Most marketers we speak to already see the value in measuring the whole marketing mix. The hesitation usually comes one step earlier, with the question of whether the right data is in place to get started.
Marketing mix modelling (MMM) estimates channel contribution by combining three things, your marketing activity, your business outcomes and the external factors that shape demand.
Line those three inputs up over a consistent period and a model can separate what each channel adds from everything else going on around it.
This guide walks through each MMM data requirement, how the data should be structured and what makes it suitable for statistical analysis. Understanding the requirements is the first step towards assessing readiness, and you may find you are closer than you think.
Here’s a breakdown of what data is required:
- Historical marketing spend
- Digital channel activity
- Offline marketing activity
- Consistent IDs, where possible
- Clear, consistent naming conventions
- Business outcomes
- External and seasonal variables
- Consistent time peroids
💡 Pro Tip
Ruler brings your marketing performance data together in one place, then applies statistical modelling to show what each channel contributes to revenue. Our MMM measures digital and offline activity side by side, and our budget scenario planner lets you test spend decisions before you commit to them. If you would like to see how your own data would work, book a demo and we will talk through your measurement requirements and how Ruler’s MMM capabilities can fit them.
What data do you need for marketing mix modelling?
The data needed for marketing mix modelling falls into eight categories, and each one plays a different role, and together they give a model enough to work with.
The precise requirements depend on your business, the channels you use, the outcomes you want to measure and the modelling approach.
A retailer with a handful of paid channels will not need the same inputs as a services business running events, print and broadcast alongside digital. The structure is the same though, and that’s what this guide covers.
There’s a good reason to get this right. According to our own research, 44% of marketers say cross-channel journeys, both online and offline, as a challenge for effective marketing attribution. MMM was built to handle exactly that problem, because it works at the level of channels and time periods rather than trying to follow every individual click.
Historical marketing spend
Spend is the foundation of every model. For each channel you want measured, you need the amount invested in each reporting period, whether that’s daily or weekly. Weekly is the most common choice, and it’s the one we see work best for most businesses.
A few things make spend data usable.
- Include every channel with a meaningful budget, not just the ones that are easy to export.
- Record actual spend rather than planned budget, since the two rarely match exactly.
- Include production, agency and sponsorship costs where they are part of the investment.
- Keep the currency consistent across the whole dataset.
We typically ask for at least two years of history, which gives 104 weekly data points.
Three years is better, because it lets the model see more than one full seasonal cycle and tell the difference between a genuine channel effect and a one-off peak. If you have less than that, you can still explore MMM, and we cover that further down.
Digital channel activity
Spend tells the model how much you invested. Activity data tells it what that investment delivered in terms of exposure.
For digital channels this usually means impressions, clicks, reach and frequency, plus the campaign or channel they belong to. Paid social, display, video and connected TV influence people who never click, so a model built only on clicks will undervalue them.
Including impression data is what allows an MMM to measure that upstream influence.
Pull this data from each advertising platform, such as Google Ads, Meta, LinkedIn and TikTok, and from your organic and email tools as well. Search volume for your brand terms is also worth adding, as it often shows the knock-on effect of other marketing.
This is also where reporting differences show up. Every ad platform reports conversions inside its own attribution window, and none of them account for what the others are claiming. Add them together and the total frequently exceeds your actual figure.
For MMM, that’s a minor point, because the model uses your own business outcomes rather than platform-reported conversions.
Offline marketing activity
Television, radio, print, out-of-home, direct mail, events and sponsorships all influence demand, and most of them leave little or no impact in digital analytics.
For each offline channel, collect spend by period along with whatever exposure measure exists.
That might be gross rating points for television and radio, circulation for print, number of sites for out-of-home or mailing volumes for direct mail. Where no exposure metric exists, spend by week is enough to start with.
Timing matters here. Record when activity actually ran, not when it was invoiced. A radio campaign that aired in March but was billed in April will confuse the model if the data says otherwise.
Offline activity also covers the conversions it generates. Customers phone your sales team, visit a showroom, reply to an email or speak to someone at an event.
Many of the highest-value sales happen this way, and when they aren’t connected back to marketing, digital campaigns look weaker than they really are. Only 48% of marketers say they are fairly confident in their ability to prove ROI, and untracked offline conversions are a big part of why.
Consistent IDs, where possible
Consistent IDs make it much easier to join datasets together. A campaign ID that appears in your ad platform, your analytics tool and your CRM means the same campaign can be matched across all three without anyone doing it by hand.
The IDs worth standardising are campaign IDs, channel IDs, region or market IDs and product or service line IDs. If you sell in several countries or run different product lines, these are what allow the model to be built at a more granular level.
We say “where possible” for a reason. Many businesses have historic data with no shared IDs at all, and that doesn’t stop an MMM from being built.
It just means more mapping work at the start. A simple lookup table that links each platform’s campaign names to a single channel category is often all that’s needed.
Clear, consistent naming conventions
Naming conventions are the least glamorous part of data preparation, and one of the most valuable.
If paid social appears as “Paid Social”, “paid_social” and “Social Paid” across three exports, the model sees three different channels unless someone cleans it up first. A workable convention does a few simple things.
- Uses one agreed label for each channel, applied everywhere.
- Follows a fixed pattern for campaign names, for example channel, objective, region and quarter.
- Applies the same UTM structure to every tracked link.
- Is written down and shared with agencies and freelancers.
In our experience, a half-day spent agreeing to the convention saves days of cleaning later. If your historic data is already inconsistent, don’t worry about it. A mapping table can bring older names into line, and the convention then protects everything you record from that point on.
Business outcomes
The outcome is what the model is trying to explain. Pick the measure that matters most to the business and record it for the same periods as your spend and activity data.
Common outcome measures include revenue, orders, qualified leads, sales opportunities, closed deals, sign-ups and footfall.
Revenue is the strongest choice where you have it, because it reflects the real value of what marketing produced. Lead volume alone can mislead, since it says nothing about quality.
This is relevant because 37% of respondents say their biggest marketing challenge is generating high-quality leads.
If your outcome measure is raw lead count, the model will optimise towards volume. If it’s pipeline value or closed revenue from your CRM, it will optimise towards the leads that actually turn into sales. We usually recommend using the CRM as the source for outcomes, as it holds the final figures rather than the early estimates.
If your sales cycle is long, record the outcome by the date it was generated or closed and be consistent about which one you choose. You can also model more than one outcome, for example leads and revenue, and compare how channels perform against each.
External and seasonal variables
Prices change, competitors launch campaigns, the economy shifts, the weather turns and public holidays come and go. A model that ignores these will wrongly give marketing the credit or the blame for movements that were really caused by something else.
External variables worth collecting include the following.
- Seasonality and calendar effects, such as holidays, paydays and school terms.
- Your own pricing, promotions and discounts.
- Competitor activity and market share where it’s available.
- Economic indicators, for example consumer confidence, inflation or interest rates.
- Weather, if demand in your sector responds to it.
- Product launches, stock shortages and site outages.
Ruler’s MMM accounts for seasonality, competitor activity, economic conditions and diminishing returns across more than 30 variables simultaneously.
You don’t need to supply all of them on day one. Start with the ones you know affect your business, such as promotions and seasonal peaks, and add further variables as the model develops.
Consistent time periods
Every dataset needs to use the same time period, and the same start and end dates. Weekly reporting is the most common standard, with each week starting on the same day. Daily data is useful where you have a lot of volume and want more detail, while monthly data gives the model less to work with.
The most frequent problem we see is mixed calendars. Ad platforms report by day, finance reports by calendar month, and the CRM reports by the date a deal closed. All of these need converting to the same weekly structure before modelling starts.
Other checks that help are listed here.
- Fill gaps with zeros where there was genuinely no activity, rather than leaving them blank.
- Make sure time zones match across platforms.
- Keep any changes to how a metric is defined clearly noted, with the date it changed.
- Cover the same date range for every variable.
Does your data need to be perfect before you start?
Data preparation doesn’t need to be finished before you explore MMM. Most businesses we work with begin with spend in spreadsheets, activity data spread across platforms and outcomes held in a CRM, and that’s a perfectly good starting point.
The quality of the inputs does affect the reliability of the outputs though, and it’s worth being clear about how. Missing spend for a channel means the model can’t measure it, so its effect gets absorbed into other channels or into the baseline.
Inconsistent naming splits one channel’s results across several labels. Too little history makes it harder to separate seasonality from channel effects. Gaps and mismatched periods add noise, which widens the range of uncertainty around each result.
The practical approach is to treat data preparation as part of the project rather than a hurdle before it. Gather what you have, list what’s missing and decide which gaps matter most. Spend and outcome data for your biggest channels is the priority, and everything else can be improved over time.
How Ruler supports MMM data preparation
Bringing marketing and revenue data together is where most teams spend their time, and it’s where Ruler is designed to help.
Ruler tracks calls, form submissions, live chat enquiries and other key conversions using a first-party JavaScript tag that follows the customer journey from first visit through to revenue.
As the tracking is first-party, it isn’t weakened by the privacy changes affecting third-party cookies. It gives you a reliable record of outcomes that connects marketing activity to real sales, including conversions that happen offline.
Ruler also ingests data from your advertising platforms, CRM and other business systems, so spend, activity and outcome data sit in one place instead of being pulled together by hand. Enriched attribution data is then pushed back out.
CRM records are updated with source, campaign and customer journey data, and ad platforms receive offline conversion and revenue signals that improve bidding and audience optimisation.
Once the data is in place, Ruler’s marketing mix modelling measures the impact of every channel, including digital, TV, radio, print and other offline activity, using both historical performance and forward-looking budget forecasts.
The budget scenario planner then uses diminishing return curves to show where each channel is approaching saturation and where extra investment is likely to generate incremental revenue. You can model efficiency, growth or custom budget scenarios before committing spend, so decisions rest on predicted business outcomes rather than assumptions.
Ready to explore your own data?
You need spend, activity, outcomes and a consistent timeline, and you can build from there. If you would like to see how your data would work in practice, book a demo with Ruler and we will go through your measurement requirements and how our MMM capabilities can support them.
Data required MMM FAQs
Marketing mix modelling requires historical spend by channel, activity data such as impressions and reach, offline activity such as TV, radio and print, a business outcome such as revenue or leads, external variables such as seasonality, pricing and competitor activity, and a consistent time period across all of it. Consistent IDs and naming conventions make the data easier to combine.
We recommend at least two years of weekly data, which gives 104 data points, and three years where you have it. That’s enough for the model to see seasonal patterns more than once and separate them from channel effects. With less history you can still start, although results will carry wider uncertainty.
No. MMM works with aggregated data by channel and time period, so it doesn’t need to follow individual customers. That makes it well suited to offline channels and to measurement that respects privacy. Customer-level data from first-party tracking can sit alongside it and give you a clearer view of outcomes.
Yes, you can run MMM with incomplete data, but the results reflect what’s missing. Channels without spend data can’t be measured individually, and short histories reduce confidence in the output. The best approach is to start with spend and outcomes for your largest channels, then fill gaps over time.
Start by listing every channel and where its spend and activity data lives. Convert everything to one weekly structure, agree one naming convention and map historic names to it, add your business outcome from your CRM, and record external factors such as promotions and seasonal peaks. Finally, check for gaps and note any changes in how a metric was defined.
