Marketing Mix Modeling (MMM)

Marketing Mix Modeling (MMM): how to allocate budget across channels in a scientific way

You’re investing €50,000 or more per month in marketing across Google Ads, Meta, email, influencers, SEO and offline channels such as TV, out-of-home or print.
The question is simple and uncomfortable: is this allocation truly optimal, or are you just repeating last year’s plan with a few “gut-feel” adjustments?

In a nutshell

Marketing Mix Modeling (MMM) is a statistical model that links spend across different channels, online and offline, to business outcomes: sales, leads, revenue and margins. It becomes really useful when the marketing budget exceeds €50,000 per month, the company is active on multiple channels (digital and traditional) and there is at least one year of historical data available. It does not replace digital attribution models, it completes them, because it allows you to include TV, connected TV, out-of-home, print, radio, PR, events and in-store activities in the picture. With tools like Robyn and Meridian you can build robust models, simulate multi-channel budget scenarios and understand not so much “which channel is the best”, but how to reallocate budget based on marginal ROI to protect investment and improve overall results.

The budget allocation problem everyone has

The problem is not theoretical: every month tens of thousands of euros are distributed across digital and offline channels. Often, however, the criteria are poorly structured.

In practice, the marketing budget is defined starting from historical spend, tweaking the channels that seem to perform better based on gut feeling and taking into account internal pressures such as the preferences of the CEO or country manager.
The result is a mix driven more by inertia, perceptions and internal politics than by quantitative analysis.

Several industry studies show that a significant share of marketing budget is wasted due to inefficient allocation: too much is invested in channels that are already over-saturated and too little in undervalued channels, including offline media.
For annual budgets of €500,000 and above, this can mean hundreds of thousands of euros potentially misallocated.

Marketing Mix Modeling was created to answer very concrete questions: where are we over-investing, where are we under-investing and what is the optimal mix between digital and offline channels to maximise return in the medium term?

What is Marketing Mix Modeling (the pragmatic version)

Marketing Mix Modeling (MMM), or media mix modeling, is a statistical model that explicitly connects marketing spend by channel to business outcomes. The question it tries to answer is simple:

If I increase investment in channel X by €1, how much do sales, qualified leads or margin increase in my business?

To answer that question, MMM analyses historical spend and performance data over a time horizon typically between 12 and 24 months.
It takes into account not only digital channels but also TV, connected TV, radio, out-of-home, print, catalogues, flyers, trade fairs and in-store activities.
In parallel, it integrates contextual variables such as seasonality, promotions, price changes, competitor actions and macroeconomic factors.

Unlike other approaches, MMM works on aggregated data and does not track individual users. It observes the system as a whole: how much we spent per channel over a given period, what results we generated and how external drivers behaved.

How it works in practice

For each week or day you collect four families of information: how much you invested in the different digital channels, how much you invested in TV, CTV, print, radio, out-of-home and events, what results you generated in terms of sales,
leads, revenue and margins, and which external variables had an impact (seasonality, promos, competitors, product availability).

An MMM model analyses thousands of these data points over time and estimates the incremental effect of individual channels, digital and offline.
This is how insights like these emerge:

every additional €10,000 on Google Search generates a certain number of incremental sales;
each €50,000 TV flight generates an uplift in brand searches and overall orders in the following weeks;
a billboard campaign in a specific area increases response to local campaigns on Google and Meta in the same area.

In addition, MMM makes explicit concepts that are often intuited but rarely measured:

saturation curves that show where a channel yields less and less as budget increases,
carryover effects of campaigns that continue to produce results over time, and
synergies between channels, for example between TV and branded searches or between out-of-home and digital performance.

MMM doesn’t replace attribution, it completes it

A recurring question is whether Marketing Mix Modeling essentially does the same job as multi-touch attribution models in Google Analytics or other platforms.

Digital attribution works at individual user or session level, reconstructing the path from impressions to clicks and conversion. It is very useful to optimise keywords, creatives, audiences and media automations. However, it has a structural limit: it only measures what is trackable, so it struggles to include TV, CTV, radio, print, out-of-home,
PR, events and all activities that do not generate direct clicks.

Marketing Mix Modeling, on the other hand, works at an aggregated level. It does not follow people but time series of spend and results. It relates the evolution of digital and offline budgets to sales, leads and margins, controlling for external drivers. It does not use cookies or individual identifiers and is, by nature, privacy-safe.

The ideal setup is not choosing between MMM and attribution, but integrating them.
Attribution guides day-to-day operational optimisation, while MMM supports strategic budget allocation decisions on a monthly or quarterly basis, including TV, CTV, out-of-home, print and other offline media.

How to set up a Marketing Mix Modeling project

At HT&T Consulting we approach MMM as a marketing analytics project that touches data, technology and strategy.
It is not an academic exercise but a tool to decide how to distribute significant budgets across digital and offline channels.

1. Clean, comparable, centralised data

Without solid data the model is weak, regardless of the technology chosen.
You need at least 12 months, ideally 24, of historical data with:

spend by channel (including media costs, creative production, agency fees where relevant);
business metrics: sales, qualified leads, revenue, margins (not just CTR, CPC, impressions);
contextual variables: seasonality, promos, launches, price changes, stock-outs, competitor activities, macro events.

In reality, many MMM projects get stuck here: data scattered everywhere, different definitions of conversion across platforms, manual Excel reports, offline disconnected from digital. Often the first concrete step is to build or consolidate a data warehouse (for example with BigQuery) and integrate data integration tools (such as Supermetrics), so you have a single coherent view.

2. Build the model: Robyn or Meridian

Today you don’t have to start from scratch. There are open frameworks and tools designed specifically for Marketing Mix Modeling.
Some of the main ones are:

Robyn, developed by Meta, is an open-source MMM package that uses machine learning techniques to estimate media channel efficiency, saturation curves and carryover effects. It is designed for those who want advanced models with a good level of automation and have skills in R and data science.

LightweightMMM, from Google, is a Bayesian library in Python for building light but robust marketing mix models, with a focus on computation speed and the ability to simulate budget allocation scenarios. It is still in use but has been superseded by Meridian.

Meridian, from Google, is the most recent evolution. It is an MMM framework based on Bayesian causal inference, designed to handle geo-level data and complex multi-channel scenarios, including TV, CTV and other offline media. Its strength is the ability to update models with new data and support advanced media planning simulations.

Alongside these open-source solutions there are specialised SaaS platforms with no-code interfaces and consulting support, which can be suitable if you prefer to outsource the technical part. The choice depends on the channel mix, number of markets, internal skills and desired level of autonomy. But Robyn and Meridian are both open source, and adopting them means you keep control and build knowledge together with the agency that supports you.

google meridiam marketing mix model

3. Interpret outputs and turn them into budget decisions

A good Marketing Mix Modeling setup delivers one central metric: the marginal ROI of each channel.
Not the historical average, but the return on the last euro invested. This is the metric that allows you to decide rigorously where it makes sense to cut budget and where it is worth increasing it, considering both digital and offline channels.

In practice, if the model shows that the last euro spent on TV yields less than the last euro spent on Meta or email marketing, the next step is to simulate scenarios where TV budget is reduced and the more efficient channels are increased.
Similarly, if part of the search spend is clearly beyond the saturation point, it makes sense to consider scaling it back in favour of channels with higher marginal ROI.

Model interpretation must always be linked to business reality: growth targets, seasonality, brand priorities, channel constraints and media contracts. MMM does not provide absolute truths, but a strong quantitative base on which to build scenarios and decisions shared by marketing, finance and management.

From the model to the media plan: how to reallocate budget

Once the MMM model has estimated the marginal ROI of the channels, the underlying idea is simple: move budget from channels with lower marginal ROI to those with higher marginal ROI until the return curves tend to align.

This principle applies both to digital campaigns (search, shopping, social, display, video) and to offline channels (TV, CTV, radio, out-of-home, print, events). In many cases, for example, part of the TV or display spend is beyond the saturation point, while email marketing, CRM, influencers or upper-funnel video still have room to grow while maintaining strong returns.

Operationally, reallocation turns into a few clear decisions: scale back over-invested channels, free up budget, concentrate resources on undervalued channels, test new combinations between on and offline.
All this with the ability to measure the impact of those choices in the next model update cycle.

Performance vs brand: including offline

MMM mainly measures effects that are observable over the period analysed, but not all channels work on the same time horizon.
Performance channels — search, shopping, direct-response Meta campaigns, retargeting — generate rapid, measurable effects on short-term sales.

Brand channels (TV, connected TV, upper-funnel video, editorial content, PR, sponsorships, out-of-home, trade press) have longer horizons. They build awareness, positioning and trust, increasing the effectiveness of performance channels over time, both online and offline.

A well-built MMM model can estimate the contribution of brand channels on sales and brand searches, but their evaluation should always sit within a strategic logic. With monthly budgets above €50,000, a split that we see working — and that is cited in the literature — is a balance with 60–70% of investment oriented towards performance and 30–40% dedicated to building and fuelling the brand. The actual weights, however, depend on the industry, brand life stage, market share objectives and time horizon.

Paid, owned, earned: including all touchpoints

The true potential of Marketing Mix Modeling emerges when all relevant touchpoints, not just paid digital channels, are included in the model.

Paid: Google, Meta, display, sponsorships

Owned: email list, blog, organic social, app

Earned: PR, reviews, word-of-mouth, media coverage

In many projects, for example, we see that a well-planned TV or ConnectedTV flight increases brand searches and search and social campaign conversion rates within a few days. Likewise, a billboard campaign in strategic urban areas can increase response to local campaigns on Google and Meta, while an article in a reputable publication or strong PR coverage can generate a measurable lift on direct traffic, branded search and conversion rate.

Example: one of our clients in the marine industry discovered that an article in a specialised magazine (earned media) increased Google Ads campaign effectiveness by 20% for six weeks. Why? Brand awareness and trust.

For this reason, in a mature MMM model it makes sense to include: paid digital channels, paid offline channels, owned assets (website, app, blog, newsletter, organic social channels) and earned levers such as PR, reviews, word-of-mouth and media coverage. The goal is not to crown a winner, but to understand how the ecosystem as a whole generates results over time.

Real case: from all-in on Google to a profitable mix

Imagine an Italian e-commerce business in the home decor sector, about €3 million in turnover and a marketing budget of around €60,000/month spread across search, social, email, influencers and some offline initiatives in trade magazines.

Before introducing MMM, most of the budget was concentrated on Google Shopping and Search, with decreasing ROI due to saturation.
Meta was used in a patchy way, influencers had almost been abandoned because they were considered untrackable, and email marketing was seen as a retention lever only. Offline campaigns in print were not systematically linked to digital performance.

After a few months of work on the MMM model, several key insights emerged:

part of the Google spend was clearly beyond the saturation point;
Meta could scale while maintaining an attractive marginal ROI; micro-influencers generated lift on brand searches and overall conversions; email, with better segmentation, had scope to grow much more;
trade magazine placements had a measurable positive impact on digital campaigns in the following weeks.

The new allocation, still around €60,000/month, slightly scaled back Google, increased investment in Meta, influencers and email and made offline activities more targeted. Within six to nine months, overall ROI increased significantly, with strong revenue growth at constant investment and less dependency on a single channel.

When it makes sense to invest in Marketing Mix Modeling

MMM is not a universal tool and it is not needed at every growth stage.
It tends to generate maximum value when the marketing budget exceeds €50,000/month, at least four or five channels are active,
including both digital and offline, and the company has a data history of at least 12 months with spend and results consistently recorded.

It is especially useful when you want to move from a logic of tactical optimisation of individual campaigns to a logic of strategic budget planning, where you think in terms of quarters and years, not just single initiatives.
In these contexts, a well-designed MMM project can quickly pay for itself by freeing up wasted budget and reallocating it to more profitable channels.

Where to start, realistically

You don’t need to launch an “enterprise” project to see value from Marketing Mix Modeling.
You can approach it step by step, especially when you manage significant budgets and want more control.

A typical path starts with mapping online and offline channels with spend and results for the last 6–12 months, defining a consistent KPI set and centralising data in a single environment, and then building a first MMM model, even a simplified one, with tools like LightweightMMM or Meridian.

From there, the model is updated regularly, budget reallocations are tested and the impact of decisions is measured.
The key is not to treat MMM as a one-off report, but as a continuous process that feeds into the plan–execute–learn cycle.

MMM, Meridian, predictive AI and LLMs

The direction is clear: integrating MMM, tools like Meridian, predictive AI and language models
to move from ex-post analysis to near real-time simulations and recommendations.

In an advanced stack, the MMM model is regularly updated with new data, simulates budget scenarios by channel, market and period, quantifies uncertainty with Bayesian approaches and proposes reallocations along with an estimate of the expected impact. Recommendations can then be connected to media platforms, from Google properties to TV and CTV, to reduce the time between analysis and action.

Large Language Models (LLMs) become the conversational layer on top of this engine: they allow C-levels and marketers to query the model in natural language, for example asking what happens if 15% of TV budget is moved to YouTube in Q3 in a given country, and they translate complex outputs into operational decisions that are understandable and defensible in the boardroom.

At HT&T Consulting we work precisely on this convergence: using MMM and frameworks like Meridian as the quantitative engine and AI as the interface to make insights accessible even to non–data scientists, with tools like gemini-cli or Claude desktop.

Conclusion: from “in my opinion” to “according to the model”

Marketing Mix Modeling does not replace the marketer’s intuition, but complements it with a strong quantitative base, especially when significant monthly budgets are at stake across a mix of digital and offline channels.

The difference in front of the CFO is obvious: you’re not defending a feeling, you are discussing scenarios based on data, explicit models and measured results. The first step is not choosing a platform, but tidying up the data and clarifying which business questions must be answered.
From there, MMM becomes a concrete tool to protect budget, invest in the right channels — including TV, CTV and out-of-home — and build growth that is less dependent on single channels or “performance heroes”.

HT&T Consulting, Google Premier Partner, Google Marketing Platform Certified, Meta Business Partner and Supermetrics Partner, uses advanced frameworks such as Google Meridian for Marketing Mix Modeling and supports companies in making budget decisions based on solid quantitative analysis: if you want to understand how to apply it to your case, get in touch.

FAQ: Marketing Mix Modeling (MMM)

What is Marketing Mix Modeling in simple words?

Marketing Mix Modeling is a statistical model that links marketing spend by channel to business outcomes
such as sales, leads, revenue or margins. It uses aggregated historical data to estimate how much each channel,
online and offline, contributes to the final result and how much return you get from the last euro invested in each channel.

How is MMM different from digital multi-touch attribution?

Digital attribution works at individual user level and reconstructs the paths between impressions, clicks and conversions,
but it only sees what is trackable. MMM works on aggregated data and includes in the model TV, CTV, radio,
out-of-home, print, events and in-store, in addition to digital channels. It does not use cookies and is privacy-safe by design.
The two approaches are complementary: attribution guides operational optimisation, MMM guides budget planning.

When does it make sense to invest in a Marketing Mix Modeling project?

MMM makes most sense when the marketing budget is above €50,000/month, at least four or five channels
across digital and offline are active, and the company has at least 12 months of historical spend and results.
Under these conditions the model has enough data to produce reliable insights and the optimisation potential
is high enough to quickly pay back the investment.

Which channels can an MMM model measure?

An MMM model can include virtually all channels that absorb marketing budget:
search, shopping, display, social, email, influencers, SEO, marketplaces, TV, connected TV,
radio, out-of-home, print, trade shows, events, in-store promotions. The condition is to have time series
of spend and results that are consistent enough for each of these channels.

What kind of data do you need to start an MMM project?

You need spend data by channel, business performance data (sales, leads, revenue, margins),
context variables (seasonality, promos, launches, price changes, stock-outs, competitor actions)
and, if possible, geographic or product-line information. Ideally all of this should cover 12–24 months,
with consistent definitions and data centralised in a single environment.

Which tools can be used for Marketing Mix Modeling?

There are open-source frameworks such as Robyn (Meta), LightweightMMM (Google) and Meridian (Google),
as well as specialised SaaS platforms. The choice depends on mix complexity, number of countries,
internal skills and the desired level of autonomy. At HT&T we help clients choose
and integrate the stack that fits their context.

How long does it take to see results from an MMM project?

A first model can be set up in a few months, depending on initial data quality.
The first insights on budget allocation come with the first version of the model,
while the biggest benefits appear in the medium term, as the model is updated,
refined and tied into the quarterly or annual media planning cycle.

Does MMM work for SMEs with limited budgets?

MMM starts to become truly useful when the marketing budget is at least €50,000/month
and there are multiple channels. Below this threshold it often makes more sense to focus on
good digital attribution, controlled tests and smart use of the data you already have.
For more structured SMEs, however, a lean MMM can still be a competitive advantage.

How does MMM integrate with AI and language models (LLMs)?

AI can help automate model updates, simulate scenarios and generate budget recommendations.
Language models can become the natural interface to query MMM in plain language,
for example asking about the impact of shifting budget between TV, CTV and search in a specific period.
This makes insights accessible even to people without statistical expertise.

What are the main risks or mistakes in a Marketing Mix Modeling project?

The main risks are incomplete or inconsistent data, models that are too complex for the data available,
over-interpretation of results and lack of connection to business reality.
That is why it is essential to start from a solid data foundation, make model assumptions transparent,
and work jointly across marketing, finance and data teams.

Essential bibliography and online resources

To dive deeper into Marketing Mix Modeling, available tools and best practices in measurement,
here are some authoritative resources:

👋 Hai un'idea? Realizziamola insieme!