Intro
By simply texting, Gary, an AI-enabled chatbot, will provide you a holistic, customized, and real-time insurance service. Say goodbye to the cumbersome, time-consuming, and almost intimidating experience of traditional insurance products. You can now get a quote, file a claim, and receive holistic services all by chatting with Gary on your mobile phone or laptop.”
I led the UX design of an insurance AI-powered chatbot that helps users get quotes, file claims, and manage bills all by chatting on mobile. The product was utilized by multiple carriers in the industry. One of them provides 2.1 million members with travel, insurance, financial and automotive-related services.
Our Goal
Insurance is one of the most confusing and complex industries out there. The agent-centric sales model was developed to better serve customers’ needs. Over 50% of insurance searches are performed on mobile devices. In our mobile-first world, many insurance searches are run on smartphones, which makes calling an agent more seamless than ever before. Auto insurance shoppers are the most likely to speak to an agent during their purchase journey, followed by home insurance, and health insurance consumers. The more complex the purchase, the more likely insurance consumers are to speak to an agent.
What if we can make all the tasks that a user will need to do all together in a streamlined fashion by chatting with a virtual agent? i.e. Get a quote, file a claim, manage policy, get support from live agents seamlessly.
Current State Pain Points
When starting on this project, I started by researching the current pain points that users are experiencing when engaging with an insurance policy online, mostly on websites. Five major issues were found:
1. Organization-Centric IA and Navigation
The number one user experience issue with most insurance websites (particularly the larger ones) is that they mirror their actual organizational structure online, making their websites unnecessarily complex — and largely pointless for customers.
An example of org-centric mega menu that works for the company, not users.
2. Digital Experience can’t replace the human touch
A website will always fail if it can’t provide information to people more quickly and more easily than they could obtain it by picking up the phone and calling their insurance agent. Lots of companies will treat their website as a huge content library, leaving the users to wander in the jungle of content, which is mostly irrelevant to what they are looking for.
This is what I kept in mind specifically when designing this chatbot experience. Virtual agents can't fool users to think that it's a real person and not everyone loves chatting with one. Aside from designing this chatbot to be as user-centric as possible, how and when to connect users with a real customer support representatives is crucial.
3. Lack of Choice & Transparency
Long, confusing, and full of jargon insurance policy is scary and not easy for users to process or understand the information, especially for people with no previous experience.
4. Fragmented and disjointed legacy systems
Expecting users to navigate through one system to pay their bills, another system to file a claim, and a third to get a quote is a complete mess if you’re looking to deliver a superior user experience. Visitors shouldn’t have to pay the price for your messy online ecosystem.
5. A time-killing wasteland filled with forms
If you asked people what they least liked about the process of purchasing insurance, you would likely hear one phrase repeated over and over — “I hate all those forms!”
With the above five key pain points of users in mind, a solution is needed to better onboard potential customers, engage existing ones, and provide a better experience.
"To fill out Form A483-32, please refer to Form D83-1 at the end of Form C933-243"
Design Process & Challenges
After a close look into what's working and what's not working nowadays in online insurance experience, I followed the design thinking process to empathize, define, ideate, prototype, and test. I analyzed the information architecture and started from low-fidelity sketches to visualize my idea and ran it by the business & development team to validate the feasibility.
One of the challenges my squad faced was that the development team don't have much experience in building a chatbot. Therefore, I did a lot of competitor analysis to see what features/experiences are commonly used in other products and also looped in the tech lead in the ideation process to make sure the design is feasible.
I conducted a quick user testing with 5 people in the target user group to see what's their reaction to the experience, with the mid-fidelity prototype on Invision. The responses were positive, agreeing that this chatbot is a more user-friendly, engaging, and delightful experience, comparing to their experience with current insurance products. Here are some quotes I've captured:
"I just got my first auto quote last month with XXX insurance company. This design is much easier to understand, making me feel like I was talking to a person and achieve my goals. Answer one question at one time." -- Male, 28
"I like the way the chatbot asked me questions and I have these pre-populated answers to respond quickly. The experience makes me feel like I'm in control." -- Female, 26
One user did mention that he thinks the chatbot is a bit confusing cause he's not sure if "Gary" is a real person or not; hence, he's not sure if he could type freely in a conversational way, like idk, brb...etc. This response made me discover this interesting pattern that people tend to speak in a robotic way when they know they are talking to a robot and talk in a more conversational style when talking to a real person.
To make this chatbot a good experience, I have no intention of fooling users to think that Gary is a real person; but I would like users to be able to chat freely with Gary as if it's a real person and the semantic system should be able to understand and react.
Final Design
After few iterations, the final design was hand-off to the development team, built, and shipped. Here are some key points of this chatbot that we believe can make people's lives easier when engaging with auto/home insurance policies:
1. Say goodbye to mega menu
The chatbot will provide you the most relevant content based on your responses. Integrated with the car monitoring device, it can proactively reach out to the users if it detects any collision or auto parts failure. The chatbot will only ask for the minimum required information to get a customized quote to let the users spring through the process.
2. Customer support representative step-in when needed
Most people are not ready to give up human interaction completely. When the system detects certain emotional responses or the users have been idle for a while, the chatbot will prompt to ask if they want to speak with a live support representative. If yes, the previous conversation or the data provided by the users will be smoothly transferred to the representative.
3. Complete transparency and flexibility
After providing the required information, the users will be prompt with their estimated monthly quote along with a list of liabilities and coverages. As mentioned above, this part could be scary for users without any experience with insurance. Therefore, there will be a short text explaining what each coverage would work under a certain circumstance. Also, the system will recommend the best option for users based on their driving behaviors. The users will have the peace in mind that they are getting the most recommended coverages they desired.
By law, the application needs to show the full name of every coverage to inform the customers. Therefore, plain text explaining how this coverage would work is added under each coverage to make sure that it’s speaking users’ language.
4. One chatbot for an end-to-end service experience
From quotation, payment, claim, billing, to another service, the users can go back to the chatbot and simply request the desired service.
5. No more forms, just chats
By working with subject experts in insurance industries, the chatbot is designed to ask only the minimum required questions to get a quote. With the pre-populated responses, the users can breeze through the quoting process and get what they want in less than 2 minutes while chatting!
This chatbot was utilized by multiple carriers in the industry. One of them provides 2.1 million members with travel, insurance, financial and automotive-related services. We will continue to monitor and receive feedback from the carriers and improve. Also, I created a future state design to envision how this tool could be utilized outside of auto insurance, like home insurance.
Envision: Home Insurance Chatbot
Takeaways
1. Always loop in subject experts in checkpoints to validate design ideas. One major design issue in the process I've encountered is that I didn't realize that the law requires policy providers to display all the coverages and details on the page directly. That changed the original hide/show design.
2. When working with a development team with little experience on the tool, making sure that they have enough time to learn and engage them since the start of the design process, thru each iteration.
3. Don't try to fool users with anything that your product is NOT. Users can easily tell if it's a bot or a real person. Even if they can't, designers shouldn't try to cheat and create dark UI.
Thank you for your time reading this article! Please let me know if you have any questions or feedback. This article is also included on ChatbotLife magazine on Medium.
