Why Most MMM RFPs Disappoint (and How to Run One That Doesn't)
By
Brian Cusick and Mike DeTrana
·
11 minute read
About the Authors
Mike DeTrana, OnTrack Analytics. Previously Marketing VP for a large auto-services retailer and Director of Analytics for a large consumer packaged goods manufacturer.
Brian Cusick, ScanmarQED. Previously in marketing analytics at PepsiCo and Kraft Foods.
A view from both sides of the table: Mike DeTrana ran MMM RFPs as a brand-side VP of Marketing, and Brian Cusick has answered them from the vendor side at ScanmarQED.
Ask any marketer how they picked their creative agency, and the answer is rarely "we read their capabilities deck." You wrote a brief, sent it to a shortlist, and watched them come back with actual scripts, storyboards, or concepts built against your business. The thinking is what gets judged, not the pitch. Marketing Mix Modeling almost never gets bought this way, and that is exactly why so many MMM requests for proposal (RFPs) fail to get brands what they actually need.
The cause is typically structural rather than a matter of bad vendors or careless brand teams. The standard MMM RFP asks providers to price and commit to a scope before anyone has reviewed the data and without a full understanding of the business needs. That single sequencing choice creates three predictable problems: an incentive structure that rewards over-asking and over-promising, a brief shaped by assumptions the brand and the vendor each make about the other, and a scope based on data that turns out to be materially different from what was described. The fix is to change what you ask for and when: start with a short RFI to get to know firms, then run paid proofs of concept to evaluate vendors.
Marketing Mix Modeling (MMM), also called media mix modeling, is a statistical method that uses historical spend and outcome data to estimate how much each marketing channel contributes to sales. In practice it is less an analysis than a process: the model output feeds budget allocation and forecasting decisions across the year. That distinction is where so many RFPs go wrong.
Problem 1
The RFP incentive structure rewards the wrong behavior
An MMM RFP sets up opposing incentives that quietly work against a good outcome. The brand team's incentive is to ask for everything it can think of, not everything it needs. That can materialize as modeling scope items such as every channel, creative, base driver, and KPI, but it can also show up as requiring every data science technique that ChatGPT can think of, all "just in case" it matters later. The provider's incentive is the mirror image: to win the deal by committing to all of it, often with the uncomfortable feeling that none of what they are proposing is likely to be beneficial, and sometimes with no real knowledge of whether the commitment is feasible given the data that may or may not be available once the engagement starts.
Brand team
Over-ask everything — channels, KPIs, techniques — just in case it matters later
Vendor
Over-promise to win the deal, often without knowing if the data will even exist
I felt this firsthand running an MMM RFP as a VP of Marketing. The proposals we got back read like product demos, heavy on what each vendor's platform could do and light on what they understood about the business problems we actually needed solved. We picked a vendor, and months in, watched them blow past the turnaround they had committed to, once data collection turned out to be a far bigger lift than anyone had scoped going in.
Neither side is acting in bad faith. The format itself rewards over-asking on one side and over-promising on the other. The brand usually discovers the gap between what was promised and what was possible only after the contract is signed, the most expensive moment to find out.
The brand assumes it knows what it needs, and the vendor assumes it knows the business
Most MMM RFPs rest on two shaky assumptions that are flip sides of the same coin. Brand teams assume they know what to ask for, and vendors assume they understand the client's business. Rarely, if ever, are both true during the RFP process, and unfortunately not even by the point the initial statement of work (SOW) is signed. Even an experienced internal team benefits from an outside opinion on the best approach, and providing that opinion is a large part of what a specialist provider is for.
Brand team assumes
""We know what to ask for.""
Treating the approach as already settled before the RFP goes out, so the proposal becomes a platform demo instead of real engagement with the decisions the brand actually needs to make.
Vendor assumes
""We understand your business.""
RFPs are usually run by very experienced senior leaders who lean on past work and project their own preconceived notions onto the client, so the right questions about that specific business never get asked.
In our experience, close to 90% of brand teams make at least one costly scoping mistake by treating the approach as settled before the RFP goes out. On the vendor side, RFPs are typically led by very experienced senior leaders, so it is easy to lean on past work and project preconceived notions onto a prospective client, which means the right questions never get asked.
90% of brand teams make at least one costly scoping mistake by treating the approach as settled before the RFP goes out.
The deeper issue is what the brief tends to focus on. Model type, whether Bayesian, frequentist, hierarchical, or something else, is a small part of the MMM puzzle. MMM is not a one-off analysis; it is a process that feeds budgeting and forecasting throughout the year. An RFP that specifies the model in detail but says little about how results will drive decisions has optimized the smallest part of the problem while leaving the largest part unaddressed.
None of this is a failing of brand teams; it is the format working as designed. An RFP built around technical scope and price naturally pulls vendors toward showing off their platform rather than showing they understand your business, so the proposals become demonstrations of each vendor's tool rather than engagement with the decisions the brand needs to make.
Marketers are the experts on those decisions: the media plans, creative calendars, and channel shifts that actually happened. What most of them are not is modelers. The RFI process described below is built around that division of labor:
"Ask brand teams for the decisions and context they know best — and leave the model specification to the people who build models for a living."
Problem 3
The data is almost never what the brief describes
Scoping an MMM engagement is real work, and the RFP process usually skips it. Because no one has reviewed the data yet, most providers responding to an RFP take their best guess at scope, timeline, and price, once again falling back on past experience and assuming the new engagement will be similar. The single input that guess depends on most, the data, is rarely described accurately.
This is not a criticism of brand teams. In the majority of cases the data actually available for MMM is materially different from what the brief describes: channels are missing, spend is recorded differently across markets, history is shorter than expected, granularity is coarser than assumed. Reviewing data properly takes both time and expertise, and that work is normally done by the vendor, not the brand, before modeling begins. Asking providers to commit to scope and price before that review has happened all but guarantees that at least one of the two numbers is wrong.
Our RFP never even addressed model type or other technical details, so it was not the presence of technical detail that caused problems, it was the absence of a data review. The vendor we selected committed to a turnaround timeline without having seen how messy our data collection actually was, and once the engagement started, that timeline slipped well past what was promised. No amount of RFP wording would have caught that. Only looking at the data first would have.
A better MMM RFP starts with a short RFI, not a specification
The most effective MMM selection process begins with a short request for information (RFI) that deliberately leaves out the two things brands usually lead with: cost and modeling specifics. Do not ask for pricing yet, and do not request MMM details like model type. Both depend on data no one has reviewed, so asking for them now only invites the over-promising described above.
Use the RFI to do two things instead: tell vendors about your business, and ask them how they would think about it. Not which modeling techniques they would use, and not whether they can replicate what you have done before, but this: if they were starting from a blank slate, how would they approach MMM, how would they want the output to be used, and, importantly, why did they arrive at those recommendations.
First, state how you want the work run. Be explicit about the operating model you are looking for:
Then share the context a vendor actually needs
- structure
- How your business is organized: marketing budgets and P&Ls across brands, product lines, and geographic hierarchies
- planning_owner
- Who decides on resource allocation, who executes, and the level of aggregation at which marketing decisions are made
- mmm_influence
- Which planning decisions you want MMM to influence
- needs
- Your business needs, stated as needs, not a desired MMM type, structure, or method
Make it concrete: share the decisions, not the model
The most useful thing you can hand a vendor is a real decision you have wrestled with. For example:
-
“Last year we had to cut budgets under operational constraints, and we genuinely didn’t know whether to haircut all spend equally, sacrifice one brand to support the others, or cut brand media to protect conversion spend.”
-
“We’re launching in three new markets next year. Should we fund them with incremental budget, or reallocate from established brands?”
-
“Our largest retail partner is pushing for more trade spend, and we can’t see what we would give up in working media to fund it, or what it costs us over the next two years.”
How a vendor responds to questions like these reveals far more than any methodology overview.
Ask each vendor for a high-level take, not a technical plan: the data they would want, and their view of what MMM should do for your business, what makes sense, and what matters most. Frame it as a discussion, and cap the reply at 20 slides or fewer so you are comparing thinking, not proposal-writing stamina.
This is the pitch stage, not the delivery stage, the same as a creative review. You are evaluating how a firm thinks about your business, not asking them to hand over finished work.
Plan for at least two calls with each vendor, not a one-way exchange of documents. Use the first call for the brand to explain its situation and answer the vendor's questions, and the second for the vendor to present their RFI response. Read that response before the second call. At the very least, let AI summarize the submission for you, since this is 2026. A brand team that is not willing to spend real time in these meetings is quietly setting itself up for long-term disappointment.
Finally, ask for a broad annual budget range, plus or minus 50%. This is the budget for the work the vendor would propose, not a quote for your exact request, and not a priced list of every scenario. It is a fit filter: if your budget is $100k and a vendor comes back at $1M, you can rule them out early, before either side has spent real time.
Then run paid proofs of concept to evaluate finalists
An RFI tells you how vendors think. A paid proof of concept (POC) tells you how they perform on your data, and it is the only way to fully vet an MMM partner. After the RFI, narrow to a short list of finalists, usually two or three, and run each through the same paid exercise:
- Sign NDAs and give all finalists the same data
- Pay each firm the same amount, which could be as much as 10% of the expected year-one contract (your expected cost, not the vendor's estimate)
- Ask each to deliver the same work: a data review, a draft model results session, budget optimizations, and a presentation of results
Paying for the POC is what makes it honest. It compensates the real work of reviewing data and building a first model, and it lets you watch each firm do the actual job rather than describe it. Running every finalist on identical data keeps the comparison clean: differences in the output reflect differences in the firms, not differences in what they were handed.
This step costs real time and money, which is the reason to reserve it for a genuine multi-year commitment rather than a one-off tactical question. For the decision it is meant to support, it is the cheapest insurance a brand can buy.
Why the extra time and money is worth it
An MMM program should last far longer than a single contract cycle. A well-built MMM process should run for 5+ years. The models will change and evolve as the business evolves, and the people involved will change on both the vendor and client side, but the process itself should hold. If a brand is switching MMM vendors and redesigning its measurement process every two to three years, something is wrong upstream, usually in how the partner was chosen.
The right comparison is to financial planning. Ask your CFO how often the financial planning process gets rebuilt from scratch; the answer is that it does not. Every company has its own characteristics, but the fundamentals are fundamental (especially for public companies, and still true for large private ones), and the core of the process does not get redesigned every couple of years. MMM deserves the same fortitude. The data will change and the statistical techniques will change, but the process should be built to last. Spending a little more up front to choose the right partner is what makes that possible.
The typical MMM RFP vs. an RFI-then-POC process
| Dimension | Typical MMM RFP | RFI, then paid POC |
|---|---|---|
|
Opening ask
|
Detailed scope, model type, and price | Business context and the vendor's high-level take |
|
What you learn first
|
Who writes the best proposal | How each vendor thinks about your business |
|
Cost and methodology
|
Committed before any data review | Set after finalists work with your real data |
|
Vendor comparison
|
Proposals built on different assumptions | Same data, same paid deliverables, side by side |
|
Time horizon in mind
|
A single contract | A 5+ year measurement process |
Key takeaways
Most MMM RFPs fail because they ask for price and model specifics before anyone has reviewed the data or understood the business needs.
Start with a short RFI that shares your business context and a real decision you've wrestled with, and asks vendors how they'd approach MMM from a blank slate — no costs, no model specs, capped or fewer at 20 slides — plus a broad budget range (±50%) as a fit filter.
Choose between finalists with paid proofs of concept: two or three firms, the same data, the same fee (As much as 10% of your expected year-one cost), and the same deliverables, run side by side.
Choose for the long term. A good MMM process should last 5+ years, like a financial planning process, not get rebuilt every two to three.
Frequently Asked Questions
Business context and a request for the vendor's thinking, not a technical brief. Share how your business is organized (budgets and P&Ls across brands, product lines, and geographies), how you plan and who makes resource-allocation decisions, and which planning decisions you want MMM to influence. It helps to include a real decision you have struggled with, so vendors respond to your situation rather than a generic one. State the operating model you want: outsourced, in-house tools only, or hybrid. Then ask for a high-level take on the data and approach, capped at 20 slides or fewer. Leave out cost and model type.
Not up front. Ask only for a broad annual budget range, plus or minus 50%, as a fit filter: if your budget is $100k and a vendor proposes $1M, you can rule them out early. Firm pricing should come later, after finalists have reviewed your actual data in a paid proof of concept. A price quoted before the data review is a guess.
A paid POC is a short exercise in which two or three finalists work with your real data and deliver a data review, a draft model, budget optimizations, and a results presentation. Paying each firm the same amount, which could be as much as 10% of your expected year-one cost, compensates real work and lets you compare how firms actually perform rather than how well they write proposals. It is the only way to fully vet an MMM partner before committing.
Usually two or three finalists. Put them all under NDA and give each the same data, so the only variable in the comparison is the firm itself. More than three multiplies cost and effort without sharpening the decision; fewer than two removes the comparison that makes the exercise worthwhile.
Rarely. A good MMM process should last 5+ years. Models and data science techniques will evolve, and people will change on both sides, but redesigning the whole process every two to three years usually signals a selection problem, not a vendor problem. As with financial planning, the process should have fortitude even as the details change.
About OnTrack Analytics
OnTrack Analytics both supports existing MMM programs and stands new ones up from scratch, using platforms like ScanmarQED to keep delivery software-driven rather than rebuilt from zero for every client. Founder Mike DeTrana started his career in analytics, then moved client-side as VP of Marketing at a $1B retail brand before founding OnTrack. That path, analyst first and marketer second, gives OnTrack a rare vantage point: fluency in the modeling itself paired with firsthand experience sitting in the seat that has to act on it. Whether a brand already runs an MMM program and needs a sharper, more actionable version of it, or is standing one up for the first time, OnTrack brings both the technical rigor and the commercial judgment to make MMM outputs actionable.
About ScanmarQED
ScanmarQED is a B2B software and analytics company specializing in marketing measurement, with a focus on Marketing Mix Modeling (MMM). ScanmarQED helps enterprise marketing teams understand which channels and campaigns drive business outcomes, optimize budget allocation, and build the business case for marketing investment, pairing measurement software with the expertise to make the results actionable.