Notification Stream
*Some content has been redacted due to PII and confidentiality.
Context
- Company
- UnitedHealth Group, broker and employer platforms
- Role
- Design Team Lead: owned the ranking strategy and stream design, led 3 designers
- Team
- 3 designers, 5 engineers, 1 PM
- Timeline
- ~8 months
- Tools
- Figma, FigJam, Dscout
- 86%
- Conversion on urgent items
- 4 clicks · 2m 46s
- To reach the first urgent item
The problem
Brokers and employers had to guess which platform to open first. Different group types and workflows lived on different sites, with no navigation between them, and key workflows hid several layers deep. Time-sensitive items in one place were missed while attention sat in another.
Discovery
I ran 10 interviews across roles and seniority, then shadowed 4 users through their first hour of the morning. One sentence came back almost word for word: "I never know which screen to start on."
Funnel data confirmed it. Users bounced between surfaces and missed deadlines in the ones they weren't watching.
Turning user pain into alignment
Consolidating channels usually triggers turf fights. Here, the user evidence did the aligning. Brokers and employers had told every team the same thing, so I framed the stream as the answer to a complaint everyone already owned.
We agreed on one non-negotiable up front: the stream had to feel calm. Dense and stressful would just rebuild the problem in one place instead of five.
Leading the ranking strategy
UX, product, and business ranked the signals together, and I drove the strategy. The key decision was giving each signal a distinct job rather than blending them into one score.
| Signal | Its job | Example |
|---|---|---|
| Urgency | Rank what needs action now | A renewal deadline today sits on top |
| Relevance | Filter to groups assigned to this broker | Other brokers’ groups never appear |
| Recency | Help users recall latest activity, not drive next action | “You last edited this quote yesterday” |
Separating "what's next" from "what happened" kept the top of the stream about decisions, not history.
What we tried and dropped
We started with a standard notification dropdown. The space was too small and the volume made it overwhelming, so it recreated the stress we were removing.
The final pattern shows the top 2 items in full detail with a summary of everything else. Users see what matters now and keep a glance at the rest.
The system we shipped
- Ranked, card-based stream with three archetypes, each with deliberately different visual weight: urgent action, market update, and system message.
- Per-user controls to snooze, mute, or pin categories.
- An empty state worth landing on, confirming nothing is due.
We de-risked it with three rounds of moderated prototype tests, weekly requirement check-ins with product and business, and a two-week guardrail dashboard before broad rollout.
Outcomes
- 86%
- Conversion on urgent items
- 4 clicks · 2m 46s
- To reach the first urgent item
- 86% conversion on urgent items [vs. baseline TBD].
- First urgent item reached in 4 clicks, 2 min 46 s [vs. baseline clicks and time TBD].
- In follow-up research, "which screen do I start on" largely disappeared from feedback.
What I'd tell another lead
"Too many notifications" is almost never a volume complaint. People were fine with forty. They couldn't tell which three mattered right now, so the design problem was ranking, not reduction.
Selected screens
