How to personalize your client communication with self-learning algorithms.
Free Whitepaper by STAT-UP with a real-life example from the financial industry how to optimize personalized client communication using self-learning algorithms.
Did you receive a marketing-letter from your bank or insurance company recently? I did not get only one letter but 5 (!) with the exact same content within the last 10 weeks. It seemed to me that the insurance company did not understand that the need they addressed was already more than covered on my side (just a quick look into my client profile would have told them so) and thus they spent a lot of money by bombarding me with SPAM. I have to admit that the last 2 letters from that company recognizable as marketing communication went unopened and directly into the dustbin. Every week I receive unwanted letters or emails from different financial companies I hold contracts with and none of them really matches my needs as a customer. Whereas the communication I receive from e-commerce companies for example is highly targeted and personalized. Why so?
The digital transformation brings a multitude of changes and challenges that banks and insurers have to face today. BigTechs such as Google, Apple, Facebook and Amazon, but also FinTechs, are occupying the various interaction points with customers, offering individually tailored, personalized financial services and thus increasing the pressure on the existing players to implement better processes and tools to quickly make highly informed business decisions based on statistical models, machine learning or AI.
In this context, however, banks and insurance companies are not newcomers to dealing with large data sets and complex algorithms. Complex data models are already used in many areas of banking, especially for specific applications in capital markets business, risk modeling, fraud detection in credit card transactions, or digital investment advice. However, comprehensive regulations, the complex internal structure of large institutes and the complex interaction of different departments and processes lead to fragmentation and inconsistency of existing data and thus still prevent the targeted use of AI in many areas.
In times of a pandemic and persistently low interest rates, many classic customer acquisition and retention measures come to nothing. It is therefore all the more important not to lose customers once they have been acquired and to optimize the customer approach in a targeted manner in order to provide them with tailored, suitable offers and to exploit the company's own knowledge of its customers in order to score points over the competition. In most cases, banks and insurance companies still communicate with their existing customers mainly by letter and e-mail. Even if the opening rates are naturally high, since many letters from banks and insurers have an official character and are not initially classified as advertising on the part of the customers, only a fraction of the communication reaches the goal of establishing a customer dialogue and ultimately a new (follow-up) contract. Even worse, if the content of the letter is regularly irrelevant to the customer, the document usually ends up directly in the wastebasket as shown in my personal example in the introduction of this article. As a result, hardly any advertising message reaches its target, as customers begin to manually or automatically sort out unwelcome advertising from their mailbox or e-mail inbox, or unsubscribe from the newsletter. In the worst case, misdirected communication can even lead to customer churn.
To counter this, many companies are introducing marketing tools that are often very expensive and are designed to improve the personalization of the customer approach. The target is "customer centricity". Most of these tools now also work with machine learning and bring their own algorithms for personalization right along with them. This involves several difficulties: on the one hand, it is questionable from a data protection point of view, since consent to processing (which is double in the case of e-mails) must have been given by a third party (the tool provider), and in many cases this is not the case for existing customers. On the other hand, a tool often does not replace the strategy and learnings in the team, which can result from an own development of the algorithms.
In our latest whitepaper, we use a concrete case study to show how the data science team can support existing customer marketing in a privacy-compliant manner while maintaining full control over the AI approaches used. Via a self-learning, automated process, emails and letter sent to customers are to be triggered, with the content of the letters best matching the expected customer interest at the time of dispatch. The open source system presented here includes interfaces for data import and export in order to fit seamlessly into the respective system landscape. The anonymous data is processed in an R or Python instance. A so-called self-learning algorithm is developed that optimizes the calculation over time based on customer reactions and the respective success of the individual measures. Possible pitfalls are pointed out and how to avoid them.
The case study presented in our whitepaper was developed for a bank, but also draws on previous experience from the insurance sector. The approach can also be used analogously for existing customers in other industries (such as automotive retail, energy providers, telephone/internet providers, luxury goods, B2B providers, and many more), where customer retention and up-/cross-selling play a major role.
You can download a free copy of our whitepaper via the link below:
Whitepaper Marketing - STAT-UP
We would be very pleased to receive your feedback and the opportunity for a personal exchange about the methods and models used.