The line, stated plainly
Software intended to inform clinical decisions is generally a regulated medical device, with the approval, evidence and liability that implies. That is a different category of project from anything discussed here.
Everything below is administrative: it changes who types what, not what a clinician decides.
Correspondence and documentation
Clinic letters, referrals and discharge summaries drafted from structured data and clinician notes, for review and signature. The clinician remains accountable for the content, which is the essential control.
The measurable benefit is turnaround time. Letters that took a week to dictate, type and check can go out the same day, which affects patient experience and onward referrals.
Coding support
Suggesting codes from documentation, with the coder deciding. Coding accuracy affects funding and reporting, and suggestion with human confirmation improves consistency without removing accountability.
This works well because the output is checkable in seconds by someone qualified, which is the property that makes AI safe to deploy.
Scheduling and capacity
- Predicting no-shows from historical patterns to inform overbooking policy
- Waiting-list management and automatic backfill of cancellations
- Reminder sequences that reduce non-attendance
- Clinic template optimisation based on actual appointment durations
No-show prediction should inform policy rather than individual treatment — using it to deprioritise particular patients raises equity concerns that outweigh the efficiency.
Data protection is the first constraint
Health data carries heightened obligations. Processing location, access control, retention, audit logging and the contractual position with any model provider all need settling before anything is built.
For many organisations this means processing within a specific jurisdiction under an agreement that excludes training on the data, plus internal access control by role.
Where the value actually lands
Administrative burden is a substantial and well-documented pressure in healthcare. Reducing documentation time returns clinical capacity, which is worth more than any efficiency metric.
That is the case to make internally: not cost saving, but clinician time returned to patients.