Health insurance
Health-insurance enrollment, from 9 steps to 5
One flow for open enrollment, life events and new hires.
Dennis has 31 days to get his newborn covered.
In the old flow, adding his daughter and covering her were two separate jobs, and every coverage added another step. In the redesign he does both in the same row.
What it takes to cover Noor, his newborn
MedicalDental
Old flow
- “Not enrolled in coverage”!
- Medical plan page → Add Noor1
- Dental plan page → Add Noor2
2extra steps, one per coverage, to cover her
Redesign
0extra steps: ticked in the row where he added her
Dennis is a persona I use to tell the story, and these are illustrations rather than the shipped screens. I counted the steps by walking through the old and new flows, and the prices are made up for the demo.
Everyone said the flow was too long
The feedback reached me from two places, users and our business analysts, and it boiled down to four complaints. The enrollment flow felt too long, members got confused about dependents, they didn't see the cost until late, and admins struggled whenever they enrolled someone on a member's behalf.
The obvious fix was to trim a few screens. Before doing that, I wanted to know why there were so many in the first place, so I walked through the whole flow step by step, the way a member would.
Flow map
Same decisions, reordered
- 01Effective date
- 02Profile
- 03Household list
- 04Add member
- 05Updated list
- ↺ Repeat per coverage: medical, dental, vision
- 06Coverage hub
- 07Browse, pick a plan
- 08Who does it cover?
- 09Confirmannual estimate
- Steps
- 9
- “Who's covered?” asked
- once, then again per coverage
- Cost shown
- at the end, as an estimate
What I found was the backend's data model, turned straight into screens. It took three screens just to manage the household. Then every coverage had its own loop: pick Medical, browse plans, pick one, choose who it covers, go back to the hub, and do it all again for Dental and Vision. So members answered 'who's covered' once for the household and again inside every coverage, and the price they saw was a yearly estimate that could still change.
The flow is too long, so trim some screens.
It asked questions in the order the backend stores them, so members kept answering the same question.
Cutting screens would have made that loop shorter, but members would still answer the same question over and over. And here a repeated question is more than an annoyance. A birth opens a window of 30 or 31 days, and a family member missed on one coverage stays uncovered until the next open enrollment.
I changed the order of the questions instead of trimming screens
Instead of trimming, I reordered the flow to follow how a family actually thinks it through: first my family, then who needs what, then which plan. 'Who needs what' is now asked once, in one grid with family members down the side and coverages across the top. Picking plans builds on that grid instead of asking the same thing again for every coverage.
I made the most common choice the default
Once the grid was in place, most of the remaining work was picking plans. If I treated every path the same, the rare cases would get as much screen space as the common one, so I designed the defaults on purpose.
A shared family plan is the default, with a clear way to pick plans per member when someone needs that. Keeping your current plan is the default too, and waiving coverage sits right there on the screen instead of being tucked away. My bet was simple: the most common path should be the shortest, and the less common one should still be one clear step away.
Family-by-coverage grid
Ask “who needs what” once
| Family | MedicalCurrent plan | DentalCurrent plan | VisionCurrent plan |
|---|---|---|---|
| DennisEmployee | |||
| AnaPartner | |||
| MayaChild | |||
| NoorNewborn |
When engineering pushed back
Engineering pushed back, and they had a fair point. Their model went coverage, then plan, then members, and I was asking them to take the answers in the reverse order.
I didn't want to push the backend's order back onto members, and asking for a full backend rewrite wasn't realistic either. So I walked the team through the order a member makes these decisions in, side by side with the actual complaints from users.
Diagram · two orders, one translation
We landed on a middle path. The UI collects the grid and translates it for the backend, and the backend changes only where translating isn't enough. We split the work and shipped it in phases instead of switching everything over at once.
Small things the new order made possible
Once the questions came in the right order, two more of my calls became possible. There's now one review page before signing, where members see what changed, edit it right there and e-sign in the same step. And document upload, which used to be a fixed step for everyone, only shows up when a member's situation needs a document.
What changed
The new flow is live in production for open enrollment, life events and new hires, and admins use the same structure when they enroll someone on a member's behalf. Using that one structure everywhere was a decision I made with product and engineering. It went from 9 steps to 5: Welcome, Profile, Dependents, Coverage and Review.
The complaints that started the redesign have dropped, and admins get through enrolling on someone's behalf faster. To be honest, these are signals we've seen rather than formal measurements, which is something I'd fix next time.
The product · before and after
The same life event, in the old flow and the new one
Before, the household was a list of names, and who each coverage applied to came later, once per coverage. After, one grid answers who needs medical, dental, and vision for the whole family, up front.
Before
After
The members who never open the flow
There was one more piece of work, on the admin side. Every open enrollment, some members do nothing at all, and admins used to sort them out by hand, one at a time. I designed a way for each employer to set a rule once and run it themselves, and anything the rule can't decide comes back to an admin with a reason. The business owned the rules, and my part was how admins set them up and handle the exceptions.
What I'd do differently
I'd get evidence in earlier. The redesign was based on feedback from users and business analysts plus my own analysis of the flow, not on watching real members use it. I'd also add tracking to the old flow before changing anything, so I could show the improvement in numbers instead of only describing it.