Hospitals Are Letting AI Handle Insurance Paperwork, With People Still in Charge

Some hospitals are handing insurance paperwork to AI software, and early results look strong. New York's Hospital for Special Surgery says AI agents now complete 1,100 claims a month. Its appeals step, where the hospital challenges a denied claim, fell from 45 minutes to five. People still oversee the work. The reports come from PYMNTS, citing MIT Technology Review Insights and Health Services Daily.

What AI agents are

An AI agent is software that can gather information, finish routine tasks and follow up when a case stalls. Infosys, a technology firm, describes how an insurer could use one. It could check coverage, collect missing documents for a treatment approval or pull together the records needed to review a disputed claim. Agents could also track follow-up care after a hospital stay and flag missed appointments.

What the hospital reports

Dr. Ashis Barad, the hospital's chief digital and technology officer, says the appeal success rate rose from 65% to 100% over nine months. The hospital now handles all claims in-house. It is also rolling out an AI service with developer Ema that helps patients find and book the right clinician. The service weighs symptoms, location, insurance coverage and appointment availability.

Sensitive or uncertain cases go to human specialists. Staff can review the service's decisions and step in at any point. An internal committee also reviews AI projects, with closer scrutiny for anything that could affect patient care.

A lesson from aged care

Wesley Mission, an Australian aged care group, uses three AI agents. One helps with care plan assessments and saves staff 30 minutes to an hour per shift. Another cut the daily task of verifying visits from six hours to about 15 minutes. A third handles calls about sick leave and canceled shifts, finds replacements and recommends a staffing change to a supervisor. All three need human approval before taking final action.

One agent seemed to be broken. The real cause was a staff member not following the process, which left the agent repeating steps. The advice: fix the process first, then add AI.

What is still early

Infosys notes that most member-service projects at insurers are still pilots or early rollouts. Its advice is to start with one focused task. Agents should recommend actions while people make the final call, and they should record what they did and why. Authority should grow only after accuracy, fairness and security are proven.

For patients, the hoped-for payoff is faster answers on claims and more time with doctors and care staff. The reports do not say when most insurers will deliver that.