Purpose

The below highlights Day1Data’s approach to Marketing Mix Models (MMM), including tips & tricks for its implementation.

What is a Marketing Mix Model?

Marketing Mix Models are a set of models which are used to measure true marketing incrementality. They were originated in the 90s by XYZ as a solution to measure marketing effectiveness & optimise budgets towards certain goals. 

The models quantify the impact of offline media (i.e. OOH advertising), which can help Marketing departments put an end to the ‘brand’ vs ‘performance’ debate. The models also capture seasonality and macroeconomic events (like unemployment rates, GDP etc.) making it a really efficient method of capturing the natural demand of your product or service. The MMM can be tuned to support the commercial mix, provide actionable insights around pricing & promotions strategy. However, most importantly, it’s a solution that doesn’t rely on cookies. Google reports that by the end of 2025 there will be no more advertising cookies due to data protection advancements. MMM represents a future-proof solution to marketing measurement. 

Is MMM right for you?

Firstly, it’s important to caveat that MMM cannot & should not be implemented by every company. 

To create an accurate MMM, a company will typically need 3 years worth of historical data, including varied spend levels, business strategies & overall performance. Additionally a variety of paid marketing channels is also crucial to help establish sub-channel effectiveness. These variables help inform the model. In our experience, a company will need to be spending over $100,000+ per annum on marketing to help the model get an accurate read & to justify the roll out of the models due to the costs associated on building it, specifically:

• Creating an econometrics team

• Spend on a vendor / consultancy

• The data storage & querying costs

The above should be considered investments in ROI calculation: you’ll need the value to justify the cost. This is the same for both D2C & B2B companies, the thing that will change is the model KPI(s) & the data inputs. When applied in the correct conditions and the business is primed to operationalise the insights, the MMM can have an overwhelmingly positive impact.

Moreover, MMM is not a pure attribution solution, such as multi-touch attribution. It is about incrementality. This means it’s an actionable methodology to support measurement, scenario planning, forecasting and budget optimisation. It should be used in conjunction with continuous experimentation to help optimise the MMM outputs. 

Despite this, the methodology isn’t without its risks. Determining model accuracy is really difficult in the absence of testing the model outputs. The model metrics can lie. It’s possible to make models statistically accurate (in terms of the MAPE or R2), but these may not be a true reflection of your business performance. That’s why we’d recommend only rolling MMM out when used in conjunction with incrementality testing to validate the output. Additionally, as referenced earlier - traditionally this solution is not the cheapest option. 

With this in mind, here’s our advice on when to use MMM:

1. Use MMM only if the internal company circumstances match what we laid out within the article introduction

   1. 3 years of historical data

   2. Large spend levels

   3. Varied media mix

2. If your company can determine Marketing diminishing returns via other means such as last-click attribution & testing, then MMM is not a necessity for you

3. Use MMM in conjunction with incrementality testing (which is the ground truth). The testing output can be included as a model feature in order to improve the accuracy

What are the options for implementing an MMM?

If you’re looking to kick off your MMM journey, there’s a variety of options you can take:

• You can build the models in-house leveraging open source versions of MMM foundational code. Robyn from Meta & Meridian from Google are great options

  • Pro - all IP stays in-house

  • Con - MMM expertise is normally required to build something accurate & meaningful

• There’s subscription based SaaS tools which are available, with extremely low barriers to entry, for example Fospha or Lifesight

  • Pro - very fast modelling turnaround time & intuitive dashboards to help bring insights to life

  • Con - dependency on a 3rd party tool, black box solution, high cost, accuracy,  internal teams need to be involved in data and infrastructure provisioning

• Varieties of consultancy based approaches:

  • Larger consultancies have proprietary MMM solutions (which are typically black box), however

    • Pro - they bring years of experience of MMM & industry benchmarks which are useful

    • Con - the solutions are black box, the price is typically extremely high, internal teams need to be involved in requirements & data provision

• Smaller consultancies who help build the models for you

  • Pro - they’re an extension of the internal team, IP stays within the company, lower prices, accuracy

  • Con - internal teams need to be involved in requirements & data provision

The Day1Data approach is in line with the small consultancy way of working. We’d build bespoke models for the company, ensuring IP stays within the firm & specific company nuances are built into the model. This is our preferred approach as it is extremely transparent. We believe transparency is crucial.

Marketing spend represents over 40% of an average companies’ expenditure - the stakes have never been higher to de-risk harmful growth decisions. More transparency into the model, leads to more trust, to help steer companies to make data-driven decisions & operationalise the outputs.