The two objections you'll hear most often when proposing media mix modeling are "we already have attribution" and "how do we know the model is right?" Both are legitimate. Both have good answers, but only if you've done the work before the meeting.
Getting MMM funded is less about explaining the methodology and more about framing the problem it solves in terms a CFO already cares about.
The cost of bad media decisions dwarfs the cost of building a proper MMM
Start With the Measurement Problem, Not the Solution
The worst way to propose MMM is to lead with the methodology. "We should invest in Bayesian media mix modeling to improve channel contribution visibility" will immediately put a CFO's eyes into the back of their head.
The right opening is a problem they already have.
"We're spending $3 million a year on advertising. Our platform attribution tells us Facebook ROAS is 4.2x. We ran a geo experiment on Facebook in March and it showed incremental ROAS of 1.4x. We've been making budget decisions based on a number that may be wrong by a factor of three. We need a better way to know where our media budget actually works."
Now you have their attention. The problem isn't "we need better measurement methodology." The problem is "we might be misallocating millions of dollars."
The Financial Framing
Budget misallocation has a direct cost, and you can estimate it.
If you're spending $3M/year on advertising and your allocation is 15% suboptimal, a conservative estimate given how common attribution distortions are, that's $450,000 in spend going to channels that aren't generating the returns they appear to. A good MMM engagement that improves allocation by even 10% recovers $300,000 in value annually.
Vendor MMM for a company at that spend level typically costs $80,000 to $200,000 per year depending on the provider and scope. At $150,000, the break-even improvement in allocation efficiency is about 5%. That's a fairly easy bar to clear if your current measurement is as unreliable as the data suggests.
Structure the financial case this way:
- Current allocation risk: estimate the cost of misallocation at your spend level (assume 10–20% inefficiency as a baseline)
- MMM investment: what the model will cost (vendor fees, or staff time for open-source)
- Required improvement to break even: the allocation improvement needed to justify the cost (usually 3–7% for mid-sized advertisers)
- Historical evidence: the measurement gap you've already observed (your incrementality test results, the sum of platform attribution vs. actual revenue, any channels that look suspicious)
Present it as a return-on-measurement calculation, not a methodology pitch.
Addressing "We Already Have Attribution"
This objection is almost always raised, and it's worth addressing directly.
Attribution tells you what happened. It doesn't tell you what caused it. The distinction matters because advertising platforms have a structural incentive to claim credit generously, they report on their own performance, within their own data, using their own methodology. An incrementality test you ran showed Facebook's contribution was 3x lower than attribution suggested. If your attribution were reliable, that gap wouldn't exist.
MMM is not a replacement for attribution. It's the calibration layer that tells you how much to trust your attribution numbers and for which decisions.
The phrasing that tends to land well: "Attribution is like asking each channel to report its own performance. MMM is like an independent audit. Both have value, but you wouldn't rely on a company's self-reported financials without an audit, and we shouldn't rely on platform self-reported attribution without independent calibration."
Addressing "How Do We Know the Model Is Right?"
This is the better objection, because it's the right question.
The answer is validation. Before you fund a full MMM engagement, you run one or two geo experiments. Then you compare the incrementality data from those experiments to what the MMM says those channels contributed. If they're directionally aligned, if the MMM says Facebook contributed 18% of revenue and your geo experiment suggests Facebook drives roughly 15–20% incremental lift, you have independent validation that the model is in the right territory.
If the MMM and incrementality tests diverge significantly, you investigate the discrepancy. Either the model has a specification problem, or the geo experiment was poorly designed. In either case, you now know you have a problem and can fix it.
The point is not that MMM will be perfectly accurate. The point is that a well-validated MMM will be more accurate than attribution data alone, and will give you a principled way to make cross-channel budget decisions that attribution cannot.
The Staged Approach (Easier to Get Approved)
A common barrier to MMM approval is that it feels like a large, uncertain investment with benefits that materialize over months. A staged proposal is easier to fund and easier to defend.
Q1: Run two incrementality tests on your top two channels. This costs roughly $20,000–$40,000 in opportunity cost (the revenue you forgo from holding out part of your audience), plus staff time. Use these results to quantify the measurement gap between attribution and actual incremental performance.
Q2: Present the measurement gap findings. Use the incrementality data to show leadership how far off your attribution is on at least two channels. This builds organizational readiness for the MMM finding and creates demand for better measurement.
Q3: Commission an exploratory MMM. With the gap evidence in hand, propose a one-year MMM engagement. Frame it as building on the incrementality validation work, not as a standalone project. Request that the vendor validate their model against the geo experiment results as part of the engagement.
Year 2: Full measurement program. After one year of MMM with validation against incrementality data, you have evidence of model reliability and initial ROI from allocation improvements. Use this to fund a sustained program.
Each stage justifies the next. You're not asking for $200,000 upfront; you're asking for $40,000 in Q1 to gather evidence that justifies the larger investment.
What to Include in the Proposal
A one-page internal proposal that gets approved typically includes:
- The measurement problem in financial terms (the cost of misallocation)
- The specific evidence you already have (incrementality test results, attribution vs. actual revenue discrepancy)
- The proposed solution (MMM type, vendor options with price ranges)
- The break-even analysis (what improvement in allocation efficiency justifies the cost)
- The validation plan (how you'll know the model is right)
- The staged approach (if appropriate for your organizational context)
Keep it short. CFOs are not interested in how Bayesian inference works. They're interested in whether spending $150,000 on a measurement upgrade will recover more than $150,000 in misallocated media budget. Answer that question clearly and the methodology becomes secondary.