Back to case studies

QBox helps us keep our customers happy by greatly improving the quality of our NLP model, resulting in a better chatbot and a higher success rate for our bot. And a better bot leads to better KPIs and happier customers.

— Jeffrey De Meulemeester, Digital Assistants Manager at SNCB

SNCB is the national railway company of Belgium.

As a company, it manages large quantities of customer queries daily regarding its services. Its chatbot, Mobi, greets customers on Facebook Messenger and WhatsApp, helping to answer questions about the services currently running and possible disruptions to their journey.


The team use a chatbot to answer the increasing volumes of queries coming through social media. Answering questions in English, Dutch and French, Mobi was launched to reduce the workload of the social media care team. The chatbot currently covers around 160 intents and gives around 500 answers per language. Mobi handles around 20,000 queries a month and enables customers to access support 24/7 rather than during the working day of a support agent.


What drove the need for change?

  • Increased volume of queries post pandemic
  • Need to understand the effectiveness of fixes
  • Lack of users with an IT background
  • Difficulty identifying the origin of the training phrases causing conflicts

Prior to QBox, the team were using a simple tool made by their own developers. With the limited usefulness it offered, there was a real need to delve more into the detail and establish whether their fixes were having the intended effect or causing more regressions.


After spending a lot of time making minor changes to improve performance, the team were working hard on upgrades but lacked a way to troubleshoot them or test the overall performance. Having sought out software to fill the gap and consulting with multiple providers, QBox was clearly the solution.

Streamlining with QBox

“QBox is like a satnav, but for NLP improvements. You could do without it, but you’re using outdated maps – and navigating the ever-growing jungle of conflicts between intents and training phrases is difficult and very time-consuming.”

Tim Lambrechts, Chatbot Specialist at SNCB

He continues: “Using QBox, we’ve easily increased our correctness percentage by around 20% in just a couple of months.

“Because QBox points out which intents need work, and, more specifically, which training phrases are causing conflicts, we immediately know where to start working. This saves us a lot of time, which we can use to expand our chatbot’s knowledge or improve the existing intents even further.


The work we can do in an hour with QBox probably took us a full working day before.


“These improvements [have] significantly lowered our fallback rate, so more customers now get a correct response. By adding more intents, testing their performance, keeping an eye on the performance of existing intents, and slightly lowering our confidence threshold (which we could analyse through QBox), the fallback rate [has gone] from 20–25% to 10–15%.

“QBox is an especially helpful tool for the SNCB team as most of the colleagues working with it don’t have a background working in IT. For our newest colleague, only some basic NLP model explanation was necessary before she could get to work with QBox, as the tool explained the rest.

“The way QBox represents data is clear and not too crowded, while not omitting the details. The diagrams are a clever way of showing conflicts, and very effective in giving a clear indication of what to start working on.”

Advice to other organisations on whether to make the move to QBox

“If you’re serious about your chatbot and investing in improving it, then yes.”

See QBox in action

If you work with natural-language data models and you’re looking to quickly and easily understand, analyse, and improve the performance and results of chatbots and conversational AI platforms, QBox is the tool for you and we will show you why.

See a demo