Koen Pauwels is a distinguished marketing professor at Northeastern University and BI Oslo. His passion for data and analytics has led him on the road of research and experiments, which have let him discover new insights in marketing productivity, metrics and social media. Today, we can find his work in numerous marketing journals such as the Journal of Advertising Research, Journal of Interactive Marketing, Journal of Marketing, Journal of Marketing Research, Marketing Science, Management Science and MIT Sloan Management Review.

As an active member of the advisory board of The House of Marketing, we regularly have the privilege of discussing with Koen Pauwels on his knowledge and experience in topics such as big data and analytics, and also the econometric modeling of long-term marketing effects. We recently had the opportunity to ask him a few questions about data analytics and the role data takes or will take within marketing. Keeping all his insights for ourselves would have been selfish, so we are happy to share them with you! This article is the first part of the interview focused on the “WHAT”: What does ‘data analytics’ mean and what can marketers gain from it?

What does data analytics mean for you?

“Data Analytics is about examining data sets to draw conclusions about the information they contain, with the aid of specialized systems and software. In marketing it is the practice of measuring, managing and analyzing all kinds of marketing performances to maximize their effectiveness and optimize return on investment. Understanding marketing analytics allows marketers to be more efficient at their jobs and minimize wasted web marketing euros. It is about discovering what the effect of a marketing action is and how we can improve it.

So, it is not just analyzing the data you possess but it’s really using data to help you make better decisions."

How is data analytics applied in marketing? Or what are the steps toward a data-driven company?

“There are several steps toward data analytics. Typically in data analytics we (1) first want to describe the data patterns:  the descriptive stage. Then (2) we want to predict what it will look like in the future, that is the predictive step. The final step (3) is prescriptive data analysis:  from all the available decision options, which one should I choose? What would be the impact of the action I choose?"

Let’s look at an example:

Together with two other professors at Harvard, Koen Pauwels studied the online behavior of customers of a bank. 

1. First, they were basically looking at an online display and search, which showed that if you get more people to look at your display ads, then you also get more people to search for your bank and then actually apply to the bank. This is the basic description of what customers do online.
2. The next step is to try to predict which product those customers want to buy next. Which one should the bank recommend to them?
3. And finally, they did some prescriptive analysis by giving some workable recommendations based on the data, including how much budget the company should allocate between display and search advertising.

What do US companies do to be or become data-driven?

“The best US companies are incorporating all three of those methods. One great example is Amazon; they optimize everything they show to their customers based on online observations and experiments. But they are also applying much bigger strategic analytics to determine for instance which kind of message they should show to their customers not only online but also offline. Likewise, Microsoft, Facebook and LinkedIn are running hundreds of online experiments a day and report about it at conferences such as MIT’s Conference on Digital Experimentation, every October in Boston.”

What can marketers get out of data analytics? How can they measure the impact of it?

“It is very easy to understand that the analysis of data will help marketers make better decisions to achieve their business objectives. Here are two concrete examples of companies who applied data analytics and who could measure the positive impact of it:

• The first example is about l’Occitane, an international retailer of fragrances and home products. The main kind of action this company did was to send direct mails to their most valuable customers and an email to the low-value customers  (the ones who bought more than two years ago) or prospects. After analyzing all past data covering thousands of consumers of six different countries (e.g. France, UK, Germany and US), they discovered that the company should do exactly the opposite. The high valued customers love the brand and will continue to buy from it; they do not need direct mails (or flyers) because they know the brand and its offer. But for the other customers and prospects a direct mail has a bigger impact because they feel valorized.  By changing this simple action, the company was able to achieve 60% more profit.
  
  
• The second example is about Inofec, a B2B company based in the Netherlands and Belgium that sells office furniture. The problem they encountered is that they couldn’t predict when an office needed furniture. To tackle this, they analyzed their whole business process. Based on the input, analysts built a dashboard model. With the model, they could test different inputs and look at the impact on their predictions. At the end, they could determine which inputs in the model provided the best predictions. The result was a 10% profit increase.

In both examples, the companies used controlled field experiments. This consists of studying the difference in impact between a treatment group  (group on which we apply a certain action) and a control group (not exposed to the action). In both cases the recommendation provided by the model helped the company towards its business objective. For more information, take a look at Koen Pauwels' paper on “Demonstrating the Value of Marketing”."

But what about companies who cannot test or experiment (with a control group)?

“Most of the companies I worked with, at least at a strategic level, don’t want to experiment. Or they just can’t. In that case what I do is to look at their past data, so there is less risk. I look for example at what happened when the company changed its price in the past and based on this I aim to predict, all other things being equal, what will happen if they change their price today. As a result, the company typically changes price in the direction I suggest (up or down), but only to the extent they are comfortable with at the moment. Once the results of that change are in, we analyze the situation to recommend what to do next. In the absence of controlled experiments, we always think about third factors that may drive the relation between price and sales (e.g. new products or quality changes not captured in the data may lead to an observed positive relation between price and sales). A funny example is that Microsoft found out that users that got more Office error messages were less likely to churn (i.e. more likely to stay with the product). Before recommending Microsoft engineers to put more errors into Office, we should first consider that heavy users both value the product more (and are thus less likely to churn) and are more likely to be exposed to error messages. Once you control for usage level in your analysis, the relation between error messages and churn becomes positive.

Can everything always be scientifically proven?

We recently read in an article published by Marketing Week your disagreement with Professor Sharp about his claim that it is “impossible” to quantify how particular perceptions drive sales.  You could show with one of your research papers that he was wrong. Now do you believe everything can always be scientifically proven? 

“I always say  ‘everything that works can be measured, but not always beforehand’. I believe everything should be tested, if not beforehand, then as proof of results afterwards. My main issue with how Sharp’s work is used, is the illusion that your creative ideas do not matter: instead, we find it matters a great deal. Already in our 2009 article in the Journal of Advertising Research, we connected marketing to perception changes and next to sales across many categories. That does not mean that all perceptions matter – we first have to show their ‘sales conversion’ and ‘marketing responsiveness’. The former means that changes to the metric drive ‘hard’ performance (e.g. sales, donations, votes) after some wear-in period we can quantify. The latter means that you as a marketer can actually change the metric with your decisions. My latest article in the Winter Sloan Management Review shows this is the case for online and offline conversations about your brand: they help predict next week’s sales and are in turn to a large extent driven by your marketing actions."

Why do not all the companies use data analytics to its full potential?

“I often hear from companies that ‘their situation is too difficult’ , e.g. they are too small to think about data analytics or don’t have the money to buy expensive technology. I respectfully disagree, in fact a recent study shows that, while larger companies are faster in adopting data analytics, it is the smaller companies that benefit most from doing so. In my experience, the key obstacle in any company is a lack of trust by decision makers in the data and the analysts who aim to share their insights.

The main hurdle for data-driven decisions is not the software or the model’s sophistication.  It’s about the mindset, courage and vision of the people within the company: don’t just hire data scientists but train the whole organization to lower the fear of decision-makers and learn lessons from similar industries abroad.” 

If you are looking for more information on how to become more data-centric, download our Yearly Marketing Survey about data maturity here.

What do you mean by: “there is more trust in the data”? Is it about the quality of the data?

“Bigger companies are spending more on data quality, but that is not what I mean. The following example can help us understand the meaning of trust and its importance within data analytics:

An international company wants to launch a new product on the market and is wondering if they should do a TV commercial and/or social media advertising. Unfortunately, they don’t know if social media will have any additional effect on top of TV advertising. So what they do is field experimentation. They look at the few Eastern European countries that are all similar in size, and in one of the countries they will launch the product with social media and in another without. The results of this test are very interesting for the company, because it gives them very specific data. But you need, of course, to have the capacity to do it.

Now can you imagine what kind of trust this can evolve?

For instance, you have to trust that if you are in one of the control groups (if you don’t get the social media), you will not be punished for lower sales at the end of the year. Usually, nobody wants to be in the control group. Any kind of model or experiment always involves a certain risk. And you need to be ready to take that risk to act on your data analytics. It is that kind of trust that is lacking in some companies.

So I think the main challenge for the decision-makers is to have trust that the whole process will benefit them.”

Could you give us five tips for Belgian marketers to evolve towards a more evidence-based marketing (using data analytics)?

These are my five tips for marketers or data analysts to process data:

1. Determine your hypotheses and assumptions before you start the analysis. What are you trying to achieve? What do you assume about the environment (government, culture), customers, competitors and your own company behavior?
2. Link your metrics in your mind and on paper and think about how data would explain these links convincingly. For example, if you want to demonstrate the impact of website visits on your sales, think about how the data from your analysis would help you to prove it.
3. Do or outsource (with close supervision) the analysis of the data. Question the findings and assumptions until you are satisfied.
4. Communicate the insights and metrics rigorously throughout the organization. Make the data visual.
5. Adapt and learn: prune what is no longer useful, propose new metrics etc.

What are the key lessons?

• Data analytics it is not just analyzing the data you possess but it’s really using data to help you make better decisions.
• Three steps when applying data analytics: descriptive, predictive and prescriptive.

• To determine which decision to make or to analyze the impact of a marketing action, companies can use controlled field experiments or look at their past data. Remember: “Everything that works can be measured, but not always beforehand”.

• The hurdle for some companies is not the software or the model sophistication. It’s about the mindset, courage and vision of the people within the company:  Don’t just hire data scientists but train the full organization about the benefits of applying data analytics.

• The main challenge for the decision-makers is to have trust that the whole process will benefit them.
  
  

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