Get a quick overview of Marketing Mix Modelling in 60 seconds
Why Marketing Mix Modelling is having a moment
Marketing Mix Modelling isn’t new. It was developed in the 1960s by consumer goods giants who needed a way to measure TV, radio, and print advertising in a world with no click data. It became the dominant measurement approach for large FMCG and CPG businesses and stayed that way through the 1990s.
Then digital arrived — and Marketing Mix Modelling fell out of fashion. Platform attribution promised more granular, real-time answers. Marketers could see exactly which ad drove which conversion. Marketing Mix Modelling’s aggregate, quarterly view felt slow by comparison.
That era is ending. And not gradually.
Three structural forces are pushing Marketing Mix Modelling back to the centre:
- Privacy constraints have broken platform attribution. The deprecation of third-party cookies, Apple’s App Tracking Transparency framework, and rising consent rates have made user-level tracking increasingly unreliable. Both Meta and Google now operate significant portions of their inventory under modelled reporting — their own attributed conversion numbers are detached from reality, and they know it. Both are now active advocates for Marketing Mix Modelling as the complement to their own reporting. Not because it flatters their numbers. Because it provides an independent, privacy-safe view of true incremental impact.
- Walled gardens have made cross-channel measurement impossible from the inside. Google, Meta, Amazon, TikTok — each platform reports its own attributed results. And none of them agree. Adding up platform-reported ROAS figures typically produces a number 2–4× higher than actual sales growth. Marketing Mix Modelling, because it operates on actual business outcomes rather than platform-reported conversions, cuts through this. It doesn’t care how each platform counts conversions. It looks at the real-world relationship between spend and sales.
- Brand marketing has always been unmeasurable by attribution. Above-the-line investment in TV, out-of-home, sponsorship, and brand awareness produces no click data. For brand marketers, Marketing Mix Modelling has never stopped being the primary measurement tool. And with brand budgets recovering after years of performance-heavy allocation — and with mounting evidence that short-termism has eroded brand equity across many categories — Marketing Mix Modelling is being picked up again by leaders who need to prove the long game pays off.
How Marketing Mix Modelling works
At its core, Marketing Mix Modelling is built on regression analysis — a statistical technique that identifies the relationship between a dependent variable (usually sales or revenue) and a set of independent variables (marketing spend, promotions, pricing, seasonality, and more).
The model is trained on historical data, typically two to five years of weekly observations. It learns the relationship between changes in each input variable and changes in sales over time — controlling for everything else simultaneously.
Here are the concepts that matter most:
Base sales vs. incremental sales
One of the most powerful outputs of Marketing Mix Modelling is the decomposition of total sales into base and incremental components.
Base sales are the sales that would happen even if you spent nothing on marketing. They’re driven by brand equity, distribution, repeat purchase behaviour, seasonality, and broader category trends. For established brands, base typically represents 40–70% of total volume.
Incremental sales are the additional sales directly attributable to specific marketing activities — a TV burst, a paid search campaign, a promotional event.
Understanding this split is strategically important. A brand with very high base sales has a different challenge than one that’s heavily dependent on promotions for volume. If your base is high, that’s brand equity working. If it’s low, you’re on a promotional treadmill — and Marketing Mix Modelling will tell you that clearly.
Adstock: how advertising keeps working after it airs
Advertising doesn’t produce an instantaneous spike and then disappear. Its effects carry over time — sometimes for weeks, sometimes longer. This carryover effect is called adstock.
Marketing Mix Modelling models the adstock of each channel separately, because different channels decay at different rates. Paid search has a very short adstock — the effect is concentrated in the days immediately around the click. TV has a much longer adstock — a well-executed campaign can produce measurable sales effects weeks after it aired.
Getting adstock right is critical. A model that ignores it will systematically underestimate the value of brand-building channels, because it only captures their immediate sales effect — not the tail.
Saturation curves: where more spend stops paying off
There is a point at which additional spend on any channel produces progressively less return. This is the saturation effect, captured in Marketing Mix Modelling through saturation — or diminishing returns — curves.
The saturation curve tells you how much incremental sales you get for each additional unit of spend on a given channel. In the early part of the curve, returns are strong. As you approach saturation, each additional pound or dollar produces a smaller and smaller return.
Knowing where you sit on the saturation curve for each channel is one of the most actionable outputs Marketing Mix Modelling produces. It tells you which channels have headroom to scale — and which you could pull back from without proportionally hurting sales.
Causal inference: why correlation isn’t enough
Traditional Marketing Mix Modelling relies on correlations in historical time-series data. But correlation is not causation. If TV spend and sales both rise in Q4, is the TV driving sales, or are both driven by Christmas?
Modern Marketing Mix Modelling methodology increasingly incorporates causal inference techniques — including geo-level experiments (geo-lift tests) that validate the model’s causal assumptions. A geo-lift test runs advertising in some geographic markets but not others, then measures the sales difference. This experimental data is fed into the model as a calibration anchor, improving causal accuracy significantly.
Google’s open-source Meridian framework and Meta’s Robyn are both built on Bayesian hierarchical modelling, which incorporates prior knowledge and uncertainty into the model rather than treating every observation as equally informative. Instead of a single-point estimate, Bayesian models produce probability distributions over outcomes — giving marketers a clearer picture of confidence and risk, not just a headline number.
Marketing Mix Modelling for brand marketers vs. performance marketers
This is the question most Marketing Mix Modelling guides don’t answer properly — because most vendors are built for one audience or the other.
Marketing Mix Modelling was originally a brand marketer’s tool. It was designed to measure the impact of large, diffuse brand campaigns on aggregate sales, in a world where digital attribution didn’t exist. For brand marketers at FMCG, CPG, retail, and financial services companies, it remains the primary framework for understanding ROI across the full media mix — TV, out-of-home, print, sponsorship, digital display, and paid social.
For brand marketers, Marketing Mix Modelling answers questions like:
- What is the true sales ROI of our TV investment, accounting for its long adstock?
- How much of our sales are driven by brand equity vs. promotional activity?
- What happens to base sales if we reduce brand advertising spend for two years?
- Is our above-the-line investment building or eroding brand equity over time?
- What’s the right balance between brand building and short-term activation?
That last question is one of the most strategically important in marketing. Les Binet and Peter Field’s work demonstrated that the optimal balance between brand building and activation for most categories sits around 60:40. Marketing Mix Modelling is the tool that shows you where you actually are on that spectrum — and what happens if you rebalance.
For performance marketers, Marketing Mix Modelling answers different questions:
- Which digital channels are genuinely driving incremental sales vs. capturing conversions that would have happened anyway?
- Where are we on the saturation curve for Meta, Google, and other paid channels?
- How should we reallocate budget across campaigns and channels to maximise incremental return?
- What is our true marketing-attributable revenue — not what platforms claim?
The convergence point: both brand and performance marketers need a single measurement framework that reflects the actual relationship between spend and outcomes — not one that’s siloed by channel type or platform boundary. Marketing Mix Modelling is currently the only methodology that provides this unified view: all channels, both paid and unpaid, both digital and traditional, privacy-safe and independent of any platform’s self-reported data.
Nepa Session – How does Marketing Mix Modelling (MMM) drive growth
A session where we explore how Marketing Mix Modeling (MMM) is helping leading brands make smarter, faster marketing decisions. Watch the on-line webinar now.
Marketing Mix Modelling in action:
Real results from real brands
Hemköp: turning brand insight into strategic clarity
Hemköp, one of Sweden’s leading grocery retailers, had high brand recognition — but struggled with differentiation. In a crowded market where most grocery brands look and sound alike, recognition alone doesn’t move the needle.
Working with Nepa, Hemköp combined Marketing Mix Modelling with a Brand Asset Study and Campaign Evaluation to understand not just what their media spend was doing, but why some investments built lasting brand value while others didn’t.
The result was a fundamentally clearer picture of where Hemköp’s brand stood in consumers’ minds, which media were genuinely building equity, and how to make every investment count. Marketing Mix Modelling provided the financial quantification; brand tracking provided the explanatory layer. Together, they gave Hemköp’s marketing leadership the confidence to make bolder strategic moves — backed by evidence, not instinct. Read the full case here.
SBAB: from trust to growth in financial services
SBAB has been a trusted name in the Swedish mortgage market since 1985. But trust alone doesn’t drive growth. When the market accelerated and competitive pressure intensified, SBAB needed to understand — with precision — what was driving their business outcomes and where they should invest next.
Together with Nepa, SBAB implemented a Continuous Marketing Mix Modelling approach, combined with Brand Tracker, Campaign Pulse, and qualitative research. This always-on model — updated regularly rather than run as an annual project — meant SBAB could make media allocation decisions based on current data, not last year’s snapshot.
The outcome was a sharpened competitive strategy, clearer channel allocation, and the ability to turn consumer insight into real commercial direction. Not a one-off report. An ongoing engine of growth intelligence. Read the full SBAB case here.
Proven results across FMCG brands
Across our FMCG clients, Nepa’s Marketing Mix Modelling has delivered an average 15% ROI increase by optimizing how they allocate media budgets across channels. This isn’t theoretical improvement — it’s measurable, bottom-line impact from redirecting spend from saturated channels to those with headroom to scale.
The pattern is consistent: brands that combine Marketing Mix Modelling with ongoing brand tracking make more confident strategic decisions, allocate budgets more effectively, and build stronger market positions over time.
Marketing Mix Modelling vs. multi-touch attribution vs. brand tracking
These three methodologies are often treated as competitors. They’re better understood as tools that answer different questions at different time horizons — and work best when they’re connected.
Multi-touch attribution (MTA)
MTA tracks individual users across digital touchpoints and assigns fractional credit to each interaction. It is designed for digital-only, performance-focused measurement, and operates at the level of individual sessions and clicks.
Good for: Understanding digital touchpoint sequences. Optimising bid strategies within digital channels.
Not good for: Measuring offline channels. Measuring brand-building effects with long time horizons. Any measurement that requires user-level data — which privacy constraints are making increasingly unreliable.
Marketing Mix Modelling
Marketing Mix Modelling operates at an aggregate level using historical time-series data. It doesn’t depend on user-level tracking. It covers all channels — online and offline — and can model both short-term sales activation and longer-term brand equity effects.
Good for: Cross-channel budget allocation. Measuring channels that produce no click data. Understanding saturation and diminishing returns. Modelling the long-term impact of brand investment. Strategic planning and scenario simulation.
Less suited for: Campaign-level creative optimisation. Real-time bid management. Very short-term (daily) fluctuations.
Brand tracking
Brand tracking measures changes in consumer perception — awareness, consideration, preference, purchase intent — typically through regular survey-based research. It doesn’t directly measure sales impact, but it captures the brand equity metrics that Marketing Mix Modelling’s base sales component reflects.
Good for: Monitoring brand health over time. Diagnosing whether communications are shifting perceptions. Early warning of brand equity erosion.
The connection between brand tracking and Marketing Mix Modelling: Brand tracking data can be incorporated directly into Marketing Mix Modelling as an input variable, allowing the model to quantify the sales impact of changes in brand equity. This is one of the most powerful combinations in marketing measurement — and the one Nepa’s platform is uniquely built around.
What does Marketing Mix Modelling output actually look like?
Most people have never seen a Marketing Mix Modelling readout. Here’s what you actually get:
Sales decomposition charts
A breakdown showing how much of your total sales comes from base (brand equity), and how much comes from each marketing channel. This is your strategic foundation — it tells you whether you’re building a brand or running on a promotional treadmill.
Channel performance scorecards
ROI, reach, and contribution metrics for every channel in your media mix. Not what platforms claim they delivered — what actually moved your sales needle.
Saturation curve analysis
Visual representations showing where each channel sits on its diminishing returns curve. The channels with headroom become your scaling priorities. The saturated ones become your efficiency opportunities.
Budget reallocation recommendations
Specific guidance on where to increase, maintain, or reduce spend for maximum incremental return. These aren’t suggestions — they’re data-driven investment instructions.
Scenario planning tools
“What if” simulators that let you model the impact of budget changes before you make them. What happens if you shift 20% of TV budget into digital? The model tells you, with confidence intervals.
What data do you need for Marketing Mix Modelling?
Data quality is the single biggest determinant of Marketing Mix Modelling quality. A well-specified model on poor data will produce unreliable outputs. Here’s what a standard Marketing Mix Modelling dataset requires.
Sales and outcome data
- Weekly sales volume or revenue by market or geography
- At least 2 years of data; 3–5 years is preferable
- Broken down by product line, channel (online vs. in-store), or region where possible
Marketing spend data
- Weekly spend by channel: TV, radio, OOH, paid search, paid social, programmatic display, email, direct mail
- For digital: impressions or GRPs in addition to spend where available
- Creative flight dates and weights for TV, to support adstock modelling
Pricing and promotional data
- Regular and promotional retail selling prices by week
- Promotional events, discount depth, distribution changes
- Trade promotional activity where relevant
External and control variables
- Seasonality indices
- Competitor spend data (optional but improves model specification)
- Macro-economic indicators where relevant
- Structural break flags (COVID disruptions, major market events)
What you don’t need: You don’t need user-level data, cookies, device IDs, or any form of personal data. Marketing Mix Modelling is entirely privacy-safe by design — it operates on aggregated data at the market or geography level.
How long does Marketing Mix Modelling take, and what does it cost?
Timeline: A first Marketing Mix Modelling build typically takes 8–12 weeks from data collection to first outputs. This includes data preparation (often the most time-consuming step), model building, validation, and stakeholder readouts.
Modern software platforms can compress this significantly — and the shift from annual project-based Marketing Mix Modelling to always-on, continuously updated models is one of the most important changes in the market. The difference matters: a model updated monthly tells you something actionable now, not what was true twelve months ago.
Cost: Nepa’s Marketing Mix Modelling is competitively priced. View our full pricing structure here to see exactly what’s included in each plan. No hidden fees, no project overruns, just clear, predictable investment in your growth intelligence.
Traditional consultancy approaches range from £30,000–£80,000 for project-based work. Open-source tools (Meta’s Robyn, Google’s Meridian) are available at no licensing cost but require significant internal data science resource to implement and maintain.
Common limitations — and how modern Marketing Mix Modelling addresses them
Every honest guide to Marketing Mix Modelling includes its limitations. Here are the most important ones, and where the field is heading.
Granularity. Traditional Marketing Mix Modelling works at the channel level, not the campaign or creative level. Modern approaches are increasingly operating at campaign and even ad-set level for digital channels — closing this gap meaningfully.
Latency. A model built on historical data can’t immediately reflect market changes. Always-on platforms with monthly or weekly refreshes reduce this lag dramatically compared to annual project cycles.
Correlation vs. causation. Observational models risk confusing correlated variables with causal ones. Geo-lift experiments and Bayesian calibration significantly improve causal accuracy.
Data requirements. Marketing Mix Modelling needs clean, consistent, long-run data. Businesses with fragmented data infrastructure find this difficult. Automated data ingestion is reducing the friction — but data quality remains the limiting factor for most organisations.
How to choose a Marketing Mix Modelling partner or platform
Whether you’re evaluating a consultancy or a SaaS platform, these are the questions that matter:
Methodology. Does the model use Bayesian inference? Can it incorporate geo-lift experiments for calibration? How does it handle adstock and saturation? Can it model long-term brand equity effects — or only short-term sales response?
Cadence. Is this a one-off project or an always-on solution? How frequently can the model be updated?
Transparency. Can you see the model specification? Are outputs explainable to non-technical stakeholders? Do they show confidence intervals or only point estimates?
Use case fit. Is the vendor primarily built for performance marketing (eCommerce, DTC) or for brand marketing (FMCG, retail, omnichannel)? Does their client base reflect businesses like yours?
Integration with brand tracking. Can the model incorporate brand health metrics as input variables? If you want to connect brand equity to business outcomes — not just media spend to sales — this is the question that separates most providers.
The bottom line
Marketing Mix Modelling isn’t a new idea. But it’s the right tool for the moment we’re in — one defined by privacy constraints, fragmented media, unreliable platform attribution, and a growing imperative to prove the value of both brand-building and performance investment.
The organisations getting the most from it aren’t treating it as a periodic reporting exercise. They’re using it as an ongoing intelligence engine that informs planning, budgeting, and optimisation decisions in near real-time.
If you’re starting from zero: get your data in order first. Clean, consistent, covering at least two years of weekly observations across all spend lines and your primary outcome metric. That’s the foundation everything else is built on.
If you already have Marketing Mix Modelling in place: the most important next question is whether you’re modelling long-term brand effects, or only short-term sales response. The two tell very different stories about where your budget should go — and most models only tell one of them.
Ready to see how Marketing Mix Modelling can transform your marketing strategy?
Nepa combines Marketing Mix Modelling with brand tracking and campaign measurement — giving you a complete view of what’s building your brand and driving growth.
Get a free consultation →
Nepa is a global marketing intelligence company combining brand tracking, campaign measurement, and marketing mix modelling. Our platform connects brand equity data to business outcomes — giving marketing leaders a single, unified view of what’s building their brand and driving their growth.
