Person smiling while using their phone
Premera Blue Cross · Healthcare · 2019–2021

Making it easier to get help with your health insurance

Premera Blue Cross is the largest health insurance company in the Pacific Northwest. I led the design of a chatbot that transformed how customers get help, making a stressful experience significantly faster and more satisfying.

Company
Premera Blue Cross
My Role
Design Lead, Research Support
Team
1 PM, 1 researcher, 5 engineers, 1 designer
Duration
6 months

35% of customers were frustrated calling customer service

Like most health insurance companies, it's difficult for Premera to earn customers' trust and satisfaction. To find opportunities to improve, we routinely sent feedback surveys to our customers. From these surveys, we discovered that 35% of people were frustrated when calling customer service.

My team was tasked with understanding what was causing this problem, and designing a solution.

Premera chatbot: initial message screen
The solution: a chatbot in the Premera app that routes customers to the right representative before they ever pick up the phone.
70%
Reduction in customer service call time
11%
Improvement in customer satisfaction
86%
Of customer service reps found the product helpful

Researching the problem

Before jumping into solutions, we needed to fully understand the problem from multiple angles. I partnered with our lead researcher to dig into both the customer experience and the business context, ensuring our eventual design would address root causes rather than surface symptoms.

Research goals

Map the existing user journey

Understand all the steps a user had to take to get connected to customer service.

Understand user pain points

What was causing frustration when customers called in?

Understand business constraints

What limitations did we have to design within?

Understand technical constraints

What was feasible for our engineering team to build?

Methods

Listened to customer service calls and read feedback surveys

Directly observing the problem helped us understand the full context of a caller's experience before talking to users.

Interviewed callers, customer service reps, and internal stakeholders

Talking to all three groups gave us a full picture, from the customer's frustration to the rep's experience to the business priorities.

Existing user journey map showing 9 steps from looking up the phone number to connecting to customer service
Existing user journey. 9 steps standing between a frustrated customer and getting help, including verification, hold times, and repeated transfers.

User pain points

Too much time in the automated phone system

It took callers about 3 minutes to go through our automated phone system, due to a strict verification process.

Callers were constantly transferred

After finally getting connected to a representative, callers were often transferred to another person, restarting the frustrating process.

Calling insurance is inherently stressful

Calls ranged from people worried they couldn't afford surgery to people frustrated that their insurance had denied coverage. The emotional stakes were high before the call even began.

Reps needed time to research before they could help

Customer service reps typically spent the first few minutes understanding and researching the caller's problem before being able to help, adding to call time and caller frustration.

Understanding constraints

Each call cost about $12

The business wanted us to shorten and prevent calls. Every minute saved had direct financial impact.

Engineering resources were limited

Our engineering team lacked resources, so the design had to be simple to develop.

Redesigning the phone system wasn't an option

Redesigning the automated phone system would be too expensive, so we had to find a solution that worked around it.

Brainstorming as a team

We presented the research findings to the team and brainstormed ways to solve the problem. Using the team's diverse expertise in engineering, business, data, and design, we generated ideas across different directions.

Workshop brainstorming sketches and sticky notes
Using the team's diverse expertise (engineering, business, data, etc.) to generate ideas.

Deciding what to build

When deciding what concepts to design and test, the main factors were the business and technical constraints. After analyzing various potential solutions, the team agreed to explore the idea of utilizing our existing mobile app to verify callers and allow them to skip the automated phone system.

2x2 matrix mapping concepts by engineering cost vs business goal
Engineering cost vs. business goal, used to evaluate which concepts to explore further.

Concept testing: chatbot vs. form

Two of the concepts that stood out were designing a chatbot or a form to replace the phone system. We decided to test both to understand their pros and cons.

Chatbot concept wireframe
A chatbot felt more informal, and users expected to get help without calling.
Form concept wireframe
A form felt more professional and was perceived as more efficient.

7 of 10 participants preferred the form

A form felt more professional and was perceived as more efficient. However, the team decided to build the chatbot because it better supported our business goal of reducing calls, while still solving the core user problem.

A chatbot felt more informal, and users expected to get help without calling.

Defining the user flow

I designed multiple chatbot flows, varying in complexity and types of questions asked. In the end, it was about keeping the experience quick and simple for the user.

Chatbot user flow showing branching paths based on ML confidence
I designed multiple chatbot flows, varying in complexity. In the end, it was about keeping the experience quick and simple for the user.

Usability testing: refining the design

After converting the user flow to a prototype, I conducted usability tests to further refine the design. Here are a few of my key findings:

Finding 01

Reduce the need for typing

Help the user quickly answer the bot's question by clicking buttons or using their voice instead of typing out responses.

Before and after showing button options replacing free-text input
Before and after showing shortened bot messages
Finding 02

Keep it short

Users tended to only scan content and often missed information if the message was too long.

Finding 03

Present the bot's top 3 guesses

When clarifying the user's question, the bot should display its top 3 guesses along with a "Something else" option, so that the user is never trapped.

Bot presenting top 3 guesses to the user
Bot failing gracefully and routing user to a human
Finding 04

Fail gracefully

If the chatbot is unable to understand the user's question the first time, don't try again. Just get them to a human.

High fidelity designs

The final design used the existing Premera mobile app to authenticate callers and route them to the right customer service team, eliminating the painful automated phone system entirely for app users.

High fidelity designs showing the complete chatbot flow across 4 screens
High fidelity designs showing the full chatbot experience: open app, start chatting, confirm guess, and call.

Successfully released on both Android and iOS

We successfully shipped this feature on both the Android and iOS Premera apps. Results were measured against the existing phone system over a 90-day post-launch period.

70%
Reduction in customer service call time
11%
Improvement in customer satisfaction scores
86%
Of customer service reps found the product helpful
Improved user flow showing old 9-step journey replaced by new 4-step chatbot flow
The new design eliminated the major pain points when calling customer service.

"I think this is an excellent step taken by Premera to help get me off the phone faster."

- App user review

What I learned

Involving engineers early is critical

Involving the engineers throughout the design process was critical in understanding technical constraints as well as promoting a culture of collaboration. This also helped reduce implementation time since they could start developing while I was still designing.

Balancing today's constraints with the ideal future state

One of the biggest challenges with this project was balancing what was possible today with an ideal future state. However, the team is committed to continuously improving the chatbot's functionality so that it can answer more questions and help users without them ever needing to call.

Neil presenting Premera Scout at a Digital CX conference
Presenting Premera Scout at a Digital CX conference - one of the more memorable moments from this project.