4 min.
Relationship marketing in the AI era
1L’art de la gestion de projet2Un projet à succès commence par une bonne gouvernance3Cascade, agilité, demandes de changement?

Relationship marketing in the AI era


In the months to come, marketing specialists will really have their hands full… and their heads. They will simultaneously need to adapt their online campaigns to a world without third-party identifiers and work towards developing more personalized loyalty and relationship marketing programs with their customers’ primary data, all while carefully adapting their practices to new legislation protecting the privacy of users.

Both these scenarios involve a more sophisticated use of company data and MarTech technologies, which is no small task, even for organizations that have more experience with analytics.

During this particularly delicate time, artificial intelligence (AI) is being asked to play a larger role within the practice of marketing, especially in relationship marketing. The reason for this is simple: At the moment, AI is the best way of managing and automating loyalty and personalization programs based on a large volume of data.

It’s possible that, at the moment, your organization doesn’t have a lot of data, or that your company’s analytics culture isn’t developed enough to seriously consider integrating AI into your marketing approach.

If so, it’s normal to instead focus your efforts on improving the management and governance of your primary data. In fact it would even be a wise decision, because it’s impossible to exploit the full potential of AI without a solid strategy and wellmanaged data environment.

However, in the meantime, I’d like to encourage you to fantasize a bit about a possible future for your loyalty and personalization program. In this article, I’ll present two types of machine learning models that could be easily merged with your relationship strategy (and effectively complete it) in addition to recommending a few interesting avenues you might like to take in the future, when the time comes to integrate AI into your practices.

Segmentation models

At many companies, the most popular segmentation models are RFM (recency, frequency, monetary) and LTV (lifetime value). Of course there are also many other segmentation models that have been created based on product age, type and preferences as well as many other factors.

While those models are a good start, it’s now possible to go even further. Thanks to data science, you can segment your customer data according to many variables, all at the same time.

One of the most often employed techniques currently used to accomplish this is clustering. By combining unsupervised algorithms, this strategy allows you to simultaneously evaluate several variables and discover new segmentation rationales that are adapted to your particular customers. Once accomplished, companies will be able to identify differentiating factors they had never even considered before, ones that are adapted to their customer, industry and business contexts.



Among the most popular segmentation algorithms is decidedly K-means, but there are a number of others inventoried in the scientific literature. Clustering models aren’t new: Customer relationship management (or CRM) teams have been using them for years.

However, the context is quickly evolving, particularly due to the volume and diversity of data available today for this type of analysis. For example, it’s increasingly common to cross-reference and unify CRM data with behavioural data collected on an analytics platform (such as those from Google or Adobe). It’s also possible to add all the information gathered from email platforms into the mix. This means a company could find itself holding tens or even hundreds of millions of entries in its data warehouse, spread across scores or hundreds of columns, containing a very large volume of information about customers.

A model like RFM, for example, is simply not capable of adequately exploiting a volume of data with this degree of complexity. However, AI segmentation models were created specifically for this context and allow you to get the most out of a collection of primary data.

Note that in a series of future articles, the team at Adviso will cover the various AI segmentation techniques and advanced analytics available in greater depth.

Personalization models

Once you’ve segmented your customers, it’s time to move to the second step: your personalization strategy. If your segments are well-integrated into your various marketing platforms, you can communicate with them based on their specific, respective needs.

But how do you discover your customers’ needs? For example, if you maintain customer service records and are able to cross-reference them with each segment identified by your algorithms, how can you analyze your customers’ communications across millions of records? Are you going to ask a team of new recruits to manually go through every email, text message, chat session and call centre recording? Not very likely.

For this task, breakthroughs in deep learning can help, especially the advancements in language understanding and generation (specifically the artificial neuron networks known as transformers) that have occurred in the past five years. These new technologies are extremely powerful. It was these advances that led to the very popular GPT-3 model created by OpenAI. This same technology also gave birth to Google’s BERT, which has been revolutionizing the way the search engine interprets user search requests since 2018.

Without getting too deep into the technical weeds, let’s just say that BERT’s and GPT-3’s secret sauce is what’s called their attention mechanism. Thanks to this layer, these models are able to attentively focus on each word of a sentence, then understand its meaning within a particular linguistic context.

The diagram below provides an overview of GPT-3’s architecture, which has achieved a level of understanding of human language never before seen in the history of artificial intelligence.


Papers with Code


How can a model like BERT or GPT-3 contribute to your personalization program?

Since these models are open source, any data scientist or machine learning engineer can access and adapt them to the context of a particular company.

That said, although there are a number of different ways they can be put to use, there are two highly complementary use cases I’d like to highlight in this article.

  1. Semantic analysis of customer service data

This essentially involves compiling the thousands or millions of records of customer communications (chat, call centre, email, etc.) and analyzing them using natural language processing in order to categorize your customers’ issues, needs, expectations and dissatisfactions.

Following this categorization to make the data more “digestible” for the human brain, it will be possible to assign communication attributes to your various segments (as long as these communications can be attributed to customer identifiers). With this important information in hand, your relationship marketing experts will be able to design a new personalization strategy related to each identified issue and need. This will make your loyalty program increasingly relevant and personal.

2Construction of a REAL chatbot

In recent years, before the emergence of GPT-3 and BERT, chatbots developed a bad reputation. Business chatbots are often not based on artificial intelligence, but on an algorithm built upon a collection of linear rules that are often too simplistic.

But today this will likely change. With models like GPT-3 (and others that are still emerging and even more powerful), a team of data scientists can now train a much more sophisticated chatbot, particularly one that is better adapted to the reality of your business and your customers.

It is now possible to train a model equipped with the full power of a transformer neural network, but still built on your primary data from your various customer service channels. Since customers are increasingly expecting 24/7 access to their brands, AI holds promise as a way of improving customer experience and reducing operating costs.



On the whole, these examples show that artificial intelligence, when combined with a good primary data strategy, can increase your company’s productivity, starting with the effectiveness of your efforts invested in marketing. However, transitioning to this technology requires developing a company culture that is authentically oriented towards an intelligent use of data.

A recent study led by PwC, covered in an article in AiThority, shows to what extent companies fully invested in AI reap greater benefits than those making only partial investments. The study also concludes that customer experience is the area that benefits most from the integration of AI into relationship marketing.



Naturally, the selection of AI use cases for relationship marketing covered in this article won’t materialize tomorrow. They require a long-term road map, or even a five-year plan with several phases. Still, you should keep in mind the famous dictum:

The journey of a thousand miles begins with a single step.

Lao Tzu

Would you like to discuss taking your first steps in AI or beginning to design a future relationship marketing and primary data strategy? Don’t hesitate to get in touch. Our team will be happy to help drive your business towards new achievements.

Join our discussion