In episode six of The Undiscovered Metric, we’re joined by Sven Meijer, CEO of the Amsterdam-based marketing consultancy, Objective Platform, and by Arno Witte, Senior Vice President of Data Science at Objective Platform, here to discuss why having a holistic view of data is so important.
Check out our podcast to find out how to take the first steps toward an automated MMM (Marketing Mix Modeling) strategy that can optimize your marketing campaigns.
Find out how MMM can help you make strategic decisions.
Thank you both for joining me, could you tell us a bit about yourselves?
AW: Yeah. So, my name is Arno Witte, I have a background in Econometrics. At Objective Platform, I’m responsible for basically creating and developing the models we have.
SM: I'm Sven. I'm the CEO of OP. My main focus has been more commercial, helping clients to get most of all out of the platform, and making sure clients are happy with their analytics use cases.
So, what exactly is Objective Platform?
AW: We’re a marketing measurement company. We use our clients’ data to make models that provide insights. We believe in full-funnel measurement, adding brand KPIs, and giving insights into relation to what brands are doing for performance KPIs.
We automate the process and update models for our clients every month to learn about changes, test, and experiment with new things. The main part of OP is the tooling we provide to make historic data predictive for the future.
Why would you say that having a holistic view of data is so important, and how do you translate that into an impact for clients?
SM: Holistic for us means seeing the full picture of your data across all your media channels, not just online but also TV, radio, and out-of-home. The goal is to make data insights accessible across the organization, from the marketers managing a single channel, right up to the big strategic questions from the C-suite.
And Arno, how does that holistic view translate into impact from a data scientist’s perspective?
AW: I think the most important thing for us is that even though we have a lot of experience with modeling, the client knows their business, so it's really important that it's a collaboration.
When we start working with the client, we always start with a blueprint:
- Which KPIs are most important?
- What kind of media are you using?
- What media do you plan to use in the future?
Then we start modeling the KPIs, taking all the media and other factors that influence that KPI into account, like pricing.
After that our job is mainly to update the models frequently and help the client use the models in practice to make better decisions. Our focus is not just on having good media mixed models, but on making them useful for the client.
With all of these models that you're creating, is that why it's so essential to have the data immediately at your fingertips?
AW: Absolutely. For our clients, it's crucial that the models are trustworthy and updated with the most recent insights. It's very important for our clients that the results for their next campaign are there fast, and they don't have to base the next decision on outdated models from a year ago, for example.
How would you say a platform like Meta adds into a client’s marketing mix?
AW: I think that's an interesting question because it really depends on what you're trying to achieve with Meta. We have seen clients successfully build their brands through Meta, and we’ve seen clients use it as a tool to drive performance.
Regardless of how they use the channel, it's essential for us to measure the impact of each campaign at different stages. That way, we can identify if a particular campaign brought traffic to their website or brought conversions for what they were advertising.
What are the details that your teams look at on channels such as Meta?
AW: For the holistic view, it's important that the channels are comparable, so you can look at the metrics coming from your ads, like impressions, reach or clicks. And I think it's even more important that we have the right mappings and the right sort of conventions between the different channels.
So, for example, when you identify the campaigns, it can be really helpful to identify major campaign types, eg. performance-based campaigns, brand-building campaigns, or product communication campaigns. If you do that consistently over all the different channels, including of course Meta, then it’s much easier to compare when you want to make holistic media decisions.
I would always recommend starting with that. Of course, you can deep dive later to explore more granular information points like which devices you targeted.
How can you use naming conventions to compare Meta to other channels to ensure that you are aligned with business KPIs?
AW: Yeah, I think it's a challenge that our clients have, and it's a challenge that we had when we started modeling and updating these models really fast — naming conventions soon became the bottleneck.
There are two ways you can approach this. You can set out your naming conventions and ask everyone internally to follow these — but in reality, we all know it's practically impossible to get everyone to follow all of your recommendations.
The other option is data mapping tools. So basically, all the input data can have a different format, and different conventions, and we use tools to map them all to the same conventions that we use in the models. Streamlining this might be the biggest timesaver in these kinds of processes.
It's better to establish naming conventions at the beginning of a campaign, but it's still possible to retrofit them later. Having mapping tools in place makes it easier to identify discrepancies, and having a feedback loop in place can save a lot of work in the future. However, to keep models and data up to date, it's best to avoid retrofitting and establish naming conventions from the start.
There seems to be a big industry debate between MMM (Marketing Mix Modeling) and MTA (Multi-Touch Attribution), why does it seem that people are leaning towards automated MMM more so than MTA now?
SM: There isn't one good answer to this. It depends on the business questions you have and where you are in terms of data maturity. We do both MMM and MTA, and I still believe that Multi-Touch Attribution can be very powerful for organizations. But what we see now with automated MMM is that it gives more strategic insights.
With automated MMM, you can answer the million-dollar question: do we have to invest more in brands or performance? This is a very strategic question and a very important one.
MMM provides all kinds of opportunities to test campaigns and compare Facebook variables with Diva-E for sales and brand KPIs. Obviously, there is a lot of third-party cookie data missing nowadays, so tracking the future of display is getting more difficult.
The value of MTA can still be very big if you have a high data maturity, but for most marketers, it's easier for them to work with the results of automated MMM, and they can achieve more results with it. That's why we are big fans of automated MMM.
For a marketer that's looking to create an automated MMM model, what are the first steps they need to take?
SM: We need data! The first step is always to determine if there is historic data available. Once data is available, we need to be very clear on the business questions that need answering. Having a clear business question makes it easier to track progress and evaluate the success of the MMM. And we also need people inside the business who can work with the insights provided by the model.
An advocate like that who is familiar with the modeling process can make a big difference. This person can champion the use of data and modeling within the organization, but it's not necessarily an indicator of high data maturity.
AW: Yeah, I think I think you touched on a very important point — of course you need people that understand the data. And I think in general you need to be ready to make decisions based on the outcomes. Once outcomes are available, the journey starts, and you begin translating the model's results into action. The results of the model should be in a format that people can understand and use to make decisions. If the insights aren’t actionable or don't fit into existing business processes, then you’re not going to see any actual impact.
Is it a common pitfall that people aren’t ready to action the insights?
SM: Clients are sometimes afraid of bringing in Marketing Mix Modeling. They’re afraid that the results will reveal negative aspects of previous media investments. However, our approach is different. We work together with clients to help them make the most out of their investments. There are always historical learnings from MMM modeling. We learned a lot during the worst test of MMM modeling, which was the COVID era. Although it was a terrible time, we gained many insights that advertisers can use to optimize media more effectively. It is essential to test and learn with Marketing Mix Modeling.
Are there any trends that you both see that marketing and advertisers should really be focusing on over the next 12 to 18 months?
AW: Meta and Google are already focused on MMM. I think we’ll see marketers take notice of this, and realize that making holistic channel decisions without MMM will be challenging. I think this shift towards MMM will go faster in the coming periods, and this will lead to more focus on upper-funnel behaviors. Automated MMM will become more important, which is why we have been focusing heavily on that part.
SM: Yeah, I think brands will become more critical, not only for bigger advertisers but for smaller companies as well. Rather than focusing only on low-funnel behaviors, companies will steer their mid-funnel to create more value from media.
In some countries, the TV industry is saturated, so it’s necessary to find new ways to drive brands. Some clients are very successful on platforms like Facebook and Instagram, and that’s their route for driving their brand — but are these going to be the future of building a brand? Are there any new platforms coming up in the future?
I think still when you talk about MMM it gives people the feeling of looking at a PowerPoint with a lot of numbers showing exactly what has happened historically. But I think the Predictive path is more interesting for advertisers. Predictive analytics or AI is the future, and we use AI nowadays to calculate the best way to optimize media across channels for different KPIs.
What advice would you want to pass on to someone starting a career in data analytics?
AW: My advice would be to start by listening to the business and understanding its challenges. As an econometrician, I learned that starting with econometrics may not always lead to the best model. Instead, it's important to gather insights from the business about what they expect from the future, which will enable you to adapt to the challenges that are coming up.
Also, I recommend that technical people focus on Bayesian statistics, which is crucial in the current world where modeling is based on all the available data, and there are areas where you may be more certain than others. It's important to adjust to those differences and experiment outcomes. In particular, when blending MTA and MMM into one model, Bayesian statistics play a crucial role.
