The same data, three times, across three different systems
In the chemical and process industry, an incoming document rarely lands in one place. A safety data sheet (SDS or MSDS) from a new supplier has to go into the ERP, the quality system and the archive. An employee copies the hazard class, the hazard pictograms, the storage conditions. Then a colleague does the same thing in a different screen, because the system does not know it has already been entered. After that, someone signs off on paper confirming it was received correctly. This is not the result of bad intentions or poor policy: it is what happens when three systems do not talk to each other and when documents arrive in hundreds of different formats.
Why automating document flows in the chemical industry is so difficult
Variation is the problem. An SDS from BASF looks different from an SDS supplied by an Asian raw materials supplier, even though both contain largely the same legally required sections. A quality certificate (CoA) for a batch of polymer has a different layout than a CoA for a solvent, even when both come from the same supplier. Traditional template-based approaches, where you build a template per supplier that says 'batch number is always on line 7, column 2', break down here. As soon as the supplier updates their letterhead or switches to a new ERP, the template stops working. In sectors with fifty or a hundred suppliers, maintaining all those templates is a full-time job in itself.
Linking a CoA to a purchase invoice: where things keep getting stuck
A widely reported problem in the chemical industry is the link between the quality certificate and the corresponding purchase invoice. The CoA arrives by email, sometimes as an attachment to the delivery note, sometimes a day later as a separate PDF file. The invoice comes through a different channel. Someone has to manually match the batch or lot numbers, verify that the certificate belongs to the correct order and then complete the file for the quality department. If this is not done immediately, loose files start to pile up. During an audit or a customer claim, the question becomes: where is the certificate for batch 2024-1187? The answer is too often: somewhere in a mailbox.
What document-AI with human-in-the-loop actually does better
Document-AI reads an incoming document without a template. The system recognizes what type of document it is, extracts the relevant fields (batch number, test values, hazard class, supplier code) and proposes a match with the corresponding purchase order or invoice. The employee sees one screen: the extracted values, the suggested link and any deviations flagged. They approve, correct where needed, and the document enters the system in a structured format. The reading and typing work is gone; the decision stays with the person. That matters in the chemical industry precisely because the data in quality and safety documents touches on liability and compliance. You do not want a system that quietly buries errors.
When should you not automate document flows in the chemical industry?
An honest answer: for suppliers who already deliver via EDI and whose messages arrive correctly in the right format, there is nothing to automate. EDI works, and document-AI adds nothing to it. The same applies if you process only a handful of documents per month from one established supplier with a stable layout. The threshold at which automation pays for itself is roughly when repetitive manual work amounts to at least a few hours per week on a structural basis. Below that threshold, it is smarter to get the process properly organized before automating it. Document-AI also does not help when the source data is structurally poor: illegible scans, incompletely filled-in certificates or documents in languages the system does not support require a different approach at the supplier level first.