Automated Virtual Assistant (Chatbot)
Challenge:
Thomson Reuters (TR) – The Answer Company has customer service operational costs close to $3M/year just for one of the business units. TR wanted to solve a multi-faceted problem of reducing customer service costs and improving efficiency. In addition, TR was looking for a solution that could be scaled across the company in the future.
Solution:
I started with secondary research to understand the problem and the market/industry competitors. Global companies of this scale were solving this issue using artificial intelligence and machine learning. After looking at industry reports and statistics, an automated virtual assistant in the form of a chatbot was proposed. The AI-based chatbot could assist in performing repetitive and mundane tasks while providing state-of-the-art experience and technology to the users.
My Role and Deliverables:
I worked as a Product Designer. I worked on the end-to-end design cycle, from strategy to production. I completed market and industry research, user research, natural language conversation design, chatbot experience design, driving and coordinating the product team along with the product owner, and usability testing of the product.
Deliverables
• Industry & Market Research
• Design Strategy & Concepts
• Visual Comps & Production-ready specs
• Prototype
• Usability Testing
• Chatbot Demo
Secondary Research:
I commenced with desktop research to understand the current state of the market and what other industries are doing to solve similar problems. One solution that stood out was artificial intelligence and machine learning techniques. Below is the summary of the secondary research.
Within five years, Chatbots would save billions of dollars by reducing an agent's time on a call to troubleshoot a problem.
Applied (AI/ML) Artificial Intelligence and Machine Learning in the form of Chatbots will be the future of customer service.
The total cost for 2018 customer calls was estimated at $3M, and thus, there was a huge potential for lowering the costs.
The secondary research aided in understanding the industry's depth, breadth, and trends.
Workshopping Use Cases:
The immediate significant step was to identify the frequent use cases. I organized and conducted a workshop with business, customer service, and technical services to discuss use cases and cost prioritization in the same room.
By conducting workshops with the stakeholders, I identified the low-hanging fruit. These tasks were high volume, high frequency, and time-consuming.
The estimated annual costs were more than $3M.
Along with the workshop, I interviewed secondary users, such as customer success reps, client management, tech support, and customer service, to empathize with their pain points and understand their priorities.
The gap between business wants and users’ needs started reducing.
Prioritizing Use Cases:
The contextual inquiry and workshops helped gather many data points and insights. But bringing order to the chaos was crucial before proceeding further. Based on the collected information, I segregated it into three different categories (routes). In close collaboration with the product owner and the technology team, I identified the level of effort required, which helped me design the product roadmap.
Conversation Design:
Before getting into the interface design, it was essential to work on the natural language conversations and probable questions the users might ask. All this information fed into the machine learning to train the algorithm, thus over time, the system could get smarter using artificial intelligence. I started designing the initial conversations by creating possible questions an end user would ask and probable answers a bot could reply with. During this exercise, the bot's tone, attitude, and language were crucial to give a consistent and helpful message. Later, I also piloted a workshop to create a bot personality.
“ If Content is King, then Context is King-Kong! ”
Visual Mockups:
The client provided me with a design system library, and I had to overlay the chatbot interface with the current system. I designed a chat interface using the brand guidelines and keeping the experience consistent with the whole website. I went through multiple iterations of designs, layouts, and micro-interactions so the visual language and feel remain consistent across different devices. I designed the animations using Adobe AfterEffects, which were used in the prototype and demoed them to the engineers. All the design changes/iterations were spec-ed out and committed directly to Zeplin to keep one source of truth for the developers.
Below are the prototype demos to show how the chatbot will help the users to complete their tasks efficiently.
Flow 1 - Check/cancel current subscriptions:
Flow 2 - Change of Address:
The detailed prototype was used for usability testing with the end users.
Usability Testing – Key Findings:
I tested the chatbot through a combination of prototype and code implementation with the end users as well as the secondary users like customer support, client managers, etc. The key findings were:
Bot Do-er than a Bot Teller:
Participants consistently indicated that the value of a bot is that it takes action versus providing how-to information.
The value of a bot is in performing a “substantive” task.
Something that helps the user to be efficient.
Perception of Hand-off to Human:
Participants reacted negatively to being handed off to a live agent.
They believed they could have contacted a live agent at the outset.
Baseline Experience:
The bot must deliver experiences that help customers be more efficient and effective than existing support options.
The bot must offer the ability to save, print, forward, and search bot conversations.
The bot needs to be promoted and explained broadly.
The bot needs to explain its privacy policy and access control.
Outcomes:
After usability testing, the feedback was incorporated to launch a pilot phase of the chatbot.
Secondary users, such as customer support and technical service teams, appreciated the virtual assistance for simple but mundane and time-consuming tasks.
It was a completely new experience for the end users. They were taking time to adapt to the technological changes. As the chatbot was in the pilot phase, achieving the users’ expected level of smartness took time.
My leadership thoughts and learnings about Chatbots can be found in this article.