Clinic practice managers review an agent's decisions by reading its record, finding the deciding bound, and checking two lines the agent must not cross: clinical judgement and patient data. Inside those lines, they approve, edit or decline with a reason. Outside them, they route to a clinician or the data protection lead.
What lands on the practice manager's desk?
A clinic agent's decisions fall into three groups.
Front office: bookings, cancellations, reminders, waiting list moves, replies to patient messages. Most run at depth 3 within the clinic's limits; exceptions are held.
Back office: invoices, payments from patients and insurers, stock orders, supplier queries. Routine items run at depth 3; anything above budget or unusual is held.
Information requests: patients or third parties asking for records, letters or copies of results. These touch special-category health data and should always be held.
The practice manager reviews held items daily and samples the rest. The record shows every decision, so nothing the agent did is invisible.
What are the steps?
- Read the verdict and the deciding bound. A hold on the clinic's own limits, such as "bulk changes over ten appointments", means a policy threshold was reached. A hold on the source bound means the agent lacked a fact, such as a missing referral.
- Ask whether the decision is clinical. If it depends on how urgent a patient is, what a result means, or whether a treatment should go ahead, route it. The practice manager does not decide clinical questions and neither does the agent.
- Check patient data handling. Where is the agent sending information? Are the contact details from the clinic's own records, or newly supplied? Is the requester entitled?
- Check the facts and the policy. Fees, cancellation rules, supplier terms, rota rules.
- Decide. Approve, edit or decline.
- Record the reason. Short and specific.
- Note repeats. A correction made every week belongs in the agent's limits, not in the practice manager's routine.
What may the agent do, and what stays with the practice manager?
| Work | Agent | Practice manager |
|---|---|---|
| Book and remind within rules | Depth 3 | Samples |
| Move a waiting list | Drafts, held | Decides, with clinician for urgency |
| Reply to routine patient messages | Depth 2 or 3 | Samples |
| Reply to complaints | Drafts, held | Decides; clinical points to clinician |
| Release patient records | Never alone | Decides with data protection lead |
| Order stock within budget | Depth 3 | Samples |
| Give clinical advice | Never | Routes to a clinician |
The hard limits are spelled out in what clinic practice managers should never let an AI agent do.
An example: the no-show policy
Take a dental clinic whose agent applies a missed-appointment fee at depth 3 under the clinic's policy. During a weekly sample, the practice manager notices the agent charged the fee to a patient who had phoned to cancel. The record shows the cancellation was taken by a receptionist and noted in free text in the patient file, which is not one of the agent's sources; the booking system still showed the appointment as booked.
The agent was correct on its sources. The process was not. The practice manager reverses the fee, records the reason, and changes the front desk routine so phone cancellations are recorded in the booking system itself. They check the past month for the same pattern and find two more. The fix was in the clinic, not the agent, and only a review of allowed decisions would have found it.
Which controls hold the review together?
- Clinical questions always route to a clinician.
- Information requests are always held.
- Allowed decisions are sampled weekly, with the method recorded.
- The practice manager's own approvals are spot-checked by the clinic owner or lead clinician.
Training the wider team is covered in how to train the practice team to review AI agent decisions. For market detail, see AI agent training for clinic practice managers in healthcare in the Netherlands. Whether the role changes shape is answered in will AI agents replace clinic practice managers. The role map is in AI agent skills by role.
Where the Academy fits
A practice manager is the person Pryme Intelligence Academy's Operator track was written for: someone running a live workspace every day. Its modules cover human review, reading the workspace, keeping knowledge current and running the rhythm. Module tests are set in the learner's own industry and role, so a practice manager is tested on clinic cases.
Questions people ask
How many decisions should the weekly sample include?
Enough to cover each type of work the agent does: some bookings, some messages, some payments. Increase it after any change to the agent or the clinic's processes.
Should clinicians review the agent's decisions too?
They should review anything clinical that reaches them and give feedback on patterns, such as urgent patients being rebooked too late. Administrative review stays with the practice manager.
What if the agent's decision was right but the outcome was poor?
Fix the outcome, then look at the process or limits behind it. The record helps separate an agent problem from a clinic problem.