How KPN uses Adobe Experience Platform and AMANDA for customer data

How KPN uses Adobe Experience Platform and AMANDA for customer data

The collection and analysis of customer data is becoming increasingly sophisticated. With the help of an Adobe Experience Platform and smart, predictive AI models, KPN wants to improve its customer experiences. KPN works with an enormous amount of data which makes this a challenging job that requires an innovative approach. 

Analyzing KPN customer data is daily business for Stephan, Product Owner Campaigning and Anastasia, Data Scientist and Business Analyst, both working at KPN. Anastasia is currently working with a self-built tool called AMANDA to better analyze data. With this tool, Anastasia maps out the value of customers. Stephan then uses Anastasia's input to streamline marketing campaigns towards customers more efficiently. In principle, the Data Office and Analytics teams work independently of each other, but in practice there is a great deal of collaboration.

Stephan: 'Traditionally, marketing is very push driven. The easiest way for marketers is to send all your customers an e-mail or a letter when, for example, a new internet subscription is available at an attractive price. But that is a bit like a hailshot and hoping you hit something. By implementing and using targeted and automated data, you serve customers better and are more in line with their wishes.’  

‘Even though we work independently of each other, you can see that developers know how to find each other when necessary', Stephan continues. ‘For example, the developers in my team recently worked together with Analytics colleagues from Anastasia to set up an interface between our systems. In addition, we often discuss how we have implemented the data models that are (or will be) used by Analytics. As for AMANDA, here I work with the Product Owner of AMANDA often, and we coordinate a lot during project meetings.' 


The impact and potential of data 
When you think of data, quickly privacy comes to mind. Within the KPN Data Office, nine data principles are maintained, with which the company ensures, for example, that data is compliant. Stephan: 'But the timeliness and reliability of data are also included. Within KPN, the Privacy Office ensures that the privacy of customers is safeguarded. If we want to implement a new type of data use, we always check it with the Privacy Office to ensure that we comply with both the AVG and the customer's expectations. Usually, we are even a bit stricter than the AVG with that.' 

The Adobe Experience Platform is currently being implemented at KPN. It is an open system, suitable for building and managing solutions for various customer experiences. This enables KPN to centralize and standardize its customer data and content from every system. Using data science and machine learning, personalized customer journeys should ultimately be improved.

This sounds wonderful on paper, but in practice KPN has an enormous amount of data and using this data has an impact on millions of people. Setting up the various systems at KPN to get the right offer to the customer is a challenging puzzle, says Anastasia. ‘There are several challenges. The first challenge is the amount of data. There is simply so much data and sometimes we can't process it all. Secondly, a mistake can have huge consequences, so we must triple check everything before we do anything and run lots of experiments to make sure it's right.' 

Then there is also the issue of noisy data; meaningless, meaningless data. How do you deal with that? Some data may not be that important or relevant and recognizing them is a matter of experience. By consulting a lot with colleagues, we learn to recognize them better and better, but smart models can also help us with this,' says Anastasia.

At the same time, the question arises: 'What would be the impact if KPN did not take the trouble to make an appropriate offer to its customers? Anastasia is honest about this: 'The impact is still quite small now, because we are not as far yet as we would like to be with personalization’s. We can do much more, use more data, unlock more sources. That has huge potential. If we do not do those experiments now and learn those lessons now, we will be empty-handed in the future. Then customers will probably get an inappropriate and boring offer that does not suit them at all. Data analytics is mostly very new, but the learning effect is huge. If we build on what we have already learned, we will be able to do much more in the future and that will make our customers happy too.' 
Stephan adds: 'We are now trying to present several generic options for all customers. We do this based on data, but it results in rough options, while we want to be much more specific. Realizing this in all systems is a huge operation, but one that we have embarked on with great conviction. 

One customer data platform​​​​​​​
The Adobe Experience Platform could therefore change a lot in the future. Built on RESTful APIs, it opens the system to developers so that business solutions can be easily integrated using existing tools. Partners can also build and integrate their own products and technologies as needed. In addition, it uses Adobe Sensei, artificial intelligence (AI) and machine learning (ML) technology, to automate tasks, personalize customer experiences and predict customer data. 

Stephan illustrates the functionalities of Adobe Experience with several examples. If you as a customer have just purchased a new digital TV subscription from KPN, it is not logical that a few hours later you will see a banner or receive an email offering the same service. Logically, that customer would not need to see that offer again. Another application is in the event of (service) disruptions, for example. If a customer has looked up a number of things and filled them in online, in an ideal world you would want the helpdesk - provided the customer has given permission of course - to have access to these details. In this way, the customer does not have to explain the same story a second time during an actual customer contact.

Smart, self-teaching models as a multi-armed bandit
Your own tools can be linked to the Adobe Experience Platform, so to speak. Anastasia, for example, is currently working on AMANDA, an advanced model that looks at the characteristics of the customer on the one hand and assesses the context of, for example, an offer on the other, to get the best match between customer and proposition. In other words, AMANDA chooses a campaign that best suits the customer. Anastasia: 'The model needs certain values, and these are in turn determined by other models. All models are constantly learning and thus becoming better at predicting customer behavior, reading the customer context, and linking the right campaigns to it. You want to move towards a situation where people who have just bought a new house, for example, and are about to move, see a moving discount banner because this offer suits their needs. In this, AMANDA decides how (through which channel) and what (what content) to show the customer based on customer context.' The model operates as a multi-armed bandit agent'.  

In marketing terms, a multi-armed bandit solution is a 'smarter' or more complex version of A/B testing that uses machine learning algorithms to dynamically allocate traffic to variations that perform well, while allocating less traffic to variations that perform less well. In theory, multi-armed bandits should produce faster results because there is no need to wait for a single winning variant.

A practical example is when marketers must decide which offers to show a visitor. Without information about the visitor, all click outcomes are unknown. The question is therefore: which offers/banners will get the most conversions? And in what order should they appear? The goal of website campaigns is to maximize engagement, but visitors have different types of content to choose from. Without data analysis and the proper use of that same data, it is difficult to follow a specific strategy. Due to the unknown outcome and associated lower probability of payout, marketing automation using smart AI models is highly desirable. These provide targeted data, maximize the payout, and can then serve up variants with the highest chance of conversion. 

Challenge and satisfaction
Even though the issues surrounding customer data present Stephan and Anastasia with the necessary challenges, the work gives both great satisfactions. 

Anastasia: 'I really like challenges; performing tasks arising from new, difficult questions and gaining new insights. Secondly, having an impact is special. If you work for the consumer market and are also a consumer of KPN yourself, then you experience how our services become better. Knowing that it is partly because of what you have contributed to something, is great.’ 

Stephan: 'It gives me satisfaction to see concrete results from the ideas that I have built up with my team. For example, at one point we took the initiative to build a new application to capture customer leads, as this was being done in all sorts of different places. When that application went live, we saw a huge improvement in the lead processing, but all sorts of new ideas also emerged that were suddenly possible. It's great that KPN gives us the freedom to do something like that.’  


Here's an overview of our vacancies.