EBR with AI — the batch ships in hours because QA only sees what departs from what was approved.

The problem

A paper batch record (or a signed PDF) is not traceability — it is archaeology.

In a GMP plant the batch release cycle takes 5 to 15 days on average because QA has to reconstruct, page by page, what happened in production. MES data on one side, SCADA ticks on another, LIMS results arriving late, the operator's notebook with pen annotations, and the photo of the encapsulator panel that the supervisor took because the machine is from 1998 and exposes nothing to the outside.

1. Every batch triggers a full QA review — the conforming ones and the problematic ones take the same time, because the system cannot tell them apart until somebody looks at them by hand.
2. WIP piles up in quarantine waiting for release, tying up capital and breaking the shipping plan.
3. Every product change, change control or deviation adds documents to the batch record stack. More paper is not more control — it is more places to hide an error.

And the day an inspection arrives, QA runs a weeks-long sprint to rebuild the dossier of a batch from two years ago, because the signatures were in format A and the analytical data in format B, and three template changes happened in between. The system works — and every release is a toll in time that nobody should still be paying.

How it fits the IRIS system

iLEAN does not replace the MES — it seals the cracks between the systems that already sign off your production.

The batch record suffers because batch information lives on islands: MES, SCADA, LIMS, ERP, the production manager's spreadsheet, the operator's notebook, the panel of the old capsule machine with no interface. iLEAN acts as the putty between all those pieces: it does not ask you to change the validated MES, nor to retire the 1998 encapsulator. It collects what each system knows, cross-checks it with the product's master template, and generates the consolidated EBR with a Part 11 audit trail.

The Tracer reads wherever it has to. The Writer assembles the batch record. The agent cross-checks against the master recipe and the validation, and flags only the exceptions. The QA person signs — never the other way round.

The three iLEAN pieces applied to review-by-exception EBR:

• iLEAN Tracer (Connect) — captures the real process data by any route: a modern API on the new MES, a SCADA file, screen reading from the old encapsulator, a photo of the panel, the operator's dictation, an email from the external lab with the analysis result. Capture graded to each machine: integrated where there is an API, intermediate where there is an old local PC, manual where there is only an analog panel — it does not force you to change anything.
• iLEAN Writer — maintains the master batch record template per product/line and fills it in for every batch with the data that came from the Tracer. The output is a structured EBR, signed with a Part 11 audit trail, ready for review.
• iLEAN Agent — cross-checks every step of the batch against the master recipe, the validation ranges and the limits of the change control in force. If everything adds up, the step is marked automatically conforming; if there is a deviation, it prepares the exception dossier for QA with a preliminary root cause, the associated data and the visual evidence. QA focuses attention only on what is worth looking at.

See the full IRIS architecture →

Before and after

Paper batch record vs. review-by-exception EBR with iLEAN

Aspect	Classic batch record (paper/signed PDF)	With iLEAN Writer + Tracer + Agent
Source of the batch data	MES, SCADA, LIMS, notebook — each on its own	Consolidated at second zero by the Tracer
QA review	Page by page, batch by batch, conforming and exception alike	Only the exceptions — what is conforming is accepted because it is validated
Average release time	5-15 days	Hours on batches with no exception, ≥30% less on the rest
Part 11 audit trail	Manual signatures + dates reconstructed after the fact	Native, cryptographic, non-editable
FDA/EMA inspection preparation	A QA sprint rebuilding the dossier of an old batch	Per-batch dossier already assembled; the inspection reads it directly
Old machine with no interface	The operator photographs the panel and types it into a spreadsheet	The Tracer reads the screen; the data enters the EBR

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

The block below is an estimate to be validated with the specific data of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• GMP plant for oral solids or steriles, 2-4 lines, hybrid EBR (part paper, part signed PDF, part vertical MES).
• First product migrated to review-by-exception EBR with iLEAN: first value in a few weeks, usually with the Pareto line that moves the most volume.
• Expected reduction in average release time: ≥30% in the first 6 months; higher as more products are brought in.
• Indicative payback between 4 and 9 months, dominated by WIP trapped in quarantine and the cost of delayed shipments to the customer.
• Quality standard in adjacent sectors already operating on review by exception (automotive): 25 PPM of acceptable defects ^[1]. That is the order of demand pharma is aligning with.

And the quality director's reasonable doubt

“What if the AI invents a piece of batch record data and QA signs it by mistake?” — hallucination is a problem of free generation, not of anchored tasks. The EBR is exactly an anchored task: the AI reads a data point from a validated system (MES, SCADA, LIMS) and recontextualizes it into the batch record format. On anchored tasks, the best models brought error below 1.5% ^[2]. And even so, what is critical is never decided alone: iLEAN prepares the dossier and the QA person signs. The three safety rings exist precisely for this.

^[1] Automotive quality standard ≈ 25 PPM. Source: Symestic.

^[2] OpenAI paper “Why Language Models Hallucinate”, 2025 — the reliability of AI in anchored tasks.