Van der Valk Hotel: Voice AI technology for hotel customers

Van der valk is the largest hotel chain of The Netherlands with more than sixty hotels in the Benelux. They are known for their excellent customer service and always strive to improve their service to maintain their service level. Van der valk always keeps track of the latest technologies which can improve customer satisfaction or can help making operational processes more efficient. Van der Valk discussed with us the opportunities voice technology could have for their customer facing and operational processes. This resulted in very challening and innovative project in which Van der Valk decided to bring voice technology to their customers.

We first made a detailed analysis to see where voice technology could fit in into their processes. After discussing and analyzing different potential processes We advised to first focus on customer facing applications, since customer facing voice technology applications where more clear and better outlined than operational processes. We decided to focus on the application of voice technology through which hotel customers are given the option to submit their feedback via voice technology. Giving feedback via voice is a very interesting and helpful way for customers to give their feedback. Since people can speak 4 to 5 times faster than they are able type, they are able to transfer much more information in the same amount of time. Futhermore since voice/speech is the most natural way of communcication for people, it is actually way more natural for people to give feedback by just telling it, than typying the feedback into a form field on a website. Hence allowing customers to give their feedback via their own voice would also be much more pleasant and non-invasive for customers than old school methods.

After having outlined the capabalitites the voice platform must adhere to, w started building the platform for them, which consists out of the following parts:

1. A user interface for hotel customers to give feedback via speech

2. The API which receives the voice feedback and transforms the feedback in order to create deep analyses on the feedback

3. Several deep learning models to classify the incoming sentiment of the voice feedback, classify entities of the incoming feedback and lastly an algorithm to classify entity sentiment to discover the sentiment of hotel amenities or services automatically

The user interface is off course a necessary part since hotel customers have to be able to give their feedback via an interface. We decided together with Van der valk this interace would be integrated into their current guest application, so customers could leave their feedback by just telling it via an interface integrated into the guest app. Since many guests were already using this guest app, integrating it into this app was logical step.

Secondly we had to create an API which could receive the speech feedback. This was more of a software engineerning task than a machine learning task but nonetheless an important part of solving the overall problem.

The last step was the most interesting part: creating the artificial intelligence algorithms extracting value from the incoming voice feedback from customers. Because of confidentially we can not tell you in which way we transformed and prepared the data before being used as input into the algorithms, but we are able to share the different models we used to extract insights. These models are as outlined above:

1. Sentiment classification of incoming voice feedback

2. Entity classificiation of incoming voice feedback

3. Entity sentiment analysis to discover the sentiment of hotel amenities and services automatically

These three models give all very useful insights into customer feedback for Van der Valk. The first model classifies the sentiment of incoming feedback. By using this model Van der Valk can quickly get insights into the overall customer satisfaction and resolve complaints very quickly while customers are still present in the hotel.

The second model classifies entities of the incoming feedback. The entity classifcation model can for example detect what a customer is talking about in his or her feedback automatically. By using this model Van der Valk can quickly learn what customers are giving feedback about. Subsequently they can focus on th entities which need attention. Entities are concepts, such as for example the hotel service level, restraunt or hotel rooms.

The last model we trained is a very interesting model and combines the first two models. This model is able to classify the sentiment of entities. For example when a guest complains about his room, this model can automatically detect the sentiment of his feedback about his room. Since guests can mention more than one entity in their feedback, this model gives Van der Valk fine grained analyses of hotel hotel amenities and services. This is very valuable for Van der Valk.

Van der Valk was very satisfied by the results of our work. The total voice technology solution is now operating in Van der Valk hotels. Currently we continously monitor the models in order to improve the models even more. Currently we are also investigating how we can scale this solution even further since many more interesting voice technology use cases exists in the hospitality sector.