AI System for Routing Patient Complaints to the Appropriate Unit

The problem

In day-to-day operations, a complaint covering several topics may be delayed within a single unit. When the complaint text, channel, date, unit and category dictionary are scattered across different screens, documents or team records, problems do not become visible in time.

Fragmented data delays preparation of the complaint routing draft and forces staff to search manually for supporting evidence again. Before making a decision, the patient experience or quality manager has to complete missing fields and verify inconsistencies one by one.

A common exception is that administrative text may require a privacy or patient safety review. Incomplete or incorrect automation that overlooks this context could treat a critical complaint as a routine request, so the output cannot be applied directly.

What we set out to improve

• Rate of routing decisions changed by the responsible staff member
• Rate of outputs corrected or rejected by the patient experience or quality manager
• Time between the source record and the review output

How the system works

1. 01 Define the scope of the review and the authorised patient experience or quality manager role
2. 02 Collect the inputs: complaint text, channel, date, unit and category dictionary
3. 03 Check source identity, date, version and data freshness
4. 04 Extract separate administrative topics
5. 05 Match topics with units and response periods
6. 06 Check the exception: administrative text may require a privacy or patient safety review
7. 07 Place low-confidence or inconsistent records in a separate human review queue
8. 08 Write the source identity, rule, output and time to the audit trail
9. 09 Have the patient experience or quality manager approve, correct, defer or reject the output

Methods we used

• Multi-label classification of administrative topics
• Source, format, version and required-field checks for complaint text, channel, date, unit and category dictionary
• A missing-data warning instead of an estimate when inputs are missing or inconsistent
• Checking exceptions separately from the main rule
• Storing the source identity, rule result, complaint routing draft and human decision together

Where people stay involved

Data and security

Access to the complaint text, channel, date, unit and category dictionary is limited to the minimum permissions required for each role. Identifiers and sensitive fields are masked where possible, and only the necessary section is sent to the model rather than the entire raw file. Access, outputs and the decision by the patient experience or quality manager are recorded; the system does not enter production until the organisation has approved the retention period and data location.

Who this suits

Frequently asked questions