Situation
Starting in February 2020, I began the journey of designing and implementing the first virtual agent (chatbot) within SWIFT. The virtual agent resides in the internal IT Service Portal and helps users with a number of different things, from searching through user written articles for keywords to creating and updating IT tickets.
Outcome: Thus far, the virtual agent has reduced the number of contacts to the Helpdesk team by 15% and provided users with working solutions to four of the tasks they most commonly need help with.
Goals
To reduce the workload on the IT Helpdesk team to allow them to attend to more complex issues and to provide users with a sense of independence and agency when resolving issues.
Deliverables
Fully functional virtual agent (VA) implemented in Production
4 custom-built conversational flows, designed and coded from scratch
Conversational components customized to the user
Tools
ServiceNow
Adobe Illustrator
Bot Society
Phase 1- Understanding and Implementation
Feb 2020 - Mar 2020
During the first phase of the VA project, I utilized online resources to teach myself how to install the VA within the Service Portal and how to implement pre-built topics to assist users with basic functions, like checking the status of an open ticket.
Results
VA was added to the live Service Portal and was able to interact with users to complete 3 different tasks based on pre-built topics.
Phase 2 - Custom Topics and User Needs
Apr 2020 - Aug 2020
In the second phase, I focused on expanding the capabilities fo the bot. I took the initiative to conduct multiple user interviews and review the Helpdesk team’s records to analyze what users most often needed assistance with that was within the capabilities of a chatbot. With the results from this research, I determined the 4 most common tasks users needed help with.
I used Bot Society to experiment with different flows and responses for these 4 different tasks and worked together with the Product Owner to finalize the conversational design. Then, I taught myself how to code these tasks into custom-built topics for the VA. Through the use of online resources, and with a background in Computer Science, I was able to not only successfully code 3 unique topics myself, but also teach a coworker the ins and outs of the VA and teach her to code custom topics as well.
Results
4 custom built topics within the VA that allowed users to complete the tasks they most often reached out to Helpdesk for. Helpdesk has seen a reduction in user contacts by over 15%, which allows them to focus their time and energy on more complex user issues.
Additionally, I taught myself and a coworker how to code and implement custom topics.
Phase 3 - NLP and Customization
Sept 2020 - Nov 2020 Next I focused on making the experience feel more personal to the user. Utilizing my training from the Conversational Design institute, I designed 6 new greeting messages that were customizable to the user. I also created a dozen unique fallback and confirmation messages for the bot. This allowed for a more custom and personal experience for the user.
I then activated the Natural Language Processing (NLP) capabilities in the bot and began training the bot on different phrasing commonly seen from users to initiate the different topics, as well as different response types. Adding the NLP helped make the conversational feel more natural for the user, and greatly expanded the bots understanding of user requests and responses.
Results
6 customized greeting messages, 6 customized fallback messages, and 6 customized confirmation messages. Implemented and began training for NLP to create a more natural conversational flow and expand the bots linguistic understanding.