For proprietary reasons, I am only able to display my work from the geico.com bot, as it is available to anyone who goes to the site.
Thank you! Samantha
Project Containment
Goal: Improve the self-service experience while also reducing our transfer rates to live agents.
Approach:
In early 2023, we found a slight majority of our customers were requesting to speak with an agent during their chat conversation. Through analysis, we identified multiple intents where a high frequency of customers would trigger a request to chat with an agent.
One of those experiences was our confused flow(intent).
At the beginning of creating our bot in late 2021, our focus was to be the most helpful possible. This included offering a live chat agent in most of our conversation flows.
Our static bot [GEICO.com] was one of the last applications to be fully implemented into the omni-bot experience in fall 2022. In the stages of creating the static bot, we opted to send a default response—or confused flow, to our customers while we continued to build out more specific responses.
First Iteration
*Initial response of our static confused flow(default response) from 2022.
Six months later, we found ourselves in a pickle. A high frequency of utterances [a customers request] were all being offered the confused flow, because of the lack of response we’d built out for the static bot. This drove substantial increase in transfers and lowered our bots overall containment rates.
Our team worked to analyze the utterances and mapped out a plan to add intents for key utterances, as well as creating a stop gap for our confused flow. (The bot also required additional training for utterance confidence).
Between updating multiple intents and adding a second attempt to our confused flow, we reached a peak containment of 60%, meaning we’d improved our transfer rate by almost 20%. Today we average a monthly transfer rate of 42%.