Opens in a new tab

What Data Do You Need for Marketing Mix Modelling?

Published-
November 11, 2026
IMS graphic for the article What Data Do You Need for Marketing Mix Modelling?, showing a data table

At minimum, a useful Marketing Mix Model needs consistent, regular data on sales and marketing spend by channel, covering at least a year and ideally two. It also needs the main non-marketing factors that move sales, with pricing, promotions and seasonality chief among them. Beyond that, more history and more detail generally mean a more confident model, but it’s rarely all or nothing. Here’s what matters, and what to do if your data isn’t perfect.

The core dataset

Every MMM project starts from roughly the same foundation:

  • Sales or revenue, at a consistent time grain
  • Marketing spend by channel, ideally split by campaign or activity type
  • Advertising activity metrics where available, such as impressions, GRPs or clicks
  • Pricing
  • Promotions and discounting
  • Distribution or availability, particularly for retail and FMCG
  • Seasonality, usually derived from the sales data itself instead of supplied separately

What time grain should the data be in?

Weekly data is the sweet spot for most businesses. It’s granular enough to separate the effects of different channels and campaigns, but not so granular that day-to-day noise swamps it. Daily data can help with specific questions, but it often needs aggregating back up before the model runs cleanly. Monthly data is usually too coarse. Squeezing a month of varied activity, a price change here and a promotion there, into one number blurs exactly the effects you’re trying to separate.

How much history do you need?

Twelve months is a realistic minimum, and twenty-four is considerably safer. The reason is simple: a model needs to see enough variation in spend, price and promotions to work out what’s driving what. A business that spends roughly the same amount on the same channels every week gives a model very little to learn from, however many months of data it has. Channels with irregular or lower spend usually need more history than the big, consistent ones before the estimate is confident.

The non-marketing data that’s easy to forget

Sales move for reasons that have nothing to do with marketing, and a model that ignores them will hand marketing the credit, or the blame, for changes it didn’t cause. Beyond pricing and promotions, gather what you can on distribution and stock availability (especially for physical retail), competitor activity where you can reasonably estimate or obtain it, broader economic indicators relevant to your category, and category-specific factors such as weather for weather-sensitive products or major events for travel and hospitality.

Not all of these will matter to every business, but rule them out on purpose instead of leaving them out by accident.

What if your data isn’t perfect?

Most businesses’ data isn’t, and that’s normal rather than disqualifying. Common gaps include a shorter history than you’d like, a channel that changed significantly partway through, or patchy competitor data. A good MMM approach, particularly a Bayesian one, can work with imperfect data by combining sensible outside knowledge with what your own numbers show. Bayesian Marketing Mix Modelling Explained shows how that works in practice. The honest answer is usually not “wait until the data is perfect” but “start with what you have, and be clear about where confidence is lower.”

A quick checklist before you start

  • At least 12 months of sales or revenue data, ideally 24
  • Marketing spend by channel, weekly if possible
  • Pricing and promotional history for the same period
  • Any known distribution, availability or stock changes
  • A note of any channels that started, stopped or changed significantly
  • Awareness of major external events, such as economic shifts or competitor launches, that might have affected sales

Frequently asked questions

Can we run an MMM with less than 12 months of data?

It’s possible, but the results carry more uncertainty and lean harder on outside assumptions than on your own data. It’s usually better to be upfront about that than to present early results as more certain than they are.

Does the data need to be perfectly clean before we start?

No. Most real-world datasets have gaps, inconsistencies or definition changes somewhere. What matters is knowing where those problems are so they can be accounted for, not assuming the data is flawless and finding out later that it wasn’t.

What if we don’t track some channels well, like out-of-home or sponsorship?

That’s common, and it’s one reason MMM exists. It can estimate the contribution of channels that are hard to track at individual level, as long as there’s a reasonable record of when and how much you spent. What Is Marketing Mix Modelling? explains how.

Who should pull this data together on our side?

Usually a mix of marketing, finance and whoever owns sales or commercial reporting, because no single team normally holds all of it. Agreeing who does what before the project starts is one of the easiest ways to avoid delays. For a wider view of what to check before committing to a partner, see How to Evaluate an MMM Vendor.

If you’re not sure whether your data is good enough to start, IMS can take a quick look and tell you honestly where you stand. Get in touch. You can also read about our data assessment service.

Background
Magda Pietrzykowska
Associate Analyst
I bring expertise in data analytics with a background in design engineering, venture capital, and data science. I enjoy presenting my work to clients and engaging with industry peers at conferences. My multi-disciplinary background and passion for data analytics provide unique insights into the latest trends and developments in data analytics.

Transform Your Data Into Your Most Valuable Asset
Connect with us and discover how our tailored analytics solutions can drive business success.

Contact IMS
Background