How to Use Marketing Mix Modeling to Plan Your Marketing Budget
A practical guide to running a marketing mix model, from the business questions and data it needs to reading diminishing returns and planning the next budget.
Most marketing budgets are set the same way: last year's plan, adjusted for whatever looked good in each platform's dashboard. The line usually attributed to the retailer John Wanamaker still describes the result: "Half of the money I spend on advertising is wasted; the trouble is I don't know which half."
Marketing mix modeling (MMM) is one of the oldest and most reliable answers to that problem. It uses your own history of sales, spend and outside factors to estimate what each channel actually contributes, then lets you plan the next budget from that evidence. If you're new to the method itself, start with our guide to what marketing mix modeling is. This guide is about putting it to work.
Start with the questions, not the model
An MMM project goes wrong most often when it starts with data and hopes the answers will appear. It goes right when it starts with the decisions you need to make. Nearly every budget question comes back to two:
- Should total media spend go up or down? Is the next dollar of marketing likely to return more than it costs?
- Is the money in the right places? Is the current split across channels balanced, or should some channels get more and others less?
Agreeing on these questions up front, with marketing and finance in the room, decides which data you need, how detailed the channel breakdown should be and what "success" looks like when the results come back.
Note that marketing mix modeling is not the same thing as the "marketing mix" of the four Ps: product, price, place and promotion. The four Ps describe the choices a business makes. MMM measures what those choices did, and a good model includes price and promotions alongside media because they move sales too.
What data the model needs
A marketing mix model learns from variation over time, so it needs history and it needs consistency.
- Sales or another outcome. Weekly or daily sales, revenue, leads or sign-ups. Two to three years is typical, enough to see seasonal patterns more than once.
- Media by channel. Spend, and ideally impressions, for every channel: TV and connected TV, search, social, display, audio, direct mail, out-of-home. Online and offline belong in the same model.
- Outside factors. Pricing, discounts and promotions, seasonality, holidays, weather, paydays, economic conditions and one-off shocks such as the pandemic. Leaving these out tempts the model to credit marketing for sales it didn't cause.
- Brand measures, where available. Larger brands often add brand tracking data, such as ad awareness, consideration and preference, to capture longer-term effects.
For retailers and consumer brands, in-store promotions deserve special attention. They can create a large spike in a short period, and a model that combines promotion data with point-of-sale data can separate that spike from the media running at the same time.
Gathering all of this is usually the hardest part. Spend, impressions and results sit with different teams, agencies and platforms, each reporting in its own format. Budget real time for collecting the data, then check it visually before modeling. A missing week or a channel that changes names halfway through can distort the results.
How the model turns history into answers
Once the data is clean, a regression estimates how sales respond to each input while holding the others constant. The output splits your sales into a base (what you would sell with no advertising) and the incremental contribution of each channel. From there you get a return on investment for each channel, measured on the same basis.
Two features make the answers realistic:
- Carryover (adstock). Advertising keeps working after it runs. A TV flight or a podcast campaign can lift sales for weeks, and the model credits those later sales back to the campaign.
- Diminishing returns. Every channel follows a response curve, often S-shaped: early spend builds momentum, the middle of the curve is efficient and eventually each extra dollar does less. Finding where each channel sits on its curve is what turns the model into a budget tool.
Turning results into a budget
The real value of MMM is not the report on last year. It is the ability to test next year's plan before you spend it.
Find the tipping points. Each channel has its own point where more spend produces little extra growth. A channel with a high average return can already be past that point, while a smaller channel may have room to grow. Average return tells you what a channel did. Marginal return tells you what the next dollar will do, and that's the number budgets should follow.
Run scenarios. With the response curves in hand, you can ask: what happens to sales if TV rises 10 percent? What if part of the search budget moves into connected TV? What is the best split if the total budget is cut? The model answers each question in terms of expected sales, not clicks or platform conversions.
Set guardrails. Optimizers will happily move everything into one channel if you let them. Setting a minimum and maximum spend for each channel keeps recommendations practical, protects channels that serve other goals and stops the model from pushing spend far beyond the levels it has actually seen.
Balance the short and long term. Some channels close sales this week. Others build the demand that later shows up as search and direct traffic. A model that includes both, and accounts for carryover, keeps you from cutting the channels that feed the rest.
Keep the model running
Traditional MMM was a once-a-year project, often run by a large consulting team at significant cost. The model would recommend a split, the agencies would execute, and nobody would know whether it worked until the next model a year later.
That cadence no longer fits how marketing works. Channels, prices, competitors and customer behavior change within a year, and so do your campaigns. A model refreshed weekly or monthly shows whether a budget shift is working while there's still time to change course, and gets more accurate as each new period of data arrives. Testing a change, such as adding a new channel to the model or splitting one channel into its parts, becomes routine rather than a new project.
The model's output still needs judgment. Statistics can tell you that a channel is past its efficient range. Someone who knows the business has to decide whether a new product launch, a competitor's move or a change in creative explains it. Our post on the complexities and pitfalls of marketing mix modeling covers where models go wrong and how to check them.
Why MMM fits privacy-first marketing
Because MMM works from aggregate data, it needs no cookies, device IDs or personal information. That matters more every year. Apple's App Tracking Transparency and browser tracking protections have cut much of the user-level data that click-based attribution relies on, while MMM's inputs are unaffected. It also measures the channels attribution struggles with, such as TV, audio, print and billboards, on the same basis as digital.
Attribution and MMM can still work together. Attribution shows detailed paths within the channels it can see. MMM shows what each channel adds to total sales. Our comparison of multi-touch attribution and marketing mix modeling covers where each fits.
The point of the exercise
Marketing mix modeling won't tell you which individual ad someone saw before buying. It will tell you how much your sales depend on each channel, where each one stops paying off and what happens to revenue if you move money around. Used that way, and refreshed often, it gives marketing and finance a shared, defensible basis for every budget decision.
Frequently asked questions
What data do I need for a marketing mix model?
At minimum, weekly or daily sales (or leads) and spend or impressions for every channel, ideally covering two to three years. You also need the non-marketing factors that move demand, such as pricing, promotions, seasonality, weather and economic conditions. Larger brands often add brand tracking data like awareness and consideration.
What business questions can marketing mix modeling answer?
The two core questions are whether total marketing spend should rise or fall, and whether the split between channels should change. A good model also shows where each channel hits diminishing returns and lets you test "what if" budget scenarios before committing money.
Is marketing mix modeling only for large brands?
It used to be, because traditional projects relied on consulting teams, took months and were expensive. Automated, regularly refreshed models have made it practical for mid-sized brands, as long as they have enough history and spend variation across channels.
How often should a marketing mix model be updated?
Classic models were rebuilt once a year, which is too slow to guide decisions that change weekly. Refreshing the model weekly or monthly lets you see whether a budget shift is working while there is still time to adjust it.
Can marketing mix modeling measure offline channels?
Yes. Because it works from aggregate spend and sales data, MMM measures TV, radio, print, direct mail, billboards and in-store promotions on the same basis as digital channels, without needing clicks or user-level tracking.