Smart Choice
A scoring-based provider recommendation system built from zero — helping members find the right provider, not just the nearest one.
Members had a provider search tool. They just didn't trust it.
UnitedHealthcare's Find Care tool let members search for providers — but search alone wasn't enough. Members defaulted to whoever was closest, or whoever their friend recommended. Quality signals, cost transparency, and personal fit weren't surfacing at the right moment. Smart Choice didn't exist. We had to define what it should be, what it should show, and how a scoring recommendation could earn trust in a high-stakes healthcare context.
What I owned
- End-to-end UX from concept to launch
- Recommendation logic surface design
- Smart Choice badge system and scoring UI
- Data science partnership — translating model outputs into human-readable signals
- Usability testing across 3 rounds
- Ongoing iteration based on member feedback
"Launching to a limited member population wasn't just a business constraint — it became a design principle. We couldn't optimize for scale yet, so we optimized for comprehension. And comprehension revealed a signal problem: the model surfaced dozens of data points, but the more we explained the recommendation, the less confident members felt. We stripped the surface down to three words — Simple. Customized. Data-driven — and let the badge carry the trust instead of the explanation."
Three decisions that defined the product
The badge lives on the card, not the detail page
Early explorations put the Smart Choice explanation inside the provider detail page — after the click. Testing showed members needed the signal before they committed to learning more. Moving the badge to the search result card increased click-through on Smart Choice providers by 32%.
Fewer signals, more confidence
The data science model had dozens of inputs. First designs tried to surface them all — quality scores, cost rankings, specialty match, distance weighting. Members felt overwhelmed. We consolidated everything into three: Simple. Customized. Data-driven. Confidence scores went up. Explanation length went down.
The introduction modal was intentional friction
We could have launched Smart Choice silently — just badges appearing in results. Instead we designed an explicit introduction moment. For a first-of-its-kind recommendation engine in healthcare, members needed to know what it was and that they were still in control. 'Maybe later' was as important as 'Learn more.'
What shipped and what it changed
Smart Choice launched to an initial member population and has been iterating based on real usage ever since. The constrained rollout wasn't a limitation — it was the product strategy. Shipping to fewer members meant every signal mattered, every piece of feedback was actionable, and every design decision could be validated before scaling.
If we had another quarter
I would have pushed for a persistent Smart Choice filter on the results list — not just a badge on individual cards. Members who trusted the recommendation enough to click once were ready to see only Smart Choice providers. That affordance didn't make v1, but it's the natural next step. I'd also have invested more in the empty state — what happens when no Smart Choice provider exists in a member's area. We solved for the happy path first. The edge cases are where trust actually breaks.