How AI risk scoring works inside AILS batch release — what the AI Risk Scorer evaluates, how it generates a recommendation, and why QA still controls every final call.
GMP batch release under 21 CFR 211 requires documented review of all production and quality records before a batch is distributed. AILS adds an AI Risk Scorer to that review — without removing QA from the decision.
All analytical results, in-process checks, and yield calculations are entered in AILS. The COC is linked to the batch. No release workflow can be initiated until the batch record is complete — enforced at the system level, not by a checklist.
👤 QA / Lab Tech actionWhen QA initiates the release review, the AI Risk Scorer automatically reads the batch record. It evaluates all results against your specifications, checks for open deviations or OOS events linked to this batch, reviews equipment calibration status for instruments used, and verifies analyst competency currency for each test performed.
🤖 Automated — no action requiredThe scorer outputs a structured risk recommendation: Release, Release with Conditions, or Hold — with the specific data points that drove the recommendation listed in order of weight. Every input to the recommendation is traceable to a source record in the system.
🤖 AI output — presented to QA for reviewQA reviews the scored recommendation and the underlying evidence. They can agree, override, or add a conditional hold. The system records the QA decision alongside the AI recommendation — so the audit trail shows both what the AI suggested and what QA decided. Disagreements are documented, not hidden.
👤 QA decision required — cannot be bypassedBefore the batch release signature is executed, AILS requires re-authentication — the QA manager must enter their credentials again, separate from their session login. This satisfies 21 CFR §11.10(g) authority checks. The signature captures name, title, timestamp, IP, and the record state at the moment of signing.
👤 QA authenticated signature — cannot be delegated to AIThe signed batch release record is locked with a SHA-256 hash. It is permanently stored in the 5-year immutable audit trail. Any attempt to alter the record after signing is detectable via the hash verification endpoint.
🤖 System-enforced — automaticThe AI Risk Scorer doesn't generate a number from nothing. These are the six data points it reads from your live AILS tenant for every batch release evaluation.
Every result on the batch COA is checked against your documented acceptance criteria. Any result at or near a specification limit is flagged, not just those outside it.
Any open deviation record linked to this batch or to the instruments used is surfaced. Deviations with no corrective action closure automatically elevate the risk score.
Every instrument used in testing this batch is checked against its calibration record. Any instrument that was out-of-calibration during the test period is flagged as a hold risk factor.
The training record for each analyst who ran tests on this batch is verified. An analyst whose method-specific competency was expired at the time of testing triggers a hold recommendation.
Any Out-of-Specification or Out-of-Trend event associated with this batch is checked for a completed investigation. An open OOS with no root cause conclusion automatically results in a Hold recommendation.
The scorer checks previous batches of the same product for trending issues — repeated near-limit results, recurring deviations, or patterns that individually pass but collectively indicate a process drift risk.
AILS is designed so QA always has the final call — but the system documents overrides clearly to maintain the integrity of the audit trail. Some override scenarios require additional justification fields before the signature can be executed.
Book a 30-minute demo and we'll walk through a batch release decision — risk score, QA override controls, and the audit trail — in real time.