Project GEICO Virtual Assistant
Goal: Build and design a chatbot experience to assist digital customers in self-servicing their policies, while also reducing the need for live agent intervention.
Initial Approach:
When GEICO changed their chatbot vendor in 2021, my team and I were enlisted to build and design a chatbot experience with the objective of keeping our customers in the digital space. As a conversational designer/project manager, this meant understanding the user and bot needs and writing experiences that answered questions and provided a live agent to chat when a customer needed additional support.
Our first Minimum Viable Product(MVP) came after a 3-month project increment (PI).
We launched with an initial batch of intents for an authenticated path and our static geico.com bot to customers on desktop, to pilot and observe initial interaction with our customers.
Every PI, we increased the volume of intents and applications to which we offered chat experiences.
Initial results found:
Automation:
Call back rates from customers that only interacted with the VA continued to drastically trend down since inception.
Overall Deflection:
Because of my changes, the overwhelming majority of customers that interacted with the VA or chat with an agent did not call.
Over time, our goal of the project shifted from not just offering guidance and a live agent in most intents, to, “what if we could start deflecting assistance to a agent all together?”
What if we could create a bot experience that not just answered customer’s questions, but also helped them self-service their policy on their own? This is when we started conceptualizing integrated flows to handle common policy needs, like making a payment.
Fast forward 3 years later, and our team is maintaining and improving our current bot experience while re-imagining the experience as a whole.
Our team’s philosophy and sprint format:
To iterate often and with purpose. The more incremental and data supported changes we can make, the better the customer’s experience. All intents made are based off a user-case that is substantiated with data.
I’ve found through analysis and testing, the best conversations with a bot are simple in format. And the language crafted should always be user-centric.
My Work on GEICO’s Bot:
As the conversational designer for GEICO’s Virtual Assistant, my role has been to create and update intents iteratively. To date, myself and a project manager keep up to date:
Over 100 Intents, including ones that complete a customer’s request for them
All applications have a bot response built in with multiple lines of business
2 languages (English and Spanish)