Current results and lessons learned from two chatbot implementations at Shell

Erik Bakhuijs, IT Advisor - Chatbots Shell & Muriël Serrurier Schepper, Senior Project Manager Chatbots at Shell during the implementation, quickly shared the results and their lessons learned with 2 Chatbot implementations at Shell. Shell started, entirely from scratch, to explore the possibilities of setting up a chatbot for internal use at Real Estate Services. During the preparations, they learned that an identical process was followed for Ethics & Compliance elsewhere in the organization. It was then decided to use the learnings of the most "simple" chatbot for the more complex variant. The following steps were taken to get both chatbots up and running:

1. Identify the goals and work processes that the chatbot will use
2. Add statements to the goals set to enable the chatbot to understand more
3. Create entities in the chatbot
4. Record the answers in a pleasant, clear and consistent manner. Always involve a conversational writer in this process and do not let developers come up with the answers themselves.
5. Make sure the chatbot behaves as it should. Practice extensively in practice with the functioning of chatbot.
6. Optimize the chatbot in the goals

As an answer to the question to what extent the chatbot uses AI and the degree of self-learning ability, it became clear that a lot of (people) work is still needed for the functioning of the chatbot. The Natural Lanqueage Processing section "what does the questioner mean?" Is AI driven. For the remaining part, the chatbot is "less intelligent".

The time turned out to be too short to explain all slides and answer all questions. We are looking forward to a follow-up within the CIO Platform Netherlands to be able to zoom in further on this! The presentation is available via the internal knowledge base (login required).