How a Healthcare Network Reversed Caregiver Decline
A healthcare network's CEO wired existing campus cameras to a 99%-accurate computer vision model — flagging exceptional care in real time and paying caregivers for it.
99%
Caregiver behavior recognition
The Challenge
A major healthcare network serving children with intellectual disabilities and autism experienced a steep post-COVID decline in caregiver quality. Despite trying bonuses, training programs, gift cards, and pay increases over multiple years, the same problems persisted: caregivers were disengaged, sometimes neglecting patients entirely. Leadership needed a fundamentally different approach to identifying and reinforcing exceptional care — one that moved beyond punishment and surveillance toward positive, incentive-based behavior change.
What They Built
Ryan Kurt and The AI Lab partnered with the organization's CEO to design and implement a computer vision system that monitors existing campus cameras to identify and reward exceptional caregiver behavior in real time. Using 99% accurate multimodal AI, the system flags positive interactions, which feed into an incentive-based pay model — caregivers get paid more for demonstrably great care. Human reviewers handle ambiguous cases and approve all pay decisions for compliance. After two and a half years of iteration, including a failed first attempt with earlier-generation models, the system has exited pilot phase into broader rollout.
Ryan Kurt and The AI Lab began with the organization's CEO to define a measurable behavioral outcome: identifying and rewarding exceptional caregiver interactions, not surveilling for failures. The first attempt, using earlier-generation AI models, failed to meet accuracy thresholds. The team iterated over two and a half years as model capabilities improved. The production system integrates with the facility's existing camera infrastructure — no hardware replacement required — and runs a multimodal AI layer achieving 99% accuracy in flagging positive caregiver interactions. Flagged moments flow into an incentive-based pay model: caregivers receive higher compensation for demonstrably great care. Human reviewers handle ambiguous or borderline cases, and all pay decisions require human approval to meet compliance requirements. The system exited pilot phase and entered broader rollout after two and a half years of iteration, with staff across departments — including employees with 20 to 30 years on campus — reporting a measurable shift in daily care quality and engagement.
Infrastructure
Custom multimodal AI/computer vision system (99% accuracy) • Existing campus camera infrastructure (no new hardware) • Incentive pay calculation and tracking system • Human review interface for ambiguous case adjudication
Integration Points
Campus cameras → multimodal AI processing layer → behavior classification • Positive interaction flags → incentive pay model → caregiver compensation adjustments • Ambiguous flags → human review queue → adjudication and approval • Pay decisions → HR/payroll system (human-approved before execution)
Impact
Care Culture Visibly Transformed
Staff and long-tenured employees (some with 20–30 years on campus) report a palpable shift in the quality of daily care. Where caregivers were previously disengaged or distracted, the campus now has a measurable sense of attentiveness and engagement — a culture shift that had eluded the organization for years despite conventional HR interventions.
99% Accuracy in Behavior Recognition
The computer vision system achieves 99% accuracy in identifying positive caregiver interactions, with a human-in-the-loop review process for ambiguous edge cases. This precision level was not achievable with earlier multimodal models and required two and a half years of iteration to reach — enabling the incentive pay system to function reliably and fairly.
Incident Reduction (Target Metric in Progress)
The organization's north-star metric is a measurable reduction in adverse incidents — outbursts, escape attempts, and other patient safety events — which correlate directly with caregiver engagement. While the rollout is still in early stages and sufficient time-series data has not yet accumulated, behavioral indicators are positive.
Implementation Complexity
Best Fit For
Mid-market CEOs ($100M–$4B in revenue, ~100–1,000 employees) in healthcare, insurance, professional services, or events/media who are personally committed to AI-driven transformation — not delegating it — and willing to invest in the internal ownership, data infrastructure, and cultural readiness required before deploying AI solutions.
