AI email bot
The design problem to solve
Autobot was originally designed as a tool for data scientists to use. However, to make it sellable solution to a broader audience beyond the realm of data science, we recognized the need to improve its user-friendliness. This meant ensuring that individuals unfamiliar with data science could confidently utilise the tool.
We needed to design a tool that would be a completely new concept new to customer service reps, Autobot is not like any tool our users would have come across before so we needed to design this with our tool with this in mind.
Process
During our usability testing, we encountered significant challenges in achieving this goal. Many participants struggled to grasp certain language and features we had implemented. For instance, we initially labeled a Refund email as an 'Action taking email' because it indeed initiates an action – processing a refund to the account holder.
However, our participants found this concept unfamiliar and, in some cases, seemingly impossible. They couldn't easily comprehend that the AI bot not only responded to the account holder but also executed the refund process automatically. Consequently, their initial assumption upon encountering the label 'Action taking email' was that they themselves needed to take an additional action to activate the email, highlighting the disconnect between their expectations and the tool's capabilities.
Solution
The Autobot email manager, powered by a uniquely trained large language model, revolutionises customer service operations in the energy industry. With the ability to address over 40 energy-specific consumer needs. This tool empowers customer service reps to efficiently manage automated email processes, from monitoring performance to toggling intents and customising template content.
When users first access this tool, they are greeted with a list of the email intents that are currently generating the highest response rates. Understanding that emails sent in large quantities were crucial for agents to monitor performance, we prioritised these intents for visibility. Typically, agents sought to review or edit intent copy either due to seasonal changes or shifts in email performance. Our goal was to create a minimal viable solution that addressed these needs without overcomplicating the product.
To achieve this, we developed a straightforward email template editor that allowed basic edits such as text styles and links. Given the tool's wide usage among thousands of customers with daily inquiries, we incorporated dynamic tags to personalise emails with account holder names, addresses, bill due dates, and more. Furthermore, recognizing the importance of tracking email performance over time, especially in an industry where seasonal variations pose challenges, we implemented a simple metrics dashboard. Leveraging Google's Looker, we included this feature as part of our minimal viable solution to test its effectiveness before further iterations.
