Facebook’s Tony Evans considers how Media Mix Modelling should be reinvented to reflect today’s marketing priorities.
Media Mix Modelling (or MMM) was first developed back in the 1970s to measure the impact of offline media such as TV, radio and press advertising. A lot has changed since then, including a plethora of digital platforms and higher expectations from CMOs toward actionability and accountability from advertising strategies. MMM can solve that, but only for those who change the way it is done.
There are people who would have you believe that marketing has changed since the internet came along. There are new tools and techniques for reaching audiences, for sure, but the very core remains the same: getting the right message to the right person at the right time in as cost-effective manner as possible.
As technology and data techniques have evolved, platforms have benefited. This technological evolution enabled smarter targeting and smarter delivery of ads fuelled by data and machine learning.
The overall experience became better for both the marketer and consumers. Better targeted and more relevant ads are great for the time-poor consumer looking for their next purchase. And, of course, they are great for the advertiser, which gets a much more cost-effective way of spending marketing budgets.
The first phase of MMM
Econometrics – the practical application of economic theory – has been in use in business for some time. Within this exists the marketing arm of the process of econometrics, namely MMM.
Statisticians ingest a number of variables that could impact sales and, through a process of regression, figure out which of the marketing channels or other factors (identified variables such as price discounting, competitor activity etc) was having the biggest impact. The data collection, analysis and then technique application to get to usable information in MMM has been a long and expensive process requiring specialists. As a result, MMM modelling was typically an annual event, neither dynamic nor driving timely actions.
Digital advertising and the rise of causality
The plethora of data available in digital marketing enabled a whole industry to look at the link between activity and sales. While many fall into the first trap of identifying only the last touchpoint before a sale is conducted online, the majority of marketers knew that was way too simplistic and incorrect.
Up until five years ago, it was not possible to easily look scientifically at different groups of people at scale, and determine whether their exposure to advertising had led to an action. Through products like conversion lift, sales lift and brand lift, marketers could now see whether their ad executions were truly driving actual behaviour.
A funny thing happened on the way to the future
The digital advertising industry is maturing, and an increased focus on privacy online has once again made MMM the unlikely hero. Advertisers will have to seek out new techniques to understand ad effectiveness. They need to be able to operate with their own data, where they are in control of complying with all necessary data regulations of ownership and processing of that data.
It’s also becoming clear that MMM is no longer a technique reserved to large, established teams within CPGs and auto brands. Mobile-app advertisers are already shifting toward MMM and innovating at a rapid pace, redefining how MMM is done for pure app players.
(Re) enter the ‘Econometric Hero’
One of the challenges is that MMM is a model: it’s a statistical assessment based on the inputs available. It looks at past impacts and makes conclusions. A study that scientifically shows causality would win every day over regression.
Ekimetrics, a data science company, has found that another challenge is that MMM tends to be slow. The challenge on data latency isn’t just on the media side, but in the fact that MMM needs a wide variety of data from across the business to run including competitor actions, market factors, distribution, product launches etc). Ekimetrics is working on low latency solutions, focused on the data pipeline and increasing access to traditionally harder areas of the business to capture.
Turning MMM into the tool for the future: Five things to ask
1. How can MMMs be built quicker?
Solutions now exist to automate MMM builds – Demand Drivers by Analytics Edge, for example. Leverage the power of cloud-based technology to help accelerate the frequency of delivery of insights. MMM can be built quicker with the right tools and data strategy.
2. How can we take more variables into the modelling?
Models should be run across as much advertising activity as possible.
The goal is to get constant learnings on as many campaigns as possible. It may mean that, in the short run, digital-only MMM may become the norm, while quick data availability and lower costs would allow for non-digital channels and other factors to be baked in quickly.
3. Could we build this ‘in-house’ in a cost-effective way?
To build in-house requires expertise. This needs investment in people and in tools. It can lead to more flexibility than outsourcing. It may mean partnering with agencies and consultancies, but in a very different way. Open source codes, including our very own project Robyn, are both a reflection and a catalyst of these changes. Set a three-year plan, starting with collaborating with your MMM provider, and ending with this MMM provider being a consultant to your in-house team of experts.
4. How can we ensure that models are accurate?
Removing bias completely from models is not achievable in the short term. However, minimising these biases is achievable now. Validation and calibration of MMM results with experiments run on and off Facebook should be a minimum requirement.
CMOs have to make multi-million-dollar decisions and holding these decisions to a higher degree of accuracy is key. Deloitte went ahead and has already developed validation and calibration techniques within MMMs it built. The 2021 Deloitte White Paper provides more details.
5. How can the modelling results show that it is worth the investment?
This is a topic for CMOs to ask themselves as much as their MMM provider. The results and efficiencies for your business are only useful if actions are taken as a result of the modelling. Put actionability at the core. Request forecasting and simulations to be done, and verify their accuracy over time.
Marketers who adopt rapid modelling alongside causal testing will be in the best position to understand advertising efficacy. As with everything companies have done in building successful businesses, if it were easy everyone would have done it. Those that make the choice now to invest time and resources in building this out will be the winners.