The Challenge

Our client is a leading global meal kit provider. It analyses customer sentiment for continuous operational improvement. However, the solution it uses for this purpose requires feedback to be analysed in English.

The challenge was to build a customer sentiment analysis model for comments in Dutch, which categorises the comments into a list of categories used by our client before analysing the sentiment.

Understanding Market Dynamics

For our client, the Dutch-speaking market is one of the largest markets without English as its primary language. Since the standard solution used by our client only analyses comments in English, comments from Dutch customers had to be translated into English before being analysed, which resulted in the analysis performance being “lost in translation”. Therefore, the problem to be solved was to develop a Native Dutch Language Customer Sentiment Analysis model that outperformed the solution already in use for Dutch comments.

The Approach

In six sprints of two weeks each, we developed a custom model that:

  1. Identifies which categories the comment belongs to from a list of about 70 categories used by our client for sorting and prioritising purposes, such as flavour, variety, delivery, etc.
  2. Analyses the sentiment of customer feedback by identified category, whether it’s positive, negative or neutral.

For example, below is a sample of the results for a specific customer comment:

Input: “Is heel gezond en lekker en scheelt zoveel tijd!”

Output: (Flavour -> positive), (Healthiness -> positive), (Time -> positive).

The Solution

We developed an end-to-end SaaS solution that outperforms the existing solution for Dutch comments. To sum up, the solution:

  1. is accessible through any web browser;
  2. notifies the end user once the analysis has taken place and the results are available;
  3. provides insights into the use of the solution.

Impact and Benefits for Our Client

As a result of our work, the client is able to prioritise improvement initiatives more efficiently based on a more accurate sentiment analysis of the feedback from the Dutch-speaking customers. Moreover, this leads to a better cooperation between the central Voice of Customer department and the local operational teams.

“We have enjoyed working with Amsterdam Data Collective in developing and rolling out a comment categorization model for Dutch language feedback. They worked closely with us to understand our challenges and improve the model’s accuracy in several iterations. ADC’s solution allows us to analyse customer comments in their native language so we can skip the translation step and evaluate them exactly as they were written. Our local teams need to be confident in the data we provide so they can plan and prioritise their efforts to improve the customer experience, and this model helps us achieve that.”

Head of Voice of Customer

Way of Working

For this project, we worked in six sprints of two weeks per sprint in order to deliver the project on time and within the budget. Furthermore, our ways of working included:

  • a kick-off and interim analysis meeting;
  • an Agile approach, working in sprints to set priorities and manage expectations;
  • a live testing phase for final acceptance.

Curious what we can do for your organisation?

Would you like to know more about this case or what Amsterdam Data Collective can do for you? Get in touch with Rik van der Woerdt at, or check our contact page.