Nothing here is clinical advice, and nothing here should touch identifiable patient information.
Use these prompts to turn a de-identified missed-appointment extract into work that can be allocated. They are for clinic administrators and service leads working under an approved local non-attendance policy.
Do not paste names, contact details, dates of birth, NHS numbers, free-text clinical notes, or any other identifiable patient information. Use a local case ID and appointment ID that your team can map back inside its approved system.
Key point
Start with the policy
The records tell you what happened. The local non-attendance policy tells you what administrative action is allowed.
Prepare the two inputs
Before you use the prompts, create two separate documents or exports:
- A de-identified missed-appointment extract. Keep one row per appointment. Include the fields named in the first prompt, even where the value is blank.
- The current local non-attendance policy. Include the relevant decision tree, contact-attempt rules, rebooking rules, escalation triggers, owner roles, and stated timescales.
Keep the policy text with the task. A model should not be asked to reconstruct local rules from general practice. If the policy is a long document, paste the sections on non-attendance and the local service appendix. Check that the appendix applies to the service in your extract.
Capabilities and input handling can vary with the model and product configuration. Check the current guidance in the xAI documentation overview before using a workflow with operational data.
Watch out
Remove free text
A notes field often contains identifiers or clinical detail. Exclude it unless your information-governance process has approved a de-identified, administrative extract.
Run the prompts in order
- Run Check the missed-appointment extract. Correct duplicate rows, contradictory attendance outcomes and missing decision fields in the source system where possible. Do not treat a clean-looking table as a corrected record.
- Run Classify each record under local policy. This is the master list. It gives every appointment one primary route: Contact, Rebook, Escalate or Needs verification.
- Split the resulting rows by primary route. Run Create the patient-contact worklist for Contact rows, Prepare the rebooking queue for Rebook rows, and Build the escalation register for Escalate and Needs verification rows.
- Run Reconcile the final follow-up lists using the source extract and all four outputs. Do not distribute work until this reports that the lists are ready.
The route labels are deliberately administrative. They do not describe a person's health need, priority of care, or suitability for any treatment. A safeguarding flag or communication flag may trigger a policy route, but the prompt must cite the policy rule rather than make its own judgement.
Decide which list owns the work
Use the master decision list as the record of why each appointment went where it did. The operational lists are for teams to act on.
| If the record says | Put it in | What the team does |
|---|---|---|
| Contact | Contact worklist | Complete only the policy-approved contact action and record its outcome. |
| Rebook | Rebooking queue | Apply the stated booking action or obtain the required approval. |
| Escalate | Escalation register | Send it to the policy-named owner with the decision required. |
| Needs verification | Escalation register | Correct the data or obtain a policy decision before any action. |
Do not use Needs verification as a quiet holding area. Give it an owner and a next action. If a rule is unclear, the appropriate local policy owner decides what it means.
Check
Check the counts
The total number of primary routes must equal the number of source appointment IDs. Every difference is an exception to resolve, not a rounding issue.
Check whether the output is wrong
Read five to ten rows from each list against both the source extract and the policy. Look for errors that can appear plausible in a neat table:
- An appointment marked as cancelled may have been treated as a DNA.
- A previous-DNA count may use the wrong policy period.
- A row may cite a policy section that does not actually permit the action shown.
- A Contact record may have exceeded the policy contact-attempt limit.
- A Rebook record may already have a booking recorded in the source system.
- An escalation may name a team that is not named by the policy.
- The same appointment ID may appear in more than one active worklist.
Pay particular attention to rows labelled Needs verification. If many records land there, the problem may be the extract design or missing policy text, not the records themselves. Do not force a route simply to reduce the exception count.
Note
Keep a decision trail
Save the policy version used, the de-identified input extract, the master decision list, and the reconciliation result in your approved local record location.
When the workflow does not work
Stop if the policy is missing, the source contains identifiable information, or the output assigns actions without a cited policy rule. Remove the data from the conversation, return to your approved source system, and ask the local policy owner or information-governance lead for the missing direction.
If the lists do not reconcile, fix the source extract or the master decision list first. Then rerun only the affected operational list and the final reconciliation. Do not manually patch a distributed list without recording the corrected route and reason.