AI – great opportunities but ‘Handle with Care’. Takeaways from the N-sight of 02 April 2019


Do you hear that buzz? Artificial Intelligence and Machine Learning technologies are firmly in line as the next step in digitising our organisations, streamlining our processes and improving our customer experience.


At our N-sight, we took a look at the technology, how companies are already making it a part of their strategy today, and  also heard about the potential pitfalls. Getting into specifics, we received an overview of market trends in customer contact centres, while in real-life cases from Digipolis Antwerp, VDAB and UCB we learned how AI is already bringing added value to some organisations. Presentations are available to Beltug members (after log-in):




Customer added-value when answer bots and customer contact employees work together


Hans Cleemput, Director at Customer Contact (the Belgian Customer Contact Association) opened by explaining that Belgium is often quite reluctant in adopting new technologies, yet the Belgian contact centre sector is a front-runner in the adoption of new opportunities, including bots. To give a few examples: IVR (Interactive Voice Response) is expected to grow by a factor of 2.6 over the next decade, chatbots will grow by a factor of 26 factor, and contact centre interactions will grow by a factor of 3.5.


Regarding the adoption of AI and chatbots, companies need to ask themselves about their customers’ expectations, which are closely related to the benefits that come with the technology. Hans doesn't see AI as a threat to the contact centre sector; he feels chatbots and humans will closely collaborate in the contact centre of the future.


The current job of the customer contact agent/call centre agent can be quite unattractive in today's job market, but  collaboration between human and bot might well increase the appeal. Hans described an experiment at Proximus, where chatbots are paired with agents: sometimes, customers can't detect if they are conversing with a bot or a human agent. But you need to ensure your customer knows who he has on the phone. And you need to clearly define the bot's tasks and limitations, and where the human agent comes in. This is often for more empathic, complex issues.


When designing and mapping the conversation at a contact centre, chatbot and human tasks can be integrated.  The agents are thus the AI trainers.  At KBC, it was in fact young customers who trained the chatbots - a much more efficient approach than training through keywords.


Customer satisfaction in the conversation with bots is mostly connected to the bots reliably providing the right answer – as long as the conversation has been properly designed, of course. Hans wrapped up with a few takeaways (slide 33).



Case: Antwerp City Platform as a Service - the key to an optimal digital experience


Tobias Verbist, Chief Architecture & Innovation at Digipolis Antwerpen then presented the Antwerp City Platform as a Service (ACPaaS).  First Tobias sketched the previous, very slow and very inefficient new project development process.  Then, Digipolis set up the ACPaaS based on the main principle of having engines manage the APIs (slide 2).


At the start of the project, Digipolis gathered all the applications in one infrastructure, created a backlog and began building the platform.


Fast, cheap and innovative - these are the main benefits of the centralised platform (slide 5). The tendering process has sped up, too.  Today the ACPaaS has some 60 components, 100 running applications, and 400 transactions per second (at peak times), unleashing the power of APIs to the fullest: from apps for the Social Service Department to apps for booking meeting rooms, to payment modules, etc.


The combination of chatbots and AI can lead to powerful results: chatbots can begin the conversation with a context head-start. Tobias ran us through the example of waste recognition and how AI supports the reporting app for citizens.


Digipolis also offers services for other cities, for instance digitisation of the street furniture inventory. OCR is also applied in numerous ways - for antique museum books, license plate recognition, etc. Numerous examples and opportunities were reviewed. And future projects include services such as customised cultural portfolios for individuals.



Pitfalls in AI


Joachim Ganseman, Research consultant at Smals, laid out some potential pitfalls in AI.  You can indeed mess up your own AI system, he began. For starters, be careful of the data and data sets! As always, 'garbage in means garbage out', he warned.  Make sure to sample your training data well, and thoroughly analyse your (statistical) data.


Biased humans also collect biased data, and these biased data result in biased algorithms.  Not all bias is unfair, Joachim commented. For example, prostate cancer data is obviously biased towards men. But unfair bias can have serious consequences on legal, economic or security decisions. So, know your data!


An AI system tries to maximise an objective, yet 'success' is often hard and complex to define correctly.

  • AI exploits bugs
  • AI exploits unexpected data properties
  • AI gets stuck in endless loops.


On YouTube for instance, the producer of a video has different objectives than the consumer.  AI is only as reliable as the data it’s trained on, and it maximises its objective, nothing more.


Even when your data set is excellent and unbiased, it can still get poisoned. Companies need to be aware that no AI system is 100% accurate - always have a back-up system in place!


A 3rd pitfall is the abuse of AI, for example for 'spear fishing' (fraudulent attempts to obtain sensitive information, directed at a specific individual/company, or for fake news, fake text and fake images (slides 24 – 36). Keep a firm hold on your common sense: if something sounds too good to be true, it probably is.


What do you need to avoid these pitfalls? For starters:

  • Transparency & explainability
  • Digital scepticism
  • Policy

(slides 36 - 45).



Case: AI @ VDAB


Three years ago, VDAB started its AI journey, described Paul Van Brabant, Solutioning Architectuur & Innovatie at VDAB. Today, VDAB is in the exploitation and production phase of this journey, with different projects in the running.  Jobnet is one of these projects, embedded in the VDAB solutions for the past year.


With Jobnet, VDAB aims to match job vacancies to applicants at one end, and job applicants to prospective employers at the other. This matching has been done using software applications for several years at VDAB, adding AI in the last year.


Via a deep learning mechanism, all complex data are matched and come out with a 'matching score’. AI enables an even broader matching ability (e.g. matching 'nanny' with 'kinderoppas').  VDAB worked with AWS to build the Jobnet solution (slide 8). Today, Jobnet is intensively used: over 3 million people receive job suggestions.


Paul zoomed in on the lessons learned from developing Jobnet: 

  • Computing power: training the model requires proper resources
  • Cloud offers the necessary flexibility - an on-premise solution couldn't have done the job
  • Avoid bias in your data

 (slide 10).


These AI projects are supporting VDAB’s evolution to a data-driven organisation. To wrap up, Paul touched upon the ethics of going digital (slide 15).



Artificial Intelligence at UCB: examples, trends and potential


How 'artificial' is AI? This was the question Sammy Sambu, Associate Director, AI Solutions at UCB used to trigger the audience. Is it 'artificial' because it sits in a machine, he probed?


In the pharmaceuticals environment, microscopy is an area that is highly suitable for automation, with numerous repetitive actions. For example, there are many types of biological constructs called organoids. By teaching AI what an organoid should look like, it can support the development of your next blockbuster drug.


Transformation is another area where AI is used at UCB. Presentation of diseases or conditions looks different from patient to patient. When it understands what exactly a disease or condition is, AI can help.  For instance, in the detection of spinal fractures (slide 9), AI can support the work of the radiologist by analysing and comparing images, and then soliciting the advice of the right doctor.


But how can we trust all this - especially when there is conflicting information, for instance?  Cooperation with human expertise is a critical part of the UCB success story.  Trust is created through 'trust by privacy' and awareness of data risks. 


UCB went through a 10-year journey to arrive at today's AI strategy (slide 17). Evangelisation was an important factor: all stakeholders needed to be on board.


The discussions after the presentations were dynamic. As one attendee commented – simultaneously impressed by the opportunities and the pitfalls - "We might compare AI to a nuclear power plant - it brings fantastic opportunities, yet we need to handle with care!"


















Dear visitor,

Access to more information about this topic and/or to download the paper is easy and fast, but exclusively for Beltug members (just login to get access).

Beltug gathers a lot of information. Here you find the advantages of Beltug membership

The Beltug Team

Click here to login

>>> Back to overview