GEICO Virtual Assistant

*Disclaimer: The use cases and work examples below are my own creations. They are representative of the work I performed at GEICO, but due to confidentiality agreements, I have not shared any classified information, screenshots, or documentation.

I fell in love with conversation design when I began working on the GEICO VA. It was a behemoth sized project that required me to wear many different hats and quickly upskill. From proof of concept to the large, omnichannel bot it is today, I’ve worked on it every step of the way for two years.

Highlights

  • 120+ intents

  • Appears on 13 different applications/lines of business

  • 10+ million interactions per year

  • Scored highest on Keynova insurance competitor VA report

  • 80% call deflection

The Goal

As with most virtual assistants, the goal for the GEICO VA consisted of two parts: provide an easy, enjoyable 24/7 channel for customers, and increase efficiency through automation.

The Challenges

We had several unique challenges when building this VA:

Insurance is a complex industry, and we needed to explain complicated answers concisely and with limited use of jargon.

Though GEICO’s brand in advertising is humorous, customers’ issues are no laughing matter. Emotions can run high, especially in claims situations, so we needed to be careful with our persona.

GEICO services many different insurance products in several different applications. Our directive was to create a bot that could work for them all. So take that 120 intent number and then multiply it by the 13 different versions of each intent we created, including a Spanish language option. As you might imagine, organizing our work alone would be a challenge at this scale.

The Persona

If you aren’t intentional about your VA persona, then customers will assign their own. With this in mind, persona creation was a top priority at inception, and we continued to refine the VA’s personality and habits as we went along. Customer interviews revealed to us that users expected the bot to be professional and empathetic, especially during such a stressful time post-March 2020. While I can’t share the details of what the GEICO VA’s persona is exactly, the details we defined included:

  • Common acknowledgements and confirmations

  • Parameters around sensitive subjects like insults, politics, data usage, etc.

  • Chit chat topics

  • Relationship with the company & how it uses “I” versus “we”

  • Core personality characteristics (3)

  • Hobbies

  • Back story

  • Preferred pronouns

  • Goals

  • Voice & tone

  • Common phrases

  • How to show empathy

  • How tone changed based on sales vs. service vs. claims

  • Reading level comprehension standards

  • Level of transparency

Quick note about empathy: this is a hot topic in conversation design, and my experience has taught me that the most empathetic thing you can do for a customer is to resolve their issue quickly. Assigning emotions to users or dwelling too long on the problem can be dangerous. Best to show the customer you understand what the problem is and then offer a solution as soon as possible.

Add a Vehicle Use Case

Auto insurance VAs often receive requests from customers to add a new vehicle to their policy, which can pose several challenges.

  • One challenge is the varied ways in which customers may refer to their vehicle, such as "car," "truck," "Honda," or "2020 model." It's important for the VA to use entities or slots to accurately identify the vehicle and distinguish between adding a new car versus replacing an existing one.

  • The online flow for self-servicing customers can be complex, so the VA may need to break the process down into smaller steps and set clear expectations for customers.

  • Many customers may be looking for a quote rather than making a change to their policy, so the VA should make the quote process clear and transparent.

  • Some information may be better displayed on the website or app, so the VA may need to direct customers to these channels to provide complete and accurate information.

  • You need to research why customers are switching to the VA in the first place and identify self-service issues.

Sample Dialogue

The easiest way to start brainstorming what your happy path flow my look like is creating a sample dialogue, which is just a an example conversation between a user and the VA. Fallbacks and error handling don’t come into play yet. Figuring out the meat of the flow is enough work.

One trick I’ve found to be useful for chatbots is to work with the conversational copywriter and role play as the user and VA by pinging each other back and forth. We both have to put thought into it, but the conversation flows more naturally from the start, and these messages alone can be the basis for a sample dialogue.

Flowchart

*Note - the process shown here is not specific to GEICO, nor does it include all steps necessary to add a vehicle, but for confidentiality reasons, I’m displaying a generic flow.

Once I have a good understanding of the process from SME’s and I’ve created a sample dialogue, I’ll create the happy path in a flowchart. I’ll notate where API checks will need to happen to give the engineering team an idea of the scope for feasibility.