Predictive and descriptive Marketing

Predictive marketing

By Didier Gaultier, Head of Coheris Datamining Business Unit

Even though it is largely used, the expression “predictive marketing” is quite restrictive because the best practices in B2C or B2B marketing leverage what we can call “descriptive data mining”. Descriptive data mining includes the fact of doing surveys, polls, descriptive analysis of the marketing databases, segmentations and clustering, with all the subsequent interpretations about the client’s behavior that this can imply.

Nowadays, people’s life has become much denser, much more active, and therefore much more complex. Many opportunities of service have appeared in about every domain. For example, the needs in terms of “personal services” have literally exploded over the last few years. Similarly, in B2B, needs that are specific to each trade have also appeared.

Purchasing behaviors have also diversified in consequence. We used to refer to the “household under 50 years old”. That was a cliché. But nowadays, this cliché doesn’t belong to any reality. For example, in many areas the man/woman factor is rarely a strong enough criterion to differentiate alone the marketing messages while it used to be in the past, except for cosmetic or specialized products.

This evolution has not come without consequences in the world of data mining. The reality has become more complex and we had to adapt ourselves to it. Many criteria which seemed strong, such as the socio-professional category, the age or the gender, which were the main criteria used to define marketing targets in the past, can no longer be used as such today. You can no longer do it, not only for ethical reasons, sometimes, but above all because it doesn’t make anymore sense! For example, a couple with a somehow low range income can come to realize important purchases, even if it means high debts, because it may correspond to a strong need at some time. These punctual behaviors depending on the stage in life do not correspond to the “cliché” type of behavior traditionally expected.

It becomes therefore harder and harder to just use intuition in marketing. A marketing analyst needs landmarks in the middle of all the data available. The role of descriptive data mining is precisely to build a real “map” of the data available.

To make an analogy, would you sail in the middle of the ocean without a good map to get directions? Nowadays, the same question applies with all the marketing databases that are available to the advertiser. We just need to cite a few to be convinced: CRM databases, transactional databases, e-commerce platforms, web analytics, call center databases, email database, SMS, etc. each of these databases (whether they are interconnected or not) can represent a different – but mainly complementary – view of the same customers.

Your map is precisely what descriptive data mining – and only it – gives you. In this respect, you won’t be able to do predictive analysis if you didn’t complete a good descriptive analysis before. It is one of the rules of data mining that is sometimes hard to admit for advertisers: in other words, data must be “understood” before trying to create a model. Going through an often very advanced descriptive stage becomes necessary. This stage itself shall include data management stages that are more or less complex depending on the multiplicity of the sources one wants to process.

Descriptive data mining answers the question “why did it happen?”.

Raw data are unfortunately impossible to understand most of the time. Not only are they very large and composed of numerous different formats, but above all, the information that is really useful can be diluted or completely hidden. You can change this data into information, which give scales, dashboards, ideas … This is all good, but you still cannot understand “why” and it’s usually not enough to act efficiently.

Only descriptive data mining can bring you the tools, sometimes using specific technical knowledge, like in finance or marketing, which will really help you to reveal the critical and actionable knowledge that lies in your databases. This is why, in data mining, we often have to wear a hat of technical expert along with our hat of statistician. The broadcasting and sharing of the results of predictive marketing happens mainly at that level nowadays. That descriptive analysis truly enables the creation of models and all the subsequent predictive computing. Let’s forget right now the wrong, and usually too common idea that only predictive marketing adds value. The opposite is usually true, and in fact, the length of the descriptive data mining stage is sometimes gigantic compared to the length of the predictive stage. It is more or less a Pareto principle: 80% of the time is used doing descriptive data mining and only 20% doing predictive data mining. In the end, the predictive data mining is a bit like the cherry on top of the cake, the last result issued from data mining, or the visible part of the iceberg that is published. But very often, a great share of the important things we learn about our clients, markets, products and campaigns mainly comes from the descriptive part.

Didier Gaultier, Head of Coheris Datamining Business Unit, and Professor of predictive marketing at EPF
Coheris is a leading French Software Vendor for Customer Relations Management, Analytical Management and Predictive Analysis.