What is Marketing Mix Modelling?

Executive Summary

✦Marketing Mix Modelling (MMM) is a statistical technique that measures the true incremental contribution of every marketing channel — including the ones last-click attribution structurally ignores. ✦Unlike pixel-based attribution, MMM works with aggregated, privacy-safe data and doesn't break when cookies disappear or platform pixels over-count. ✦Modern Bayesian MMM adds prior knowledge and quantifies uncertainty, so you see not just a point estimate but how confident the model actually is. ✦For DTC brands spending across 3+ paid channels, MMM answers the question attribution can't: 'If I move £20k from Meta to TikTok, what happens to revenue?' ✦Pangolin combines Bayesian MMM with autonomous execution — the model doesn't just recommend a budget shift, it can action it across platforms in real time. ✦This guide covers how MMM works, what inputs it needs, why the Bayesian approach matters, what outputs you actually get, and how to tell whether your brand is ready for it.

The Attribution Crisis

Why your current measurement is lying to you

Every ad platform takes credit for the exact same transaction. If a customer clicks a Google Search ad, views a Meta video, and encounters a TikTok trend before purchasing, all three platforms will report a successful conversion on their respective dashboards. If you add up the platform-reported conversions, you will frequently find they exceed your actual bank-level revenue by 30% to 80%.

Relying on last-click attribution forces your marketing team to make high-stakes decisions in a structural vacuum. Last-click measurement is inherently blind to organic baseline demand, brand equity, offline word-of-mouth, and top-of-funnel discovery. Because you only reward the final touchpoint, you systematically starve the very campaigns that feed the funnel, ultimately driving down blended efficiency.

The cost of optimising the wrong number

Brands that switch from last-click allocation to MMM-informed budgets typically find 8–15% incremental revenue from the same total spend — not by spending more, but by moving money to where it actually works.

Mathematical Foundations

How MMM works

Marketing Mix Modelling is a statistical technique that measures the contribution of each marketing channel to a chosen business outcome — typically revenue or new-customer acquisition — using aggregated, historical data rather than user-level tracking.

Instead of tracking a single browser cookie, MMM looks at how peaks and troughs in your marketing spends match up with peaks and troughs in your total sales over time. By analyzing this historic covariance, the regression model isolates the unique lift generated by each channel. This math accounts for carry-over effects (adstock) as well as diminishing returns (saturation), ensuring that you don't over-fund highly saturated channels.

The inputs MMM needs

[01]Spend data

Daily or weekly spend per channel, ideally spanning 12+ months to account for long-term cycles.

[02]Outcome data

The foundational metric you are optimizing, such as raw revenue, total orders, or verified new customers.

[03]External factors

Contextual variables like regional seasonality, promotion periods, pricing adjustments, or competitor activities.

Probabilistic Precision

Why Bayesian — and why it matters for DTC

Frequentist models demand years of daily records to arrive at a stable point. But DTC brands move too fast for that. Bayesian statistics solve this by incorporating "priors" — existing business knowledge from incrementality tests or platform data. The model updates these priors with incoming weekly spends, mapping out a posterior probability distribution that shows you the exact likelihood of a channel's contribution.

What MMM measures that attribution cannot

✓Incremental contribution: The net new revenue directly caused by a marketing channel after mathematically stripping away baseline demand. ✓Diminishing returns (saturation curves): Dynamic curves that indicate exactly when additional spend on a channel stops scaling efficiently. ✓Carry-over effects (adstock): How long a pound spent on marketing today continues to influence conversions over future weeks.

Limitations & scepticism What MMM cannot do — and when to be sceptical

•It cannot attribute at the individual customer or session level — it is macro, not micro. •It needs enough historical data to detect signal (thin history = wide uncertainty bounds). •It is only as good as the outcome variable you choose — optimize the wrong metric and the model faithfully optimises the wrong thing. •A wide posterior on a channel coefficient is not a flaw — it is the model telling you the truth about what it can and cannot see.

On confidence intervals: A wide posterior distribution on a channel coefficient is not a flaw — it is the model telling you the truth. If the data genuinely cannot distinguish a channel effect from noise, a good Bayesian model says so. That honesty is the point.

Actionable Outputs

What you get — from model output to live budget

01 Channel contribution See exactly how much each channel actually drove, net of organic baselines and external factors, presented as clear percentage contributions.

02 Response curves The diminishing-returns curve for each channel, mapping out the precise mathematical point where scaling up your spend stops paying back.

03 Budget scenarios Interactive 'what if' simulations that let you stress-test reallocation scenarios before risking any real capital.

Autonomous Reallocation

How Pangolin closes the loop

Traditional media mix modeling suffers from a latency problem: you hand over your historical files to a consulting agency, wait six to twelve weeks, and receive a bulky PDF of strategic suggestions that are already out of date. Pangolin transforms this dead retrospective into an active execution system.

The Pangolin difference

Most MMM vendors stop at the PDF. Pangolin connects the model to your ad accounts. When the model says move £3k from Generic Search to TikTok TopView, Pangolin can execute that shift — with human approval or autonomously, depending on your comfort level. The model updates weekly, the agent acts on the latest posterior, and the loop between measurement and execution shrinks from quarterly to continuous.

See how Pangolin works — Book a demo

Fit Assessment

Is Pangolin right for your brand?

Pangolin's algorithms are tailored for growing DTC companies, but mathematics require minimum data densities to isolate signals from noise.

If your total marketing spend is below £100k/month across channels, MMM works as an interesting hypothesis-generator, but the math may be too thin for highly confident channel-level decisions. Once your blended budget scales above £150k/month across 3+ active paid channels, MMM becomes a genuinely actionable system of record.

Ask your agency this

"Next time your agency presents a channel plan, ask: What would happen to revenue if we halved spend on this channel? If the answer is a guess, you don't have measurement — you have an opinion with a spreadsheet."

Glossary of terms

Key Terms Explained

Do we need an in-house data science team?

No. Pangolin is a complete software-as-a-service solution. Our pipeline automatically cleans your data, fits the Bayesian algorithms, and presents the output in an intuitive interface. We handle the hard mathematics so you can focus on allocation decisions.

How often do Pangolin's models update?

Models update automatically every single week. We ingest daily transaction data, process baseline adjustments over the weekend, and deliver the final, validated attribution outputs and recommendations on Monday morning.

What data history is required to start modeling?

We recommend at least 12 months (ideally 24 months) of historical daily sales and advertising spend data. This history is crucial to train the model to understand seasonality and baseline organic performance levels.

Does Pangolin replace our existing analytics stack, or sit alongside it?

Pangolin sits alongside your existing tools. We don't ask you to rip out GA4, your ad platform dashboards, or your CRM. We ingest data from them and turn it into a single incremental-revenue view your team can act on.

How long until we see our first output?

Once your data sources are connected, initial model outputs are typically available within days, not the 3 to 6 months a traditional MMM consultancy takes. Full confidence intervals stabilise over the following few weekly refreshes as the model sees more data.