Chelsey Wiley
AI · DECISION SYSTEMS
When the brief is wrong: reframing AI triage as a decision platform.
Baylor Scott & White Health asked for a symptom checker. The research showed it would have repeated a failed pattern. I led the reframe into First Contact Navigation — a care platform integrating AI triage, Epic EHR, and clinical follow-up.
CLIENT
Baylor Scott & White
ROLE
Product Manager
APPROACH
Research-led product
INTEGRATIONS
98Point6
Epic EHR
30%
of users named the EHR-integrated appointment view as a deciding factor in their care choice
90%
of patients already searched their symptoms online before seeking care
80%
felt neutral or worse after searching their symptoms
20
research participants whose findings reframed the entire scope
EXECUTIVE SUMMARY
Baylor Scott & White Health asked the product team to build a symptom checker to reduce unnecessary ED visits — 37–43% of which are deemed medically unnecessary. I declined to ship the solution as briefed. The research showed that 90% of patients already searched their symptoms online, and 80% felt neutral-to-worse after doing so. A symptom checker would have repeated a failed pattern at greater cost.
I led the reframe into First Contact Navigation — a care navigation platform integrating Baylor's Epic EHR, an AI triage layer (98Point6), and clinical follow-up. The product shifted volume to telehealth and urgent care, and surfaced socio-economic drivers of ED use that the original brief had hidden. 30% of users specifically named the EHR-integrated appointment view as a deciding factor in their care choice.
THE BUSINESS PROBLEM
Excellent local work, compounding into a globally costly system.
The Digital Health Organization had a clear, measurable goal: reduce non-emergent ED visits, which represent 37–43% of all ED traffic nationally. The proposed solution was a symptom checker integrated into MyBSWHealth — a tool that would tell patients whether their symptoms warranted an ED visit.
The brief was crisp. The metric was clear. The budget was approved. It was also, on closer inspection, a solution looking for a problem.
THE STRATEGIC DECISION
Validate the brief before architecting against it.
I conducted information-gathering research with 20 participants across genders, ages, and insurance brackets — all of whom had visited an ED in the past six months. The findings reframed the entire project.
RESEARCH FINDINGS THAT REFRAMED THE SCOPE
abandoned their care-seeking after a symptom search.
2%
felt neutral or worse after searching their symptoms.
80%
90%
of patients already searched their symptoms online before seeking care
Patients didn't need a better symptom checker. They needed direction on where and how to get treatment — guidance that started after the symptom search, not before it.
The strategic decision was to push back on the brief, take the research to the Digital Health Organization, and propose a different product entirely: a care navigation layer that started with triage and ended with concrete care options across the BSW system.
"
The most consequential product decision was made before any design started: refusing to build the wrong thing.
- REFLECTION · FIRST CONTACT NAVIGATION STRATEGY
THE SYSTEM INTERVENTION
Build the decision layer between patient and care.
Reframing the ask meant negotiating with executive stakeholders who had already approved the symptom-checker solution. I led that conversation with research evidence and a clearer model of patient behavior, and the org agreed to expand scope.
We partnered with 98Point6 for the AI-driven clinical triage layer and built the integration into BSW's Epic EHR. I led the cross-functional work between the Digital Health Organization, internal development, the 98Point6 integrations team, and the Appointments & Scheduling team responsible for mapping triage outcomes to care recommendations.
A four-layer decision architecture: patient entry feeds the AI triage layer, which routes into three care pathways — all unified by a single Epic EHR record.
The reframe turned a feature request into a platform.
OUTCOMES & MEASUREABLE IMPACT
What changed, in numbers.
30%
of users in VOC feedback specifically cited the EHR-integrated appointment view as a factor in their care decision.
=
ED visit rates remained roughly unchanged at 3 months — a strategic insight, not a metric miss. The drivers of non-emergent ED use are socio-economic, not informational.
↑
Telehealth call volume and urgent care visits both increased post-launch — patient volume shifted toward appropriate alternatives to the ED.
∞
The diagnostic finding now shapes how the business scopes future ED-reduction initiatives — a permanent shift in problem framing.
WHAT THIS DEMONSTRATES
Treat the brief as a hypothesis. Validate it before you architect against it.
The original brief, if shipped, would have generated a feature, a launch, and an unchanged ED metric — with no diagnostic insight into why. The reframe turned a likely failure into a platform that surfaced the real problem and shifted patient volume toward more appropriate care.
That's the pattern that matters at AI-platform scale. AI doesn't deliver value as a feature; it delivers value as a decision layer integrated into operator and customer workflows. The same discipline now sits at the center of my systems work: research the actual decision being made, build the orchestration and governance around it, and have the standing to push back when the brief and the evidence disagree.