BlogBatch Record Review by Exception: Faster Release Without Cutting Corners

The batch is made. It has been tested and it meets its specification. And yet it sits in quarantine, waiting, because a reviewer still has to work through a 200 page record before it can be released. For many manufacturers, this final review is the slowest step between the production floor and the patient.

It is also one of the most closely watched. Production record review and the investigation of discrepancies, both covered by 21 CFR 211.192, were cited in 164 FDA inspection observations in fiscal year 2025, making it the second most cited section of the drug GMP rules. Slow review and weak review are two sides of the same problem. Review by exception, made practical by AI, addresses both.

What GMP requires

Under 21 CFR 211.186, every product needs a master production record, the approved recipe for how a batch must be made. Under 21 CFR 211.188, every batch needs a batch production and control record showing what actually happened. Under 21 CFR 211.192, the quality unit must review and approve these records before release and thoroughly investigate any unexplained discrepancy.

In the EU, Chapter 4 of the GMP guide sets out documentation requirements, and Annex 16 describes how a Qualified Person certifies each batch. The principle is the same on both sides of the Atlantic: no batch is released until its record has been reviewed and found to be correct.

Why review takes so long

Batch records are long by nature. Each step carries its own entries, measurements, signatures and second person checks. Paper records add handwriting, crossed out corrections and late entries, all of which demand close reading. Because review usually happens at the very end of the process, any error that is found takes the longest to resolve and holds the entire batch back.

The frustrating truth is that most of this effort confirms what was already right. Reviewers spend hours reading correct entries in order to find the few that are not. Fatigue sets in, and the very errors the review exists to catch become easier to miss.

The idea behind review by exception

Review by exception turns this around. Instead of reading every line, a system checks each entry against predefined, approved rules: limits, sequences, required signatures and calculations. Entries that pass are accepted. Entries that fail, or that cannot be checked, are listed as exceptions for a person to review. Reviewer time goes where the risk is.

The benefit can be dramatic. EY has described how a well designed electronic batch record system can reduce a 150 page review to a short exception report. The catch is that review by exception has traditionally required a full manufacturing execution system or electronic batch record platform. Many plants, especially small and mid size sites and contract manufacturers, still run on paper or hybrid records, and replacing those systems can take years and a significant investment.

How AI brings review by exception to paper and hybrid records

AI closes this gap by working with the records plants already have. Scanned pages or exported records are read in full: typed text, handwriting, checkboxes, signatures and dates, each with a confidence score. Every entry is then checked against the master record, including the steps, their order, the approved limits, the yields and the calculations.

Anything that does not match is listed as an exception, with the page, the entry and the reason. Typical exceptions include missing or late signatures, values outside the approved range, steps performed out of order, calculation errors, unexplained overwrites and material lot numbers that do not match linked records. Any entry the AI cannot read with confidence goes to a person rather than being guessed.

Packaging records deserve special attention. GMP requires labeling to be issued, used and reconciled under strict control, and the batch packaging record shows exactly which labels were used. When the printed labels have already been checked against the approved artwork on arrival, using AI that can read scanned or photographed labels, the review of the packaging record starts from much firmer ground. Label errors are stopped before they reach the line, rather than discovered in the record afterwards.

The reviewer works through the exceptions, resolves each one and signs. The full review, including every exception and decision, becomes part of the batch record.

Building trust in the exceptions

Review by exception only works if reviewers can trust what the system passes, not just what it flags. That trust has to be earned before the tool is relied on. The strongest programmes test the AI on historical records with known errors, to prove it finds what it should. In the first weeks of live use, they also sample some of the pages the AI passed, to confirm nothing is being missed.

The rules that define an exception should be agreed and approved by quality in advance. The model version should be locked and placed under change control, and every flag, decision and signature should be kept in a complete audit trail.

It is worth being precise about what regulators expect. GMP requires the quality unit to review production records before release. It does not require every line to be read by hand, provided the review method is validated, documented and every exception is properly reviewed. As with any change to a GMP process, the approach should be agreed with your QA and regulatory teams before it goes live.

Measuring the difference

The value shows up in a few clear measures: review hours per batch, time from the end of production to release, the share of batch records that are right first time, and the number of errors found after release. Recording these before you start makes the improvement easy to demonstrate and builds the case for extending the approach to more products and sites.

Faster release does not have to mean less control. With AI handling the detailed reading, reviewers can give every real exception the attention it deserves, and batches reach patients sooner.

Benchmark IT Solutions helps manufacturers apply AI to batch record review. Talk to us about reviewing your own batch records by exception.