Chatbot AI: feedback from the city of Asnières-sur-Seine

The town of Asnières-sur-Seine, located in the Hauts-de-Seine department in the Île-de-France region, called on Wikit to deploy a chatbot based on generative AI on its website and citizen portal. Thomas Tour, IS project manager, looks back at the implementation of this chatbot and its benefits.

Pouvez-vous vous présenter et décrire votre rôle au sein de la ville d’Asnières-sur-Seine ?

Hello, my name is Thomas, and I've been IS project manager with the DISN team at the town of Asnières-sur-Seine for almost two years now. I'm in charge of integrating new solutions into our information system to best meet the needs of municipal services and citizens.

Quelles ont été les raisons qui ont poussé la ville d’Asnières-sur-Seine à envisager le déploiement d’un chatbot basé sur l’IA générative ?

We hadn't specifically identified a need, but we heard about Wikit and its chatbot solution. The idea was born! The prospect of providing a permanent link with residents particularly appealed to our administration, and we decided to embark on a project. We decided to explore different solutions available on the market to assess their added value for our services and our constituents.

Quels étaient les besoins spécifiques que vous souhaitiez adresser avec ce chatbot ?

The first need identified with Wikit was to relieve the pressure on telephone reception in the city's departments , by reducing waiting times for residents. In other words, to facilitate access to information in order to optimize the service provided to our citizens, and to relieve our agents so that they can concentrate on higher value-added tasks.

Pourquoi avez-vous choisi de collaborer avec Wikit pour ce projet ? Qu’est-ce qui a différencié l’entreprise des autres solutions que vous avez étudié ?

First of all, we had a very good impression of the Wikit team, who were available and attentive.

Côté produit, l’interface no-code de la solution Wikit a marqué des points par rapport à sa simplicité de prise en main. La première version du chatbot que nous avons déployé avec Wikit ne s’appuyait pas encore sur l’IA générative et à ce stade, Wikit disposait d’une avance avec des packs de compétences préétablis. Le fait de pouvoir en disposer, nous a permis de nous projeter plus facilement en les consultant afin de pouvoir imaginer les nôtres.

In a second phase, Wikit helped us migrate to a chatbot based on generative AI to transform our classic chatbot into an intelligent assistant. Concerned about the quality of the results delivered, we appreciated the expertise of the Wikit team, who implemented methods to limit AI hallucinations and reinforce the reliability of responses.

Quels ont été les étapes du projet, depuis la conception jusqu’à la mise en service du chatbot ?

First of all, we identified the specific needs and defined the objectives of the chatbot. This involved identifying and consulting various departments. We then selected three suppliers and compared the solutions. We also identified the two communication channels for the chatbot: our citizen portal and the city's future website.

Nous avons ensuite nommé une personne ressource dans chaque service pilote. Une personne au contact des usagers, que ce soit par l’accueil physique ou téléphonique, et en capacité d’identifier les questions les plus fréquemment posées.

We then moved on to the design stage, with weekly updates, which enabled us to maintain the mobilization of our departments while limiting disruption to their day-to-day work.

This was followed by an acceptance phase, so that as many people as possible in the pilot departments could take part in the tests and check the quality of the answers given by the chatbot. Adjustments were made based on user feedback.

Initially, we deployed the bot on an experimental basis on our citizen portal, in order to gauge how well it was received by residents. It was a great success, and we've now deployed the chatbot on the city's new website!

We continue to monitor and improve the chatbot thanks to the data provided by the Wikit solution's administration console.

Comment les citoyens de la ville peuvent-ils interagir avec le chatbot et sur quels sujets ?

The chatbot is available on our citizen portal and our website www.asnieres-sur-seine.fr. With the help of our Wikit project manager, we were able to scrape all the data from our website. Our conversational agent is now able to answer almost all questions (except those concerning events).

Pour vous, quel est l’apport principal de l’IA générative pour le chatbot ?

The first chatbot we deployed with Wikit was not yet based on generative AI. At this stage, the challenge was to mobilize the departments to produce the content and design the word-by-word conversational trees. Given the time-consuming nature of this approach, we initially deliberately limited the bot's scope to three services.

Then, still with Wikit, we migrated to a chatbot based on generative AI (LLM connector), which then enabled us to save precious time and rapidly expand the scope. The chatbot now has the ability to integrate documents and write responses itself.

Today, on the city side, I'd say that the arrival of generative AI has saved us a considerable amount of time! A number of guides and content for citizens already existed. Being able to integrate them into our conversational agent with just a few clicks is a revolution. On the citizen's side, generative AI provides instant access to information without having to spend time browsing the site or poring over official documents.

Convinced of the benefits of generative AI, we are currently working on a second chatbot, this time for municipal employees. This one will have dual skills: IT & HR support.

Quels sont les premiers bénéfices que vous avez observés depuis le déploiement du chatbot ?

Firstly, I'd say that the chatbot has enabled us to gather information on the concerns and needs of end-users. Contrary to what we had imagined, we were able to see that requests relating to the "Etat Civil" came first, ahead of those concerning the "Espace Familles".

Secondly, this information helped us prioritize the missing information to be implemented in the chatbot.

We also observed a large number of requests outside public opening hours, confirming the need for information to be available 24/7.

Quels conseils donneriez-vous à d’autres villes qui envisagent de lancer un projet similaire ?

Get started, test, iterate and adjust!

Beyond deployment, it's important that departments remain mobilized and keep an eye on the solution's administration console. In retrospect, I think it's important to have regular informal discussions, in addition to the more formal quarterly meetings, so that the tool continues to improve.

Y a-t-il des éléments particuliers sur lesquels vous insisteriez pour garantir le succès du déploiement d’un chatbot basé sur l’IA générative ?

Il est essentiel de veiller à la fiabilité de la donnée ! Un point très positif : le projet nous a permis de questionner les processus internes et d’effectuer des ajustements.

Autre point important, lors des phases de recette, il faut faire appel à un large public. Chacun ayant une façon différente d’exprimer son besoin. Depuis la connexion avec les LLM, le travail est grandement facilité et la recette également. Attention toutefois à ne pas négliger le test et travailler le prompt qui formalise la manière dans le chatbot construira ses réponses

Don't miss our next resources