The Challenge
A healthcare network serving children with intellectual disabilities, autism, and neglected youth watched caregiver quality collapse after COVID. Hourly caregivers — responsible for feeding, playing with, and supervising children who cannot be left alone — grew disengaged, sometimes ignoring patients for extended periods. The organization tried bonuses, training programs, gift card incentives, and pay increases. Nothing worked. With a 55-year reputation for exceptional care on the line, the CEO came searching for a fundamentally different kind of solution.
What They Built
The AI Lab partnered with computer vision specialists to build a positive-reinforcement recognition system using 300 cameras already installed on campus — identifying exceptional caregiver behavior at 99% accuracy and converting those detections into weekly incentive scores reviewed by managers.
The AI Lab's design premise was inversion: instead of using cameras to catch problems, use them to catch excellence. Working with leading computer vision specialists, they trained a model to identify specific caregiver behaviors associated with high-quality patient interaction — engaged play, attentive feeding, responsive supervision — using the 300 cameras already installed across campus. The model runs continuously and flags positive interactions at 99% accuracy, counting each toward a weekly score per caregiver. At the end of each week, managers receive an AI-generated evidence packet for top performers, review the flagged interactions, and approve incentive pay for those who qualify. No automated payroll changes occur — humans remain in the decision loop at every stage, satisfying compliance requirements for a regulated healthcare environment. The pilot launched approximately 4–6 months into development. Early results showed veteran staff reporting that the culture on campus felt like it did before COVID. The organization is now planning a campus-wide rollout.
AI Role
Veteran staff report care quality returning to pre-COVID levels, visible across the entire campus
Infrastructure
• Computer vision model (trained for positive caregiver behavior detection) • 300 existing on-campus cameras (repurposed) • Custom scoring and evidence portal for managers
Integration Points
• Computer vision model connected to camera feed processing pipeline • Weekly score aggregation feeding manager review portal • Evidence packet generation linked to incentive approval workflow
