AI & Business

How to prepare a manufacturing record book in fewer hours

For suppliers to oil and gas and process plants, the manufacturing record book is often what the last payment waits on. Where the hours go, how to measure them, and when software is not the answer.

16 September 2026 · 8 min read

The valves have been dispatched. The inspector has released them, the truck has left, and the invoice has gone out. Then, weeks later, the order is still not closed, because the manufacturing record book is not.

If you supply valves, castings, forgings, fittings or pressure equipment to refineries and process plants, you know this situation. The book is the final bundle of certificates, test reports and approved drawings that proves what you built matches what was ordered. Preparing it is rarely anyone's full-time job, so it gets done in between everything else, and the order stays open until it is accepted.

This is not only an administrative irritation. In a 2014 Engineers India Limited enquiry for gate, globe and check valves for Bharat Petroleum, the payment terms for orders that call for vendor documentation release 90% after the material is received and accepted at site, including receipt of the relevant technical documentation, and the remaining 10% only after receipt of the final technical documentation (EIL enquiry MR 3314, clause 15.1.2, page 94). That enquiry is old and your own purchase orders may say something different, so read them. But where terms like these apply, the book is where your last payment waits.

What goes into a manufacturing record book?

The contents are set by the purchase order and its documentation specification, not by habit, so the list below is typical rather than complete. For a valve or casting order it usually includes:

  • an index in the numbering and order the client asked for
  • the approved quality plan or inspection and test plan
  • approved drawings and datasheets, at their final revision
  • material test certificates, each traceable by heat number to the parts it covers
  • test reports: pressure or hydro tests, non-destructive testing, dimensional checks, material identification where required
  • third-party or client inspection reports and release notes
  • records of any non-conformance and how it was closed

Much of it does not start in your office. Raw material certificates come from mills and traders, casting and forging certificates from sub-suppliers, test reports from your shop floor and outside labs, and inspection reports from the agency. The book is where all of it has to agree.

Where do the hours in a manufacturing record book actually go?

Ask most people and they will say "compiling", as if the work were scanning and merging PDFs. That part is quick. The hours sit elsewhere.

Chasing. A sub-supplier's certificate is missing, or it arrived as a phone photograph, or it carries a different heat number from the one stamped on the part. Every gap means a call, an email and a wait.

Matching. Each certificate has to be tied to the specific items it covers: this heat number went into these body castings, which became these valve serial numbers. If that link was not recorded when the material was issued, someone reconstructs it later from registers and memory.

Checking. Reported chemistry and mechanical values have to be compared against the limits in the specification. A person reads the certificate, finds the limits in the purchase order, and compares them line by line, again for every certificate.

Reformatting. Every client wants its own index, numbering and cover sheets. The same documents get renamed and reordered for each one.

Revising. The reviewer returns comments. One certificate is replaced, and the index, page numbers and cross-references around it have to be redone.

Notice that almost none of this is typing. It is reading, matching and checking, done by the one or two people who understand the order well enough to do it.

How do you measure the hours on your own orders?

Before you change anything, find out what the book costs you. You do not need a consultant for this; your own records hold most of it.

Take your last five orders that needed a record book and note, for each:

  1. Documents in the book — count the certificates and reports in the final index.
  2. Review rounds — how many times the book, or parts of it, came back with comments.
  3. Days from dispatch to documentation acceptance — the dispatch date against the date the client accepted the final documents.
  4. Days from acceptance to the final payment, and the amount that was waiting.
  5. Who did the work — the names, and whether they were pulled away from other jobs.

Then time one book as it is prepared. Not an estimate afterwards; a simple log of start and stop times against each activity above.

The arithmetic is simple enough to check yourself. For example, if a book holds 80 certificates and reports, and finding, matching, checking and filing each one takes 5 minutes, that is 400 minutes, close to 7 hours, before a single review comment. If the book comes back twice and each round takes half as long as the first pass, that adds another 400 minutes, and the total passes 13 hours. Multiply by your orders per month and you have a number worth discussing.

Do the same for money. If 10% of a Rs 40 lakh order waits on documentation, Rs 4 lakh is held until the book is accepted. The days it waits are in your own records.

Your numbers may be much smaller than this example, or much larger. That is the point of measuring them.

Do you need software, or a better way of filing?

Often the first fix is not software at all. Several of the worst hours come from decisions made, or not made, at the start of the order.

  • Agree the index before fabrication starts. Get the client's documentation requirements, numbering and format confirmed when the order is placed, not when the goods are ready.
  • File as each inspection stage closes. A test report added to the book the day the test is signed off is minutes of work. The same report found three months later is not.
  • Keep a heat-number register from the day material is issued. Record which heat went into which parts at the time. A register kept at the time avoids rebuilding that link from memory later.
  • Make certificates a condition of receiving material. If a sub-supplier's material arrives without a legible certificate, it waits until one comes.
  • Keep a checklist per client. Repeat clients ask for the same things; write them down once.

If you handle a few record books a year, this discipline may be all you need. If you handle many, with standard client formats, dedicated document-control software may be enough: it tracks what has been submitted, what came back and at which revision.

When is AI the wrong tool for this?

AI is not always the answer, whatever a vendor says. Be sceptical if:

  • Volume is low. A few books a year rarely justify building or buying anything beyond good filing.
  • The delay is on the other side. If your books sit for weeks in the client's review queue, faster preparation will not bring the payment forward.
  • The problem is a missing document. No software can produce a certificate a sub-supplier never issued. That is a purchasing problem.
  • Your records are already clean and consistent. If certificates arrive digitally in a predictable format and the heat register is kept, most of the reading work has already been removed.

What does a working system look like?

Where volume is real and the hours are in reading and checking, a well-built system does the reading, not the judging. In practice it should:

  • Read each certificate and report, whatever the layout, including scans and photographs, and pull out the heat number, material grade, chemistry, mechanical values and test results as named fields.
  • Keep the source. Every extracted value points back to the page it came from, so anyone can check it in seconds. We wrote about why this matters in OCR is more than reading PDFs.
  • Check against the specification. Compare each value against the limits in the purchase order and show which pass, which fail and why.
  • Build the traceability map. Link heat numbers to parts and serial numbers from the records you already keep, and show gaps instead of filling them.
  • Flag doubts for a person. An unreadable value, a heat number that matches nothing, a certificate type the order does not accept: these go to your QC person, not into the book.
  • Assemble the index in the client's format, and redo it when one document changes.

The QC person still signs off. The difference is that they review a checked draft instead of building it from scratch. Our document intelligence implementation example shows the same approach applied to documents in general.

What should you ask before buying anything?

Whether the offer is software, a service or a custom build, ask:

  1. How many hours does our book take today, measured, and how many will this remove?
  2. What happens when the system is unsure? Does it guess, or does it flag?
  3. Can every value in the book be traced back to the original page?
  4. Does it work with the scans and photographs we actually receive, not only clean PDFs?
  5. Will it produce each client's index format, or only its own?
  6. What does our QC person still have to do, and is that the part that needs their judgement?
  7. If our volume is low, will you tell us this is not worth doing?

If the answer to the last question is anything but yes, keep looking.

Where to start

Measure one order. Count the documents, the review rounds and the days between dispatch and the final payment. That number will tell you whether you need better filing, a document tool, or something that reads and checks certificates for you.

If you would like a second pair of eyes on it, book a 30-minute call about this one process at prakreon.com/contact. We will ask who prepares your record books, how often and how long it takes, and we will tell you plainly if AI is not the answer.

Build with Prakreon.

Start with one process that takes up your team’s time. A free 30-minute call is enough to tell whether it is worth measuring.