Standard Operating Procedure · SOP-QC-001
Onboarding Quality Control operators
A 30-day path from day one to certified independent review. Built around a single principle: nobody touches live decisions until their judgment has been measured, not assumed.
01Purpose
To bring a new Quality Control operator from zero to independent, audited review capability in 30 days — with a documented accuracy baseline before they carry live caseload, and a named owner for every step of the ramp.
This SOP exists because the alternative is the default: a new hire shadows someone for a week, gets handed a queue, and their real accuracy is discovered a quarter later in an audit. That is expensive in exactly the way this function cannot afford.
02Scope
- Applies to: all new QC operators, internal transfers into QC, and any operator returning after 90+ days away from live review.
- Does not cover: role-specific tooling permissions (see IT access request process), HR pre-boarding, or payroll setup.
- Prerequisite: signed confidentiality agreement and completed data-handling training before day 1. No exceptions — this gate is legal, not procedural.
03Roles & responsibilities
| Role | Owns | Accountable for |
|---|---|---|
| Talent Manager | The ramp plan, certification decision, escalations | Operator is certified or exited by day 30 — no indefinite ramp |
| Assigned Mentor | Daily contact, shadowing, first-line questions | Operator is never blocked more than half a day |
| QC Lead | Calibration sessions, blind-set scoring | Accuracy measurement is honest and consistent |
| Operator | Their own reason codes and escalation discipline | Flagging uncertainty rather than guessing |
"Ask anyone on the team" reliably means asking no one. A single named mentor with an explicit half-day response expectation is the cheapest intervention available and the one that most consistently moves week-one accuracy.
04Procedure
Phase 1 · Days 1–5 — Foundation
Day 1 — Context before tooling
Why this function exists, what a bad decision costs downstream, and who is harmed by a false accept and a false reject. Tooling comes after the operator understands the stakes; it lands differently once they do.
Days 2–3 — Regulatory grounding
KYC/AML obligations, what the regulator actually requires, and where the organisation's obligations begin and end. Operators who don't know the rule cannot apply judgment to its edges.
Days 4–5 — The review sequence
The checkpoint order, the reason-code taxonomy, and the escalation path. Operator runs the sequence aloud against known-good examples until the order is automatic.
Phase 2 · Days 6–15 — Supervised review
Days 6–8 — Shadow
Operator observes the mentor and narrates what they'd decide before the mentor decides. Discussion focuses on the reasoning, not the verdict — a right answer for the wrong reason is a future failure.
Days 9–12 — Reverse shadow
Operator decides; mentor reviews every case before it commits. Nothing reaches production without a second signature. Disagreements are logged, not smoothed over.
Days 13–15 — First blind set
Operator scores a curated set with known outcomes, unaided. This is the first honest accuracy number and it is diagnostic, not disciplinary — its purpose is to target the back half of the ramp.
Phase 3 · Days 16–25 — Calibrated independence
Days 16–20 — Live queue, sampled
Operator carries live caseload at reduced volume. 100% of rejects and 20% of accepts are audited. Reject-heavy sampling is deliberate: a false reject harms a real person and rarely complains.
Day 20 — Calibration session
Operator joins the full-team calibration on contested cases. Hearing experienced reviewers disagree with each other is the fastest available lesson in where genuine ambiguity lives.
Days 21–25 — Volume ramp
Volume steps toward full load while audit sampling holds. If accuracy degrades as volume rises, the ramp pauses — throughput is never allowed to buy itself with accuracy.
Phase 4 · Days 26–30 — Certification
Day 26 — Second blind set
A fresh curated set, scored unaided. Compared against the day-13 baseline to measure the ramp itself, not just the endpoint.
Day 28 — Review board
Talent Manager, QC Lead, and Mentor review the full record: blind sets, audit log, escalation behaviour, calibration participation. Certify, extend once, or exit.
Day 30 — Certified or extended
Certified operators move to standard audit sampling. Extensions are capped at one 15-day cycle with a written gap plan — an operator who cannot certify in 45 days is a hiring problem, not a training problem.
05Certification thresholds
| Measure | Threshold | Rationale |
|---|---|---|
| Blind-set accuracy | ≥ 95% | Below this, live caseload costs more in audit than the operator returns |
| False accept rate | ≤ 1% | The expensive error — it clears the gate and lands downstream |
| Reason-code completeness | 100% | An undocumented decision cannot be audited or appealed |
| Appropriate escalation | No missed mandatory escalations | Knowing the limit of your own authority is the job |
| Throughput | ≥ 80% of team median | Deliberately last. Speed is trainable; judgment is the constraint |
Accuracy thresholds are published alongside throughput, never instead of it. The moment throughput is reported alone, operators correctly infer which one actually matters — and accuracy starts paying for speed within a fortnight.
06Post-certification
- Ongoing audit: 100% of rejects reviewed for the first 30 days post-certification, then dropped to standard team sampling.
- Monthly calibration: mandatory for all operators, not just new ones. Drift is universal and invisible from the inside.
- Quarterly blind set: everyone, including leads. An accuracy number nobody re-measures is a number nobody has.
- Threat refresh: as new attack patterns are identified, updated within the cycle they're found — not at the annual review.
07Measured outcomes
This structure — mentor-led ramp, measured baselines, published thresholds — is the approach behind the 25% performance lift I delivered across my QC team, and the environment that sustained a 98% compliance standard.
08Revision history
| Version | Date | Change | Author |
|---|---|---|---|
| 1.0 | 2026-07 | Initial issue | J. M. Tapia |