“If you want it done right, do it yourself.” It’s an old phrase that Team Select Home Care embraced when looking to deepen AI’s role in its operations.
At the Home Care Innovation Forum, the company’s CEO, Fred Johnson, VP of Technology Meghan Willson, and Chief Clinical Officer Shari Phillips shared the story behind CareSight, a proprietary AI model that predicts hospitalizations in medically fragile pediatric patients three to five days in advance, and the harder story of how they got skeptical nurses to actually believe in it.
Across 16 states, Team Select provides high-acuity, near-hospital-level care in the home, primarily for medically fragile children. The stakes are steep: pediatric patients make up just 0.5% of Medicaid's pediatric census but account for 30-40% of all pediatric healthcare spend, driven largely by hospitalizations. When a child is admitted, the fallout ripples outward: families are displaced, nurses lose income, and hospitals absorb costs that, with the right intervention, are often preventable.
Team Select built CareSight in-house to get ahead of that curve, trained on over 10 years of clinical data across tens of thousands of patients to continuously monitor eight or more vital signs in real time. But before the model could work clinically, the team had to solve a much messier problem first: their data.
Wilson didn't sugarcoat the starting point. Vital signs were arriving unstructured, words where numbers should be, entries that made no sense. The team spent roughly 18 months on data cleansing and rebuilding their data warehouse before the model work could even begin in earnest.
An early piece of advice from a CareSight advisor reframed the scale of the challenge: "You're trying to boil the ocean, which you don't have the time or money for," the advisor told leadership. That led the team to narrow its focus to one "cup of water" first — respiratory diagnoses, which represented both a high volume of cases and the highest frequency of hospitalizations in their patient population.
The technical build was only half the challenge. The real hurdle was adoption. Early on, nurses pushed back hard on a flood of false positives, and the team faced a choice: rush the rollout or listen.
They chose the latter, extending their pilot from three months to six. As Phillips put it, "we knew that we had to really listen and refine the model instead of pushing it out quickly."
That decision paid off roughly three weeks into the extended pilot, when the model flagged a young patient for multiple days of stridor, high pulse, and low oxygen — despite the family and nursing team initially attributing his symptoms to a known pneumonia diagnosis.
Persistent alerts prompted the care team to escalate to the physician, who found a secondary infection that required additional treatment. That case became the proof point that shifted the room. As Wilson described it, nurses had spent their entire careers trying to prevent hospitalizations, and suddenly a technical tool claimed it could help with that instinct. Winning them over meant pairing the model with things nurses actually valued — fewer missed visits, less documentation — rather than positioning it as a replacement for their judgment.
The team was deliberate about fitting CareSight into the care team’s efforts: the model flags risk, but a nurse still reviews the data, assesses the patient, and decides on interventions. "This does not replace clinicians," Johnson emphasized. "It never will."
The results speak to what disciplined, clinically-embedded AI development can look like: what began as a single respiratory model has since expanded to cardiac and infection diagnoses, with the company now reporting that over 90% of its patients are supported by some form of predictive modeling.
Team Select's experience offers a broader lesson for home-based care providers eyeing AI investments: the technology is often the easier half of the equation. The harder, more essential work is building the clinical trust and iterative feedback loops that make frontline staff willing to use it — and believe in it — every day.