Structuring feedback processing to elevate the guest experience
Pairi Daiza welcomed nearly 2.9 million visitors in 2025, while hotel guests shared feedback on different parts of their experience. Those responses covered everything from animal enclosures to hotels, restaurants, and services. As feedback volumes grew, the team found it harder to read, classify, and identify recurring themes across responses.
The challenge
Pairi Daiza welcomes millions of visitors every year, and as the park expanded its accommodation capacity, attendance climbed to nearly 2.9 million visitors in 2025. Every one of those hotel guests could leave feedback through a post-visit survey, and the support team had to make sense of all of it.
Each comment could carry more than one message at once, since a visitor might praise the staff in one line and complain about the restaurant in the next, and there was no way to predict whether it would arrive in French, Dutch, or English, meaning every response had to be read in full before the team could understand what it actually covered.
Every additional response meant more reading and tagging by hand, and Pairi Daiza needed a way to handle that growing volume without losing the detail behind each comment or the attention that kept guest experience at the centre of the park's growth.
The solution
Our team built a generative AI assistant and integrated it into Pairi Daiza's existing Freshdesk environment, designed to take on the reading and sorting of visitor feedback so the support team no longer had to do it by hand.
Every response is read for its intent, since a single comment can carry both praise and a complaint. Each complaint is then tagged by what it's about, the specific issue, how serious it is, and where in the park it happened.
Once sorted, feedback is automatically redirected to the right team, and every reply the assistant drafts still passes through a support agent before it's sent, so nothing reaches a visitor unchecked.
That shift resulted in a 25% reduction in manual feedback processing time for the customer support team. Corrective actions now reach the right team faster, with staff spending less time sorting information and more time on guest follow-up.
The approach
Our team began by studying how Pairi Daiza’s support team handled visitor feedback in Freshdesk. This gave the team a shared basis for defining what counted as a complaint, which languages the assistant needed to recognise, and how the park’s 400-plus existing tags could be replaced with a clearer taxonomy.
Pairi Daiza’s internal teams helped shape that taxonomy and validate the rules behind it. Together, the teams defined the feedback types the assistant needed to recognise, including compliments, remarks, disappointments, and complaints, as well as the topics, subtopics, criteria, and locations used to classify them.
Those definitions also guided how replies were constructed. Depending on the survey response, assigned tags, and available visitor information, the assistant could adapt an existing template or generate a new draft for the support team to review.
A blended team spanning project management, data and machine learning engineering, and software engineering developed the assistant alongside Pairi Daiza’s IT director. They worked in short sprints, giving the support team a working version to test at each stage and using their feedback to shape the next iteration.
