PrymeIntelligence

Training teams for AI agents·Playbook

How to run calibration sessions for patient services teams

Calibration for patient services teams: decide the same held rebookings, carer requests and messages alone, compare clinical routing, set a standard.

Patient services teams calibrate on the messages and bookings an agent held: rebooking requests, carer enquiries, cancellations and anything with a possible clinical signal. Each coordinator decides the same anonymised set alone, including whether to route to a clinician. Compare, have the patient services manager and a clinical lead agree the standard, and change the agent's red lines to match.

Why do patient services teams need calibration?

A patient services agent in a hospital or clinic group typically confirms appointments and sends reminders at depth 3 from approved templates, drafts rebookings and replies to patient messages at depth 2, and holds anything that touches clinical matters, consent or records for a coordinator.

Coordinators each draw the clinical line in a slightly different place. One rebooks a patient who mentions their symptoms are worse; another passes the message to the clinical team. One gives appointment details to a patient's daughter; another asks for the patient's consent first. Patients get different answers, and the agent's red lines are taught inconsistently. Calibration fixes both.

The patient services manager owns the sessions. A senior coordinator facilitates. A clinical lead and the information governance lead attend when the standard touches their areas.

Which cases belong in the set?

Eight to ten held items from the past fortnight, anonymised:

  • A rebooking request that mentions worsening symptoms in the last line.
  • A carer asking for appointment details for a parent.
  • A patient cancelling a follow-up after surgery, giving no reason.
  • A request for a copy of a clinic letter.
  • A complaint about a long wait with no clinical content.
  • A transport booking for a patient with mobility needs.
  • A message the agent refused to answer, such as a request for test results.

The session, step by step

  1. Person (patient services manager): selects and anonymises items, with information governance approval.
  2. Person (each coordinator): decides each item alone: the action, whether to route to a clinician, and the reply.
  3. Agent, depth 1: shows its verdict and the red line or bound that held each item.
  4. Person (facilitator): reveals the spread, starting with clinical routing.
  5. Person (coordinators): explain their reasoning.
  6. Person (patient services manager, with the clinical lead and information governance lead): agrees the standard.
  7. Person (patient services manager): sends the standard to the brief owner, including changes to the agent's red lines.
  8. Person (patient services manager): samples held and sent items for a month.
Step The agent does Depth Stays with a person
Select items Nothing 2 or 3 on live messages Governance approval
Decide alone Nothing Unchanged Action, routing, reply
Agent's verdict Shows red line or bound 1 Comparing
Discussion Nothing Unchanged Clinical line, consent
Standard Nothing Unchanged Manager, clinical and governance leads
Update red lines Holds per new standard 2 or 3 Brief owner
Sampling Works live messages 2 or 3 Manager

What does a gold standard look like for patient services?

Clear enough to apply at the desk. For example: "Any message mentioning new or worsening symptoms, pain, bleeding or a reaction to treatment goes to the clinical team the same day, whatever else it asks. The coordinator handles the administrative request separately."

On carers: "Appointment details are shared with a carer only where the patient's record shows consent. Otherwise the coordinator explains why and offers to pass a message to the patient."

On cancellations after surgery: "A cancelled post-operative follow-up is flagged to the clinical team before it is closed, so they can decide whether to contact the patient."

On test results: "The agent refuses requests for results. Coordinators direct patients to the clinician or the agreed results process, never read results out."

What does a session look like?

Take a hypothetical outpatients team of twelve coordinators. On the rebooking with worsening symptoms in the last line, seven rebooked without routing, five routed. All twelve had routed a similar message in training, where the symptom was in the first line. The clinical lead agreed the standard above, and the brief owner widened the agent's red line so it reads the whole message and holds it for clinical routing wherever the words appear.

Which controls does calibration support, and what evidence does it leave?

Clinical routing. Evidence: the red-line standard and sampled messages.

Consent and confidentiality. Evidence: the carer standard, approved by information governance.

Post-operative safety. Evidence: flagged cancellations and clinical responses.

Agent alignment. Evidence: red-line changes and approvers.

What should the patient services manager measure?

  • Agreement on clinical routing. The target is full agreement.
  • Messages with a clinical signal that were rebooked without routing, found by sampling.
  • Carer disclosures without recorded consent. The target is none.
  • Time from a held clinical message to the clinical team seeing it.

Where does it go wrong in patient services?

No clinical lead. Coordinators agree on a line clinicians would draw elsewhere.

Testing only obvious cases. Signals buried in long messages are the ones missed.

Leaving the red lines unchanged. A standard in a team note is forgotten by the weekend shift. A red line in the agent's bounds is not.

Student services teams face a close parallel with welfare signals; see calibration for student services teams. Payments operations staff use the same method on money rather than care; see calibration for payments operations staff. New coordinators should start with the reviewer onboarding checklist for patient services teams, and planning is covered in the AI agent training plan template for patient services teams. The topic guide is training teams for AI agents.

Where the Academy fits

Pryme Intelligence Academy's Operator track is for the people who run a live workspace every day, and its practicum requires two weeks of real review decisions with no missed red line. For patient services, the red line is a clinical signal handled as an admin task, which is exactly what calibration rehearses.

Questions people ask

How often should patient services teams calibrate?

Every two weeks while the agent is new, then monthly, and after any change to clinical red lines.

Should clinicians attend every session?

A clinical lead should attend whenever routing is discussed. They need not stay for administrative items.

Can the agent rebook appointments on its own?

Within approved templates and the clinic's own limits, at depth 3, once certified. Messages with clinical signals are held whatever they ask for.

Train the people who run your agents.

Pryme Intelligence Academy: free courses on the real product, audio lessons, labs in a free workspace, and certificates anyone can check on VerifyDoc.ai. Teams get seats, an admin view and a free readiness assessment.