TechnicalReplatforming ManagementPlatform SelectionProject DeliveryRequirements GatheringRFP Management
Vervaunt is a London-based ecommerce consultancy and paid media agency, working alongside ambitious ecommerce brands and retailers to improve key areas of their business and drive growth.
Businesses need to be able to understand the impact of different marketing efforts in order to make informed decisions when it comes to strategic prioritisation and budget management. Media mix modelling can act as the solution for this, but what is it?
Media Mix Modeling (MMM), sometimes referred to as Marketing Mix Modeling, is an analysis technique that uses aggregate data to provide stakeholders with a bird’s eye view of their marketing ecosystem and to demonstrate how different elements contribute to their main business objective of driving sales / conversions.
MMM originated from the 1960s-70s, and Kraft was one of the early adopters with their Jell-O product. This is when the notion of “big data” started to emerge and stakeholders started to view marketing as more of a scientific entity. Increased investment in data collection resulted in a shift in thinking that is increasingly relevant today, with the growth of digital marketing.
It is a high-level, holistic evaluation that takes into account all marketing channels, including traditional and digital platforms (billboard and TV advertising plus paid search and paid social), as well as non-marketing sources including external factors (like seasonality and the economic climate) and internal variables (such as mark-downs), to determine the extent to which each input is impacting overall sales. This then gives an indication of efficiency by channel and where the budget should be allocated going forward in order to maximise sales volume.
Through this holistic analysis, which incorporates the entire media portfolio, data can be leveraged for forecasting to make predictions about expected sales growth based on the changes applied to media spend across different forms of activity. MMM analysis relies on data that has been aggregated over a longer period of time, usually a multi-year historical period (three years of data is recommended), to allow for a fair representation of trends and patterns.
DDA is an effective analysis technique when considering person-level data and customisation, which Media Mix Modeling, as a holistic channel-focused strategy, is unable to support.
In summary, DDA uses real-time data instead of years of historical data, looks at user-level information including customer experience, and is therefore more agile and better suited to day-to-day optimisations. This is why digital marketers (in particular, paid search and paid social specialists), usually harness this model for campaign management.
However, unlike MMM, which provides a birds-eye view of the entire ecosystem, DDA models are digital-focused and limited when it comes to measuring offline conversions. DDA models aren’t able to provide a view of the whole ecosystem so MMM is used to inform high-level budgeting decisions and for wider sales forecasting.
By looking at how different factors contribute to a business’ marketing objective, Media Mix Modeling is closely intertwined with the marketing 4Ps: Product, Price, Place, and Promotion.
As mentioned, MMM takes into consideration all forms of marketing (Promotion) as well as other variables like mark-downs (Price). Paid search and paid social activity play an important role in the promotion side, and the MMM should include all paid channel segments to provide a complete view of the impact of each activity under promotion.
As MMM is a holistic view which considers all inputs, factors related to Product and Place will also be built into the model to provide a complete picture which helps to determine the impact that pulling a particular lever has on the sales output.
In order for Media Mix Modeling to provide a business with meaningful insights, it is crucial that:
Data is of a high quality, which means companies need to dedicate resource to aggregating and cleansing data from all sources
There is enough data, to allow for analysis covering a multi-year historical period (three years is recommended)
Analysis is conducted at the right level of frequency (annually or bi-annually is the recommended frequency)
The right tools are leveraged to pull big data into a unified platform for measurement
Nielsen summarises the Media Mix Modeling process as incorporating 4 key stages: Collect, Model, Analyse and Optimise.
Collect - gathering all of the relevant data, including all marketing inputs, external factors, and internal variables: price, promotion, online media, offline media, distribution, competitive data.
Model - using advanced statistical models to connect marketing and other activities to performance.
Analyse - crunch the numbers to assess the impact of all activities on business performance.
Optimise - adjust marketing spend through simulations to test the impact on results.
Preparation is key. Google recommends following 4 steps in order to effectively launch media mix modelling:
This is crucial for making sure that the right data is collected from across the business. It will save a lot of time and effort if goals are clearly laid out ahead of requesting information from the different data gatekeepers. You’ll be more likely to benefit from streamlined communications and it will prevent unnecessary back and forth.
Focus on one particular metric to avoid confusion and inaccurate data collection. Get this right at the beginning to ensure the right information is being tracked.
In order to ensure a smooth process for collecting this large amount of data, it is important to assign the responsibility of data sharing to one key gatekeeper per data set.
Provide these representatives with an induction into the MMM including a breakdown of their responsibilities, a demonstration on the format in which this data should be shared, and a timeline which should be followed for data processing.
Make sure to engage all of the relevant people to ensure all data is encompassed in the model. This will most likely include setting up communication with CMOs, TV and media agency partners, marketing agency partners including paid search and social teams, CRM managers, and marketing executives.
Data needs to include all channels, be mutually exclusive to avoid unfair attribution, and provide a granular channel view in order to inform budgeting decisions.
Create a detailed visualisation for the data ecosystem, highlighting any blockers (for example payments and subscriptions needed for data access from third parties, or time delays related to offline data access).
Media mix modeling uses regression analysis (multi-linear regression), a statistical analysis technique that tests the impact that independent variables have on a dependent variable. The MMM ratio is made up of three core elements:
The inputs being applied
The spend on each input
Results per input
Large organisations dealing with a large number of inputs should look at leveraging MMM AI-powered and automated solutions to set up MMM. These platforms avoid manual analysis and data cleaning work and instead provide a solution that enables businesses to use AI and advanced analytics to identify the true impact of different variables and allows for advanced forecasting capabilities.
There are a host of different platforms available but businesses should choose a tool that can collect data from diverse sources, weigh up long and short term growth, and take into consideration external factors. Here is a list of some of the AI-powered solutions available:
Proof Analytics - Run "What if" Wargame Scenarios, Get Transparent Visualizations, A Dynamic GTM Analytics Dashboard.
Nielsen Marketing Mix Modeling - provides answers to critical marketing questions using industry-leading performance models for data coverage and granularity
MassTer by MASS Analytics - user-friendly and intuitive interface
Sellforte - Always-on MMM solution for companies that invest both in offline and online marketing
Scanmar QED - Quickly build robust and reliable models to make smarter business decisions
GfK Marketing Mix Optimizer by GfK - Quantify your drivers of sales and understand the effectiveness of your marketing across all channels to optimise your investments
Marketscience Studio - provides a modern, integrated environment for advanced marketing analytics
If you’re a business or marketer dealing with a smaller data set, there is the option of building the Media Mix Model yourself. This article from Towards Data Science guides you through the steps of building the model from scratch in Python.
Although MMM is designed to be used by businesses for a holistic view of the organisation’s entire ecosystem, MMM principles can be adopted into a scaled-down version by specialist teams, as a way of analysing the impact of different channels.
For example, paid media agencies can build their own MMM - pulling in data from all channels and variables, in order to make forecasting simulations and inform decisions about budget allocation within their own department.
If you’re a paid media agency building your own Media Mix Modelling microcosm, all of the same principles discussed in this article apply with regards to collecting data and the stages of the MMM process. Remember, MMM relies on a multi-year historical period so make sure the inputs have been live for around three years.
For paid media teams, the inputs of your model can be taken to a more granular level to identify the impact of different activity - segmenting into channel and format.
An example of a data input structure could be:
Google text ads
Bing text ads
+ “X” search engine + format
Facebook/Instagram image ads
Facebook/Instagram video ads
Facebook/Instagram carousel ads
Facebook/Instagram collection ads
+ “X” social platform + format
In the same way as a traditional MMM, this can help to create simulations and forecasting and inform important budgeting decisions across channels and formats.
In its true form, media mix modelling enables stakeholders to assess the impact of different inputs across the organisation’s entire ecosystem. It is crucial that the data collected is of a high quality and that data covers a multi-year historical period to allow for a representation of trends. When dealing with a large volume of data, AI-powered and automated solutions are worth investigating as these platforms harness advanced analytics for simulations and forecasting to help inform important budgeting decisions. It’s important to pull all channels and variables into the model to allow for a true picture of the ecosystem.
Our paid media team are experts in data reporting and are ready to integrate our channels into your media mix model. If you have any questions about our services and want to discuss with our team, feel free to get in touch.