A Fast Trainee, a Slow Review

The speed is not a benefit. She produces 6 files per hour against a norm of 4 - about 50% more output - but at 94% accuracy. If accuracy is measured per 100 lines, that is roughly 6 errors in every 100 lines; on a 12-minute dictation of about 1,200 words, that is dozens of errors per report. At 4 files per hour with 98% accuracy the unit delivers usable text; at 6 files per hour with 94% it delivers text that someone must re-read, re-listen to and correct, so the true cost is higher and the service level is at risk. Speed bought with accuracy is borrowed time.

Critical versus minor: '15 mg' typed as '50 mg' is a critical error because it changes a dose and can harm a patient - no overall percentage excuses it. Misspelled drug names are major because they can be misread as a different drug. An unflagged [inaudible] is major, because the doctor cannot correct what he is never told about. Spacing and formatting are minor.

Immediate plan: pull back every report she has submitted but not yet delivered and check those line by line before release; move her to 100% review; hold a short feedback session using her own error log rather than a general lecture; put dosage and drug names on a mandatory checklist; and tell the supervisor that, with the client audit next week, no unreviewed text leaves the unit.

Now, in order:

1. Record the report factually and immediately - who saw what, when, which patient, which file - before memory fades. Do not handle it informally.
2. Inform the supervisor or designated privacy officer the same day; the hospital client must be told under the contract, because this is patient-identifiable information.
3. Secure the phone and the chat as evidence. Do not delete anything yourself and do not let the trainee delete it; preserve it for HR and for the client's investigation.
4. Protect the reporting colleague's identity as far as possible. Punishing or sidelining the reporter would guarantee that the next person stays silent.
5. Follow the disciplinary process, refresh the signed confidentiality undertaking, and document the corrective action. Her intention - 'I only wanted help with a sentence' - is not a defence: unauthorised disclosure is the breach, and the employee signed for it on day one.

Systemic changes: no personal phones at production workstations; a named in-house expert or supervisor to answer difficult queries; role-based access so a trainee cannot open files she is not assigned; access audit logs; a secure internal channel for questions; and short, repeated privacy training rather than a one-time signature.

Stage 1 (weeks 1–2): 100% review, with every error logged by type - dosage, drug name, numbers, omissions, formatting. A daily ten-minute feedback session using the trainee's own lines, plus short drills on the two or three error types that dominate. Stage 2 (weeks 3–4): review falls to 50% of files, then 20%, but only after two consecutive weeks at 98% or better on audited files - never on self-reported accuracy. Any critical error resets the clock regardless of the overall score. Stage 3: routine random sampling (about 10% of files, or 20 lines per file) plus 100% review of any file containing high-risk content such as dosages or oncology terms.

Support that raises speed honestly: approved abbreviation and drug-name lists, autotext/expansion for recurring phrases, foot-pedal shortcuts, a standard template, and a quiet zone for dictation. Measure audited lines per hour, not files submitted, so volume cannot be gamed.

Management: a documented QA sheet per file, a monthly error trend reviewed with the client, and enough buffer inside the 24-hour SLA that a file can be rechecked without breaching TAT. Never set piece-rate targets that reward unverified volume - that is what produces a 94% trainee.