Paper on glass is still paper

Every few weeks I meet a technology leader from another Indian API company who has just finished converting batch records to electronic form. The project usually has the same shape. The paper batch manufacturing record, page by page, is reproduced on a tablet. Operators tap where they used to sign, and fields that took handwriting now take typing. Quality reviewers read the record on a screen instead of at a desk stacked with folders. The plant announces that it is paperless, and customer auditors, for a while, are pleased.

I understand the attraction, because I proposed something close to it myself at Dahej before we changed our minds. It is quick, it needs little redesign, and it can be rolled out across a whole plant within one budget year. My argument is that it buys the least useful part of an electronic record and leaves the most useful part on the table. Paper on glass is still paper. It records what people say they did. It does not stop them doing the wrong thing, and it does not spare the reviewer from checking everything.

What the paper record was really doing

A batch manufacturing record in an API plant does three jobs at once. It instructs: charge this material into that reactor and hold it at this temperature for so long. It records who did each step, when, with which lot of which material and on which equipment. And it is the evidence quality reviews before deciding whether the batch can be released. On paper the same sheet does all three jobs, and people do the checking between them. The operator reads the instruction, performs the step and writes the result. A second person verifies critical entries. The reviewer later confirms that entries are complete, that calculations are right, that the equipment was clean and calibrated, and that every deviation has been raised.

Most of the effort in that arrangement goes into catching errors after they happen. When we studied a month of Dahej's paper records in 2022, the commonest documentation errors were missed signatures, arithmetic mistakes in yield and quantity calculations, entries made out of sequence and references to equipment whose cleaning status had not been recorded. Reviewers spent most of their 6.5 days per record finding and returning these errors, not judging the chemistry.

Now consider what paper on glass does to that list. A missed signature becomes a missed tap, which the screen may flag, though often only at the end. The arithmetic is still done by the operator unless someone builds the calculation in. Steps can still be recorded out of order if the form allows free navigation, which most converted forms do, because the paper allowed it. And the equipment's cleaning status still lives somewhere else. The reviewer still has to check everything, now on a smaller screen.

Enforce the sequence, review the exceptions

The alternative is to let the system do the checking that people used to do afterwards. At Dahej the execution system now refuses to open a step until the previous one is complete. It will not let an operator charge into a vessel that is not recorded as cleaned and released. It reads the weight from the balance instead of accepting a typed number, and it compares the material lot with the bill of materials before the drum is tipped. It calculates yields itself. When a value falls outside its limit, the system demands a comment and raises an exception that quality must see.

The reviewer's task then changes. Instead of re-reading 140 pages, the reviewer examines the exceptions the system raised, the comments attached to them and any deviations, and confirms that the configured checks were active for that batch. That is what review by exception means, and it is defensible only when the checks have been validated and nobody can configure the system to skip them without leaving a trace. For the products we built this way, review time fell from 6.5 days to 1.8, and documentation errors from 31 to 4 per 100 batches. Those numbers came from enforcement. The screens themselves made almost no difference.

Enforcement has costs that paper on glass avoids. Every product's recipe has to be authored in the system, with each step, limit and calculation defined and verified. At Dahej that took fourteen weeks per product, more than twice my estimate, because writing a recipe exposed ambiguities the paper had tolerated for years: a hold time stated as approximately two hours, a wash repeated until the rinse looked clear. Each ambiguity needed a decision from process development and quality before the system could enforce anything. I now regard that as the most valuable part of the project, though it did not feel so at the time.

There is also a cost on the floor. A system that refuses to proceed is a system shift chemists will argue with, sometimes rightly. In the first three months we raised more exceptions than paper had ever produced, some because our limits were tighter than the process needed. Operators need a quick, recorded way to escalate, and engineers need to review those escalations every week, or the plant will start looking for ways around the system.

Fewer products, deeper records

Because enforcement is expensive per product, I believe a plant should digitise fewer products deeply before it digitises all of them thinly. We began with 14 of Dahej's 38 products, chosen because they accounted for 61 per cent of its batches. The plant ran two methods for two years, which supervisors disliked and customer auditors had to examine twice. Our head of manufacturing had warned that it would be awkward, and it was.

The alternative would have been a plant fully on tablets within eighteen months and a validation package covering 38 converted forms. We would have validated, at considerable expense, a system that did little more than store signatures. When customers later asked what the electronic record prevented, we would have had little to show them.

For directors and executives judging such a programme, I would suggest three questions. Which checks that reviewers used to perform does the system now perform, and how was that validated? What share of batches now goes through review by exception, and how long does that review take? And how many product recipes have been authored, against how many were planned? A programme that can answer only with a percentage of the plant converted is measuring coverage rather than control.

By March 2026 Dahej had 26 products on enforced electronic records, covering 78 per cent of its batches, and we are taking the same approach to Kurkumbh, where smaller campaigns and a high-potency suite make every recipe harder to write. It will take until late 2027. I would rather explain that timetable to a board than explain, after an inspection, why a paperless plant still relied on people to catch their own mistakes.