Creating individual content flows for COVID-19 across multiple chatbots.
As each client had different information to communicate regarding the crisis, we needed to design, write and train COVID-19 quickly.
Our challenges were on 3 fronts:
- Design new flows
- Re-write COVID-19 policies
- Train our NLU agents daily
Design New Flows
We divided the content into themed topics such as health precautions, booking rules, refund policies, insurance policies and more.
Due to the fluid nature of the policies, we grouped the content which allowed for quick updates of the text only.
Re-write COVID-19 Policies
The policies were lengthy pieces of content and a major challenge to adapt. We wanted to communicate important information as effectively as we could, dividing content messages bubbles, hard-coded bold text and appropriate emoji use.
Training The NLU
Daily training data was added from real interactions to keep up with the variance in queries/utterances regarding the virus. We added substantial amounts of synonyms for COVID-19 (which changed names frequently) as well as terminology for crisis, lock-down and quarantine.
Generally words outside the scope of a normal chatbot.
Final results: Communicated COVID-19 updates faster than website and emails, alleviated pressure on call centres.
Kris
As this was a travel agency assistant, the information had to communicate agency policies as well as countless of airlines.
The fluidity of new information meant that changes were daily such as updated refund policies and new airline contacts.
Kris would experience 10,000 triggers per day of this intent.
Toby
With Toby, unfortunately the airline client suspended all flights.
We found a clear message using hard-coded bold text to be most effective, with button options for more information.
The "your bookings/credit" button efficiently linked users to relevant updates on their bookings.
Donna
Donna is the virtual assistant for a Canadian airline.
Their content required to communicate a lot of information which changed daily as well. In addition, their customer service agents were overwhelmed.
The balance of not adding too much content without losing important information was a definite challenge within the confines of chatbot window.
We added button links to all greeting/intro messages to direct users as early as possible to relevant information.
