Brand: CommBank
Collaborated with: Content Designers, UX designers, Product Owner, Call Centre Specialists, Editors, Product Managers, Compliance and Business Analysts.
Project length: One year
Brief
Collaborate with designers, product managers, and developers to create a chatbot available to customers in the app and online banking. Refine chatbot flows, anticipate user needs, and develop responses that answer customer questions and help complete banking tasks.
Problem statement
How might we use a chatbot to reduce call volume by efficiently answering customer queries and guiding them to relevant resources or actions online?
Approach
Data: We used call centre data, top search terms and website visit data to determine the top 100 customer queries to focus on for day one. At every stage, we used data to inform content and iterations.
UI design: Through user testing, we established some design guiding principles to inform content, which included:
No more than three speech bubbles per response to keep it short and scannable
One sentence per speech bubble to aid scannability
Active CTAs with no more than three words to encourage engagement
Clear, active and conversational language (slightly more human than our usual brand voice)
Conversation design and writing: Given the size of the project, three content designers and three UX designers contributed to the end product. Our process was:
Create a confluence page for an intent and have the call centre specialists add information and links on how to answer the query.
Content designers would collaborate with a UX designer on the conversation tree before moving into the writing phase. Complex queries with large trees would be sent for comprehension testing before iterating.
Copy and conversation tree would be added to the confluence page for approval from product and compliance teams.
Business analysts built out the intent in the CMS.
Training the bot: As part of the content delivery, we also had to write as many variations as possible for how a query could be asked. This was uploaded to the backend to help train the bot on responses.
Content guidelines: Once we had tested the design and proposed content structure, we created a set of content guidelines on Confluence that other writers and teams could use to produce conversation intents in the future.
Content iterations: After going live, we monitored conversations and used customer data to inform how to improve conversation designs and copy. This was reviewed daily in the early days, and then an operations team was trained to update and create new conversations.
Deliverables
Contributed to over 150 customer queries ready for launch and a further 100 informed by data after launch.
Insights
This was one of the first projects that showed me the power of UX and content designers collaborating and working towards a shared goal.
People believed the bot was a person - we had to emphasise bot versus human. One of the ways we did this was by updating the welcome message to mention the user is talking to a bot and referring to humans when handing over to a human channel.
No matter how many variations of a question we wrote, people still asked questions in obscure ways
The language used by customers typing to the bot is often very different to how we refer to banking products - we had to adjust, even when in some instances it didn't align with brand guidelines. This learning was usually also passed on to the website team and other content designers.
Comments