What is health care automation?
The AI Glossary
In our classes, we often pause to explain the ideas behind AI and the language used to describe it. These posts grew out of those conversations. They’re for anyone who would like to understand AI better, with no technical background assumed.
Health care automation uses software to carry out parts of clinical or administrative work with less step-by-step manual effort. It may use fixed rules, AI or both. Sending a reminder at a scheduled time does not necessarily require AI; interpreting a varied collection of referral letters may benefit from a learned model.
Automation is a property of the workflow, not a promise of complete independence. A system can prepare information for review, perform a narrow action under standing authorization or handle a sequence until an exception requires a person. These arrangements should be described separately.
Start by identifying the actual task. “Automate referrals” is too broad to evaluate. “Identify missing documents and prepare a follow-up for an authorized coordinator” specifies an activity, a boundary and a possible result. It also makes the remaining human work visible.
Administrative does not automatically mean low risk. A scheduling error can delay care. A wrong eligibility decision can prevent access. An incorrect bill can harm a patient financially, and a misdirected message can expose private information. Risk depends on consequences, reversibility, detectability and the people affected—not simply the department using the software.
A second issue is where the work goes. Time saved in one step may create checking, correction or exception-handling work elsewhere. Staff may need to monitor a queue that did not previously exist. A meaningful assessment follows the task through to completion and includes that additional effort.
For example, automating requests for missing referral documents might reduce manual preparation but generate repeated messages to the wrong office. Counting messages sent would make the system look productive while concealing the failure. Better measures include completed referrals, elapsed time, duplicate contacts, correction effort and patient experience.
Automation can also change roles and expectations. The organization should decide how any capacity released will be used, rather than assuming that software alone will reduce workload or improve care. The aim is a better service, with dependable handoffs and clear responsibility, not simply fewer visible human steps.
Sources: Anthropic — Building Effective Agents (2024); ASTP/ONC — Health IT Playbook: Electronic Health Records
To learn more, take a look at the AI in Health Care Certificate of Specialization from the Advanced Learning Academy at Harvard T.H. Chan School of Public Health. You can take an individual course or work toward the full certificate.
AI is a fast-moving field. This explanation reflects our class discussions and related material at the time of writing. Terminology, capabilities and tools may change.
For suggested additions or corrections, please email me at tpanch@hsph.harvard.edu.
Dr. Trishan Panch
Program Director, AI in Health Care
Harvard T.H. Chan School of Public Health — Advanced Learning Academy