Case study: Conversation Design at Vattenfall
“Claire and Willem have taken our chatbot to the next level. Nina is now smarter and can better serve our customers. They not only wrote good texts, but also made the team grow thanks to their insights. ”
At the end of 2019 Nuon got a new name: Vattenfall. This led to a complete rebranding, including the website and the chatbot. All of the content was rewritten according to the new tone-of-voice, so the chatbot could not be left behind.
Her name remained the same: Nina. But her personality got a complete makeover.
The chatbot’s content also needed to be optimised..The content was added to the chatbot with enthusiasm, but it did not always match the customer’s demand.
Fortunately, we weren’t alone. We were part of a special chatbot team with (ex-)customer service employees, two developers and a tester. Our Product Owner and the Scrum Master led us in an agile way of working.
We took the following steps to optimise the dialogue of the chatbot:
Step 1. Analysis existing dialogue
Every optimisation starts with an extensive (data) analysis of the existing dialogue. Secondly we analyse the use of a dialogue over a fixed period.
Step 2. Proposal optimisation
The analysis is the basis for our optimisation proposal.. In this proposal, we advise to place topics earlier in the dialogue, so that customers reach the correct answer sooner.
Paths are built more logically and we add content that is still missing. We make a rough sketch of the dialogue. We then submit this proposal to the content specialist.
Step 3. Refinement with the content specialist
We discuss the proposal with the content specialist. Where we, as Conversation Designers, monitor the structure and customer journey, the content specialist mainly looks at the content. Is it true what the chatbot says?
We put the recommendations and comments of the content specialist in an adjusted final proposal. We check this again with the content specialist for a final sign-off. After the final approval, we can implement the optimisations.
Step 4. Writing , building and testing
One of us writes the content according to the correct tone-of-voice. The other Conversation Designer then double checks the text. After this review round, we build the dialogue in the chatbot platform.
As soon as it is built, we test the dialogue in the offline environment. We look at the following:
- the text is clear and legible
- the links are working
- the structure is logical
Step 5. Feedback, multimodal design and the launch
When the test is done, we add the "feedback bubbles". Customers are then able to indicate per answer whether they have been helped (thumbs up) or not (thumbs down).
We then ensure that the content is also correct in the app and for the business market. These chatbots have slightly different writing rules. After this, the optimised dialogue is ready to go live.
Step 6. Continuous monitoring
We continue to monitor even after the optimised dialogue is live. After a month we take another look at the data, such as the drop-out rate, the percentage of negative feedback. Where necessary, we fine-tune the dialogue. And a month later we check whether the results have improved.
Want to know more about conversation design?
Curious about what this new specialism entails? And what we do for our customers?
