Highlights
The hurricane has passed. The power has not returned.
For many members, this disruption is an inconvenience. For a member whose insulin requires refrigeration, one who depends on home oxygen or a CPAP machine, or who cannot reach dialysis because transportation and local services have been disrupted, it may be the beginning of a preventable clinical event.
The member’s diagnosis has not changed. Their risk score may not have changed. Their actual risk has.
The opportunity beyond the risk score is knowing what to do in the moments before a claim tells a payer what happened.
The signal is visible. The member’s clinical vulnerabilities may already be known. Yet most payer risk models were not designed to recognise what happens when a change in the member’s environment intersects with a specific clinical condition—and creates a narrowing window in which intervention could still prevent deterioration, avoidable utilisation, and cost.
It is not simply to find out which individual's risk will escalate and become high-cost. It is to recognise when a member’s trajectory begins to shift, understand the impact of that shift on the individual, and mobilise the right response. Intervention should occur well before the first definitive evidence emerge, such as an emergency department visit, hospital admission, or insurance claim.
Risk changes before the record does
Traditional risk stratification remains essential. It helps payer organisations identify members with chronic conditions, rising utilisation, gaps in care, and a higher probability of future costs. But it often provides a periodic view of risk based largely on events that have already occurred. The moments that alter a member’s trajectory may begin earlier—and outside the traditional medical record.
Signals that may precede claims
Medication cost and affordability, known or undisclosed adverse effects, caregiver disruption, transportation issues, poor air quality, and care after discharge from the hospital can each become a clinical incident depending on the member’s condition and response window.
Five decisions before risk becomes cost
Dynamic risk stratification is not simply a more frequently refreshed score. It is a different way of managing the period between an emerging signal and a potentially preventable event.
From prediction to orchestration
Agentic AI can help link and contextualise emerging signals to known clinical vulnerabilities, prioritise urgency of clinical interventions, initiate approved actions, coordinate outreach, schedule interactions, and escalate members whose needs require clinical judgment.
Consider the power-outage scenario. The goal is not to send the same alert to every member in the affected area. It is to recognise which members may have a clinically meaningful dependency on electricity, transportation, medication storage, or local services—and determine which situations require information, logistical support, clinical outreach, or immediate escalation.
One member may just need a message. Another may need help with an alternate power source for oxygen supply or a replacement for a compromised medication. A third may require clinical assessment by a trained medical professional.
Important distinction
Agentic AI does not replace the care manager. It enables ensuring the care manager reaches the right individual, with the right context, at the right time.
Signal |
Clinical context |
Agentic orchestration |
Human judgment/outcome |
Power outage/disrupted services |
Insulin, oxygen, CPAP, dialysis, complex post-discharge needs |
Prioritise urgency, initiate approved outreach, coordinate resources, schedule touchpoints |
Clinical assessment, escalation, stabilisation at home, avoided ED/admission |
Access or affordability change |
Medication adherence or treatment continuity risk |
Trigger benefit/pharmacy pathway, resource navigation, follow-up sequencing |
Care plan adjustment, member support, cost avoidance |
Behavioural or environmental drift |
Respiratory, cardiovascular, frailty, caregiver dependency |
Monitor trajectory, route to appropriate program, escalate exceptions |
Earlier intervention before deterioration |
From activity to impact
Most payer organisations already have care-management and disease-management programs. But cases can still sit in queues, interventions can vary, and clinicians can spend substantial time gathering information and coordinating routine tasks. Meanwhile, the member’s trajectory continues.
A dynamic clinical operating model allows routine needs to move through digitally enabled pathways. Meanwhile, it prioritises complex or urgent circumstances for human intervention. Measuring success differs considerably with the dynamic operating model.
The shift is from measuring whether work was completed to measuring whether the member’s trajectory changed.
Instead of measuring call attempts, the measure pivots from assessments completed to preventing disease progression; from members enrolled to members remaining stable at home; from cases closed to avoided utilisation; and from interventions delivered to timely action before the intervention window closes, ultimately reducing the total cost of care.
Payer organisations do not lack risk scores, predictive models, or talented clinical teams. The larger opportunity is connecting those capabilities around the way risk actually unfolds.
Risk changes with the interaction of clinical conditions, therapy, access, cost and affordability, environment, patient and care-giver behaviour, and daily life. Speed of addressing risk varies at different rates and creates intervention windows that may remain open for months, days, or only hours.
Dynamic risk stratification recognises meaningful trajectory changes.
Agentic AI orchestration responds at scale while the window is still open.
Clinical professionals apply judgement and expertise to change the outcome.
A claim can tell an organisation what happened. The greater opportunity lies in understanding how to act before those signals emerge.
Agentic AI-led clinical model is built around how risk actually unfolds.
I want to thank Ramesh Gudalur for his ideas and contribution to shaping this blog.