Contact Us

The blog

Discover smart tips, personal stories and our take on the latest marketing trends
26 November 2018

Behind the scenes of data analytics tools & models with Koen Pauwels (2/3)

This second part of our 3-piece article series based on an interview with marketing professor Koen Pauwels focuses on the “HOW”: How do marketers use data analytics and how well are they using their tools? Indeed, today the market is filled with an infinite number of data analytics software and tools (e.g. Google Analytics, Marketo, Hubspot, Adobe, Marketing Cloud etc.) that help marketers better understand their customers and answer their needs. Let’s have a look at what Koen Pauwels thinks about it.

 How do marketers use data analytics and how well are they using their tools?

Do you believe there is a risk that marketers jump too quickly to use the software and tools and do not do their “analytical homework” well enough beforehand?

“Yes, definitely, those tools create a disconnection between the data and the decision-making. Companies should first figure out what they want, what their objective is and what they want to achieve before diving directly into a specific tool.

Marketers usually look at how many people are visiting their website or are looking at their Youtube videos, but this data doesn’t help their company to move forward. This is a key issue.

Let’s have a look at an example:

automotive brand was proud to have the 2nd best automotive Facebook page of the European country, despite having only 1% market share there. They were spending 25% of their marketing budget on their Facebook page. However, the question they should ask themselves is what really drives sales? They know that before buying a car most of their customers ask to test-drive a car. We checked the conversion of their Facebook following to test driving – it was nearly 0%. While the Facebook followers were happy to discuss about the brand on Facebook, they never talked about test-driving. Hence, the challenge of this company was to understand how they could use their main channel to get more people applying to test-drive a car and not invest money just to increase their Facebook performance.”

Should marketers then pay attention before using their models and tools?

Indeed, marketers should understand what models and tools boost them with additional information to their own analysis.

In my course, I usually do a small exercise with my students. Based on raw data, I ask them to choose what a company should keep doing between two marketing actions. My students start without any model and apply some correlations without complicated statistics to this data and come to the conclusion that the company should do one of the actions less and the other action more. Then we discuss and specify a model to regress sales on the two actions to quantitatively separate their impact. When I give them the model’s results, they are delighted it suggest the same direction of change that they themselves recommended. This is called: “face validity”. The next question I ask is what the model tells them more than their own analysis. The answer is about “quantifying”: the model tells them exactly how much you should increase and decrease each marketing action. So the model gives you an objective measurement about how much you should change. Incorporating the quantitative model results into their more qualitative narrative also helps them to better ‘sell’ the recommendation to their boss. Hence, the model is important but marketers should also work on the data before to understand and trust the data, and then challenge the results of the model.

All in all, marketers should first figure out what they want and then bring in a platform, model or tool and see how it fits and what additional insights it could bring them.”

How far do US companies use marketing attribution? 

In the example described previously you explain how important it is for marketers to rely on what they want to achieve and to analyze the touchpoints and actions that make them move towards their objective. In our Yearly Marketing Survey of this year, we deep-dived in the current state of marketing attribution.  We predict for the future that marketing attribution will represent the main challenge for marketing teams. What is your opinion on this?

Companies in the US do a lot of marketing attribution but mostly for their online actions only. Indeed, online actions are easier to track thanks to cookies and insights you can collect about your individual consumers: you could know everything the customer has done until the final online purchase. So when you hear about marketing attribution in the US it is typically only the online attribution model. It is extremely detailed but it is only online. They don’t know to the same extent what the TV commercial or billboard brings them.”

Why don’t they use it yet for their offline actions?

“For the offline actions, companies can’t have the same kind of details in data. However, the problem lies more in the variety of data and less in the volume of data. It is the fact that you have different data sources. So in a typical data project you need to spend more than half of your time to understand how data sources talk to each other.

We should distinguish two models: 

  • Marketing Mix Modeling is used for strategic decisions on budget allocation on all your marketing channels. It is a technique to measure and forecast the impact of various marketing activities. For instance, you know how much you spend on TV and radio, and know how many people talked about your brand or visited your website. These are aggregated numbers. But then I don’t know what customers did individually. Hence, thanks to Marketing Mix Modeling I get the strategic information about how to allocate my resources. You can measure the overall marketing effectiveness and know what is the optimal budget to spend on your different channels (on- and off-line). The aim is to find the best mix between the channels. (You typically use product data/product feature, conversion data, competitive and target audience data.)

    For instance, Uber just validated Google’s ‘Unified App Campaign’ (which allows you to promote your app across Google Search, Google Play, YouTube, Gmail) with a marketing mix model: they regress their overall app usage on their ad spending on these 4 Google platforms, controlling for external factors such as seasonality. Next, they validate the model by completely shutting down spending on 1 channel and observing the impact. Note that you don’t need any individual (consumer) data, which Google withholds due to privacy reasons.

  • Attribution Modeling is used to understand what is the perfect mix of online channels. It typically assumes that everything you do offline it is ‘in the baseline’, it is considered as the environment. Attribution Modeling focuses on the individual consumer’s online actions and which action you should take in response. When you know everything a specific consumer has done online: e.g. what they searched for, which display online they saw, you can use the Attribution Model to tell you, for instance, how much should you bid on the next paid search ad for a specific customer, which email to send him, which landing page to give him, etc.

Marketing Mix Modeling and Attribution Modeling is what companies are working on right now in the US. Being able to combine both is for me the holy grail. All companies want to know everything about what their customers are exposed to, online but also offline. Moreover, they want to know what message to serve to their customer next. At the same time they also want to see how much percentage of their budget they need to spend on display, TV, radio, …”

What about the future? Where will companies focus their attention in the future?

“There is this question that was asked on LinkedIn in 2013, about which marketing principle would still be on our minds ten years from now. There were all kind of buzzwords such as multi-channel marketing, social media, return on marketing investment and so on. For all these topics, less than half of the people thought they would still be a problem in 2023. So people in 2013 thought that we will solve Facebook measurement by 2023. The only topic the majority of respondents would still be worthy of discussion in 2023 was….return on marketing investment. Boldened by this assertion, I named my Twitter account @romimarketer and focused on this topic in my first book ‘It’s not the Size of the Data – It’s how you Use it’, giving a step-by-step guide to measure and improve marketing ROI. Five years later, it turns out the respondents were correct: return on marketing investments remains a challenge.

One of the reasons to explain this is that every year we have new marketing tools and technological evolutions but customer behaviors are evolving as well. The consumers become very smart in playing with us: they go online to find pricing information and then they go to the dealer to have the best negotiation. Everything is going very fast.

I believe attribution is like a moving target. I don’t think we will ever solve it at 100%.”

What are the key lessons?

  • Marketers should first figure out what they want to do with their data before bring it into a model or tool.
  • Companies in the US do a lot of Marketing Attribution, but mostly for their online actions only. The Holy Grail is to combine Marketing Mix Modeling and Attribution Modeling to know everything about what your customers are exposed to, online but also offline.
  • Return on marketing investment is still a moving target and Koen Pauwels does not think we will ever solve it at 100% because we are in an ever-changing market where new marketing tools and technology evolve and customer behaviors are evolving as well.

Want to know more about data analytics?

Wondering what ‘data analytics’ means and what can marketers gain from it? How do marketers use data analytics and how well are they using their tools? And which data skills should marketers have now, but also in the future?

Register for our 3-part article series and discover the full interview with marketing professor Koen Pauwels on data analytics for marketing. 


Want to know more about how to become more data-driven? We got you covered!