AI, machine learning, reinforcement learning… Into the heart of predictive marketing

Written by Zoé Baillieux on Feb 17, 2020 12:03:20 PM

Today, ever more relevant exploitation of Big Data is made possible thanks to Artificial Intelligence techniques such as Machine Learning or Reinforcement Learning ...They are a critical part of predictive marketing within the Advalo platform. Encounter and decode with Nicolas, Data Scientist at Advalo.


"These techniques transform mass marketing into a more acceptable personalized relationship that makes more sense to the consumer. A real relationship, such as a store salesperson could have with his customers 50 years ago." Nicolas, Data Scientist at Advalo


Can you tell us more about your job?

A Data Scientist is responsible for the management, analysis, and exploitation of Big data. It's a bit like being an extension of a statistician in the era of Big Data. Specifically my job consists of making use of state of the art AI (Artificial Intelligence) and all its techniques to decide if it can be used in specific marketing cases. There is therefore a large part of R&D with the adaptation of open source algorithms (from research and developer sharing) to create exploitable models for our retail and automotive customers. When will an individual who owns a vehicle think of changing models? How to detect if a client intends to buy a new pair of jeans according to their internet browsing, their likes on social networks and even their travels IRL (In Real Life)… AI allows us to answer precisely these kinds of questions : key moments, finding new insights. In addition to their expertise, a good Data Scientist must have a multidisciplinary approach taking into account the constraints of the field in which they work and must excel in being able to communicate results to non-experts.

So AI is the new buzz word in digital marketing, but what are we talking about exactly?

To give a global definition, I would say that AI is the set of theories and techniques used to make machines capable of simulating human intelligence. That said, we can clearly distinguish strong AI from weak AI:

Strong AI seeks to draw near human cognitive capacities (as in Science Fiction films), it is a good target from a theoretical research point of view but it remains a distant dream despite the many advances made . Weak AI, on the other hand, is used from an operational point of view, as in the context of digital marketing: The idea is to put into place ultra-specialized algorithms for a specific task for operating a device which exploits Big Data in total autonomy.

What is Machine Learning ?

Machine Learning is an AI technology that allows computers to learn without having been explicitly programmed for it. This means a very large set of algorithms with learning which can range from a simple mathematical function, some parameters must be optimized to multi-layer neural networks (“Deep Learning”) where billions of parameters can be calibrated by the machine during learning. We can for example set up an algorithm which is capable of estimating the fair value of an apartment from different information that it will enrich itself (surface, antiquity, facilities, municipality, district, distance from schools, transport and shops, values of similar assets, etc.).

Machine Learning actually covers 3 categories: supervised learning, unsupervised learning and reinforcement learning. Of course in the fields of digital marketing, these different techniques complement or even mix.

What is Reinforcement Learning ?

Here and unlike Machine Learning (supervised or unsupervised), the dataset (associations between observations and variables) on which the algorithm will work is not initially given: it will be discovered as the agent interacts with its environment. I can give you a concrete example with the one-armed bandit algorithm which can be seen as an extension of the classic A / B testing: instead of waiting for the end of the test period to draw a conclusion, we can both "exploit" (make the optimal decision as learned at a time T) and "explore" (test suboptimal options, to continue learning): it is therefore possible during the course of an emailing campaign to test several combinations of email content, objects, offers, etc. until you get to the most effective mix.

How do these techniques intervene in the Advalo platform ?

They are at the heart of the solutions we offer since they make it possible to set up predictive models. The primary step involves reconciling offline and online data to create a "classic" DMP. We can then obtain an omnichannel vision thanks to pure statistical techniques (Business Intelligence, Analytics, Datamining ...) ... but without an automated model, it is simply an "image" taken at a specific time. With the help of AI, we can really take action in real time to optimize targeting, offers, campaigns and find consumer insights: this is individualized and above all automated marketing. Note that these techniques can also be used in a less visible way, similar to the enrichment of our product repositories using NLP (Natural Language Processing - the ability of a program to understand human language) or Deep Learning (for example the implementation of product image recognition).

How have these technological advances helped brands ?

I think that the contributions are numerous. Take the example of the automobile: the different models using the above techniques allow us to set up intention scores to purchase a new vehicle, detect churn (and prevent it), predict the mileage of a vehicle, its maintenance, the ideal time to renew it, set up product recommendation engines and imminently optimize campaigns for maximum gain. All this allows the brand to reduce unnecessary spending on AdWords, Facebook, etc. by targeting consumers more precisely (stronger commercial impact for a reduced marketing budget). We then work on real “life moments” of prospects and customers of the brand to offer them the right product with the right offer, at the right time and on the right channel…

On the other hand, the automation in the production process and distribution of a campaign makes it possible to reduce purely operational tasks: an algorithm can make it possible to manage, for example, 1000 targeting for the same campaign which would be impossible for a human. A real gain in efficiency and time that allows customer marketing teams to focus on strategy.

Is this a step forward for the end consumer?

The algorithms allow better targeting for when to speak to the consumer, what products to recommend (recommendation engine), via which channel (email, Facebook, sms, ...).

This transforms mass marketing into a personalized relationship as a more acceptable form and that makes more sense for the consumer. A relationship much like one that a store salesperson could have with their customers 50 years ago. In the long run and if done properly, the consumer wins since we no longer send them messages that do not interest him, we even save the consumer time by offering them what they need when they need it.

What alternative applications can you think of for tomorrow ?

I think it will be a continuation of what we do today with even more performance and relevance: better targeting, better recommendations, optimization of increasingly advanced campaigns. We can then imagine applying it to other fields. For example, the “creation” part could also be managed by AI which would offer an email template, an iconographic and optimal wording per target, or ultra personalized advertising spots with hundreds of automatically generated versions… The future still has great surprises in store in terms of predictive marketing!


Topics: Predictive Marketing, Exclusive interviews

Back to Home Page