5/5 (1)

We have all been there: listening to repetitive music, on hold for ten minutes, just wanting to ask a simple question about our bank account. This frustrating reality shows that there is a broken system. For ING, one of the biggest banks in the Netherlands, relying on traditional, physical call centres creates an expensive bottleneck. The average inbound call costs nearly $7 and lasts seven minutes! (ContactBabel, 2024)

To solve this problem, our team asked a simple question: what if you could just ask?

Our proposal is integrating a Voice Assistant that is build on AI into the mobile app of ING. This is a practical solution that helps customers use their voice to book an appointment or block their lost card in an instant, 24/7

Beyond Cost Savings: A Strategic Leap

The financial upside of this product is undeniable. By automating much of the routine calls, the AI assistant saves ING millions of euros annually (Calabrio, 2024). But only looking at the cost reduction misses the real big picture: customer service will turn into a value-creating channel instead of a back-office cost centre.

Instead of having an unnatural text-based chatbot, a voice-first approach just feels natural. Especially for those that are visually impaired or those that simply find typing a hassle.

A Prototype That Proves the Possibilities

Our prototype brings this vision to life. Imagine a customer saying: “My card is lost.” The assistant will immediately confirm which card is lost, and after a quick voice authentication, block the card. It will then proactively ask if the customers wants a new card that will get sent to their address as soon as possible. The Ai isn’t just for emergencies either, it can switch languages mid-conversation if it has to. A user can start talking to the AI in Dutch and ask it to switch to Moroccan Arabic if the user is struggling with the Dutch language.

Built with a Human Touch and European Trust

Of course, none of this matters without trust because of the massive amount of personal data that the AI will have access to. That is why our design is built on three fundamental pillars:

1. Accuracy: With the use of Retrieval-Augmented Generation (RAG), the assistant will only get its information from pre-selected data that is stored in its database. This will drastically reduce the amount of errors (Arslan et al., 2024).
2. Transparency: The AI will clearly identify itself as an AI and the customer will get a transcript of the conversation, if they want.
3. Human Escalation: For issues that are sensitive or complex, a human agent will always be available who will also get a transcript of the previous conversation with the Ai so the user doesn’t have to repeat themselves.

These fundamental pillars will ensure that the AI is in compliance with the EU AI Act and GDPR

By embracing this new approach, ING will build a customer loyalty that lasts and they will position themselves as leaders in the banking industry.

Reference list:

Arslan, M., Munawar, S., & Cruz, C. (2024). Business insights using RAG–LLMs: A review and case study. Journal of Decision Systems, 1–30. https://doi.org/10.1080/12460125.2024.2410040

Calabrio. (2024). Demystifying chatbot containment rates with analytics. https://www.calabrio.com/wfo/contact-center-ai/understanding-chatbot-containment-rates/

ContactBabel. (2024). The 2024 US Contact Center Decision-Makers’ Guide (16th edition). RingCentral. https://assets.ringcentral.com/us/report/us-dmg-2024.pdf

5/5 (2)

I have always struggled with cooking. I would find a promising recipe somewhere on the internet and I would be eager to make it. However, I was immediately thrown off by the comment section on the page. “I baked it for 45 minutes, not 30!” one user said. “It needs twice the garlic,” argued another. To solve this issue I looked up different recipes for the same dish, but on different websites. The result? Even more conflicting opinions.

That’s why I decided to ask AI for help. Instead of looking through thousands of blogposts, I typed a direct and simple prompt: “Give me a simple recipe for a Spaghetti Bolognese. There needs to be numbered and clear instructions with no optional ingredients.”

The result was amazing.

In a few seconds I had a clean and simple recipe, exactly what I was looking for. There was no life story about a trip to Tuscany and how it changed the authors life, or a whole explanation of the authors family history. Just an uncomplicated list of ingredients and easy to follow step-by-step instructions. The AI had made the core principles of the dish into a single guide. No longer did it offer conflicting opinions, it gave me an easy path to follow.

I have discovered that having AI as your kitchen assistant has some issues though. For example, when I asked for a stir-fry sauce recipe, it said that I should use a tablespoon of soy sauce. The brand that I used was particularly salty, so I am glad that I tasted and adjusted before I made the dish inedible.

Nevertheless, the AI has changed cooking for me. I can ask it to explain techniques, convert measurements and adjust recipes for whatever I have left in my fridge. For anyone else that wants to try asking AI for help with recipes, I have one piece of advice: just ask the AI, but make sure you double check everything.

No ratings yet.

Chris Anderson’s (2004) theory of the long tail was revolutionary for its time: the internet would destroy the tyranny of the blockbuster. Our culture wouldn’t be dominated by a few select hits that everyone likes. Instead, online retailers like Netflix and Amazon could succeed by selling a large number of niche products, they would all be sold in small quantities but together they would create a massive market. The long tail was supposed to be a theory wherein every niche could find its audience, if you knew where to look. But those times are gone. The value today isn’t in holding all the niche products, but it’s in making those niche products. We have moved from a long tail to a algorithmic tail.

Look at TikTok for example. Imagine that you have a slight interest in the Harry Potter movies because you just watched some of the movies on TV. In the old long tail model you would have found a list of other generic fantasy films, but now, TikTok’s algorithm will give you something different. Their algorithm doesn’t just see the fantasy tag, they see everything. They see the highly specific elements like magical world-building and British boarding schools. Your ‘for you’ page will turn into content about ‘mystery novels with a magical school setting’ or ‘fantasy with an academic rivalry in it’. Authors will respond to this and will create books and content that fit these categories that were created by the algorithm, which will further reinforce the trend.

This sounds amazing because your niche tastes will get catered to, but is it really that great? We might get put into ‘filter bubbles’. Filter bubbles are personalized filters that can limit the amount of diverse viewpoints that get shown to you (Pariser, 2011). Because of this, the algorithmic tail has the possibility of becoming the world’s biggest echo chamber.

What do you think? Is this new algorithmic tail a great way to find new content that you find amazing, or is it simply handing us a menu of preferences that were chosen for us by a computer.  

References:

Anderson, C. (2004, 1 October). The long tail. WIRED.  https://www.wired.com/2004/10/tail/

Pariser, E. (2011). The Filter Bubble: What the Internet Is Hiding from You. Penguin Press.