Health insurance is without a doubt one of the least human-centered, most convoluted products out there. Ensuring one of our most basic needs is one of the most complex endeavors.
Over the last few months, we were tasked with translating this elaborate world of services into an AI chatbot: a platform expected to be efficient, engaging and easy to understand — a challenge, to say the least.
We came out alive, and the process left us with invaluable insights into conversation design, chatbot functionality and user expectations, which we’d love to share. (Sidenote: there’s still endless health insurance challenges to tackle in Israel and abroad, but one step at a time!)
Get your information from the front lines
When designing conversations to be implemented in a chatbot, it is best to gather the information you need from…real conversations! Our most valuable learnings throughout the project came not from the head of marketing, digital, customer experience or even the head of the health division, but rather from the customer service representatives — those taking health insurance related calls day in, day out. Focus groups with representatives and recorded customer service calls served as the true foundation for our chatbot, from inspiring the design of the conversation, to guiding difficult decision making in terms of the processes and features for the chatbot platform.
The bot is not a threat to jobs, but rather a gift
In the same vein (we love customer service reps), a chatbot should not be perceived as a threat to customer service employees, but rather a tool that heightens their credibility, responsibility, and sense of purpose. A chatbot, no matter how well designed, is still a chatbot. Conversations about health insurance can be intricate, sensitive and often times emotional. A bot exits to deliver simple, immediate information and execute simple, immediate tasks. A human does far more, from solving highly complex inquiries and requests, to imparting care and empathy. A chatbot should serve to empower, rather than threaten employees.
Easy on the persona design
There’s a lot of emphasis on personality design these days. A bot’s personality, conveyed primarily through language, sets the tone for the user’s experience and distinguishes it from other bots. It introduces a human element to a mechanical interaction. However, we should be cautious in highly emphazising or embellishing personality in a utility bot (one meant to get the job done, and get it done quickly). Utility comes first, persona second. When a user is seeking to file a claim for a recent surgery or check coverage for a certain medication, overly kind words, puns and emojis can make an otherwise expedient process annoying and gimmicky. And no level of chatbot empathy will make up for subpar service.
Endless opportunity is limiting
AI-based chatbots are especially exciting because they allow users to engage in an experience that comes close to an authentic conversation. Exceptionally designed bots deliver experiences that are difficult to distinguish from interactions with humans. (Just see these love notes to “Amy Ingram”.) Still, a few fixed choices here and there are not the enemy. In the attempt to maintain an authentic conversational feel, our first chatbot prototype was based entirely on open text, without any buttons to guide users along the way. Users’ reactions were similar to mine the first time I tried to pick a body soap at Walmart: freeze, panic, retreat. Our final prototype included three guiding buttons at the start of the conversation, there to help users navigate their conversations within the endless world of health insurance.
Prototype from day 0
Finally, and perhaps most importantly, designing a successful chatbot is a complex process that takes time, and requires ample information. Do not wait until you have your concept, persona, processes, features and dialogue trees perfected to begin prototyping conversations, as you will immediately find that they are far from perfect. Start prototyping from day one. Simply experiencing the look and feel of the platform will already help you reach important UX conclusions, regardless of the conversational content. As you begin designing dialogues, prototyping will continue to help you and your clients make important decisions and adjustments. We worked with DialogFlow and landbot.io, two excellent, intuitive tools for prototyping Telegram, Facebook Messenger, Slack, and web-based chatbots.
We’re excited to continue our exploration and practice of conversation design, a medium we will see more and more of in the years to come (whether text or voice-based). We’ll be sure to share additional learnings that might help create more thoughtful, effective experiences, and of course, we’d love to hear any input of your own!
