. 

“They lost a lot of money, and organic didn’t make it up. They needed that branded search budget.”

Every business gets its own number, built from its own test.

How Incrementality Testing Actually Works

Three approaches, ordered roughly by how much data you need.

1. Matched Market Test

The most rigorous. Using historical data, Silverback’s team finds pairs of markets that have historically moved together. For example, when Charlottesville, Virginia revenue goes up, Dayton, Ohio revenue also goes up.

Once enough pairs are identified, one group becomes a test, the other control. A change happens only in the test group — launching a new channel, scaling spend up or down, turning a channel off — while the control stays unchanged and nothing else shifts during the window. 

Attribution gets set aside completely. The only thing measured: how revenue moved differently between the two groups.

2. Directional Lift Test

Used when there’s not enough data for full statistical significance. For one Silverback client, Jordan’s team cut branded search paid spend in half and watched organic clicks. 

If organic rose by roughly the amount paid fell, that’s a strong signal, even without a full test, that the paid spend was largely redundant with organic traffic that would’ve happened regardless.

3. Media Mix Modeling (MMM)

A statistical model built from roughly two years of historical spend across all channels, including offline, and the business’s revenue. These models come from approaches open-sourced by companies like Uber, Google, and Facebook after years of internal data science investment. 

The output: a prediction of how revenue responds if spend in a given channel changes. MMMs work especially well alongside testing. Results can feed back into the model to improve it over time.

Learn More About Silverback’s Media Mix Modeling Approach

From Insight to Reallocation: What Happens After the Tests

The retail entertainment client example and the branded search multiplier example point to the same kind of decision: once a business knows its real target CAC and its real incremental cost per channel, budget moves stop being arguments about gut feel and start being simple math.

That doesn’t make the conversation easy. A CMO who’s spent ten years optimizing toward a platform’s reported ROAS needs real time to trust a multiplier-adjusted number instead. 

But once that trust is there, reallocation follows the numbers: cut what’s expensive once incrementality is accounted for, fund what testing shows is working, and keep testing as the business and competitors change.

This shift is about giving intuition something real to stand on: a target CAC grounded in the business’s own economics and multipliers grounded in what actually happened when the business tested its assumptions.