It is 3:40 on a Friday. A customer's quality engineer wants the material certs, the certificate of conformance, and the country of origin for an order that shipped two weeks ago. Your quality manager walks to the filing cabinet.
The traveler is there, folded in thirds, with a coffee ring over the operation sign-offs. The heat number on the receiving log has a 3 that might be an 8. The cut sheet references a bundle tag that nobody photographed. The MTR is in someone's inbox, attached to an email with the subject line "certs."
By 5:15 the packet is done, and it is probably right, though nobody would bet the account on it.
Most manufacturers have a version of this Friday, whether the packet is mill certs for a fabricator, pressure test records for a flow line, or lot history for a circuit board build. The fix starts with the forms, well before any AI system.
Software can only help with jobs it can read
Before software can help with a job, it has to be able to read the job. Paper forms, and spreadsheets everyone fills in differently, are hard for anyone to read reliably, including your own people on a busy afternoon.
Digitizing your operation means turning the forms you already use (travelers, receiving logs, inspection sheets, cert requests) into digital forms with fixed fields. It is the least exciting step in any AI project, and it decides whether the rest works.
What structured data means
Take a photo of a filled-in receiving log. The photo holds the information, but only as a picture. To know that "A36" is the grade and "4471823" is the heat number, a computer has to work out where each one sits on the page and what each squiggle says. AI can read that photo and will usually get it right, which means it will sometimes be wrong about whether a digit is a 3 or an 8, and just as sure of itself either way.
Now picture the same receiving log as a form on a tablet, with boxes labeled Heat Number, Grade, and PO. Whatever goes in the Heat Number box is a heat number, so the computer never has to guess. That is structured data: information stored in labeled fields, the same way every time.
A photo of a page is unstructured. A form with fields is structured. AI can work with both, but it is far more accurate, and far easier to check, when the data is structured from the moment someone enters it.
One fabrication job on paper
Follow a single fabrication job through the shop the way many shops run it today. A circuit board assembler or a flow line maker runs the same handoffs under other names.
At receiving, a truck arrives with plate. The receiver checks it against the PO, writes the heat number on the receiving log, and drops the MTR in a tray. If the MTR did not come with the load, someone may or may not make a note to chase it.
At cut, the traveler goes out to the saw or the table and the operator writes which plate he pulled. Sometimes he writes the heat number, and sometimes he writes "same as last job."
At weld, the welder signs off the operation, and the procedure number ends up a smudge.
At inspection, the inspector fills in a paper sheet with dimensions and a pass or fail. That sheet goes in a binder, and the traveler goes in a different folder.
At shipping, someone pulls the order, builds the packing list, and asks quality for the cert packet. Quality now has to rebuild the story of the job from four pieces of paper, two spreadsheets, and one email thread.
Information leaks out at every handoff. Three people write the heat number down three times, and nothing guarantees all three match. The link between this plate, this part, and this shipment lives in people's heads, and when one of those people is on vacation, the link is gone.
The same job with digital forms
The shop floor, the people, and the steps stay the same. Only the way people fill in the forms changes.
At receiving, the receiver opens the receiving form on a tablet, picks the PO from a list, and enters the heat number in its own field. He scans the MTR and attaches it to that receipt. If the heat number on the form does not match the one on the MTR, the form says so before the truck leaves.
At cut, the digital traveler lists the material received for this job, and the operator picks the plate he pulled. The heat number comes along automatically because the receiver captured it once.
At weld, the welder signs off the operation on the traveler, and the sign-off carries a name, a time, and the procedure used.
At inspection, the inspector fills in a form tied to the same job, entering dimensions in fields instead of margins. The supervisor sees a failed check the moment the inspector enters it.
At shipping, someone builds the shipment from the job record, and every part on the packing list already carries the heat it came from and the inspection it passed.
Nobody on the floor is doing a new job. The same information lands in one place, in labeled fields, connected to the job.
What full traceability gives you
We did this with a client, a manufacturer whose travelers, receiving logs, inspection sheets, and cert requests were all paper or spreadsheets. We turned each of those into a structured digital form.
The company could then see every job: where it was, who had touched it, and what was waiting. It also had full traceability from the mill heat to the shipment. Pick any part on any packing list and you can walk it back through inspection, weld, cut, and receiving to the MTR it came from, and that traceability is what AI needs.
Some customers put that walk-back in the contract. On a federally funded highway job, the FHWA rule requires every process on the steel, down to the coating, to happen in the US. The only exception is the small foreign allowance shown at the top. An aerospace supplier working under an FAA production certificate has to keep those records for years. In both cases a heat number with a 3 that might be an 8 is a real problem.
How software assembles the compliance packet
The software pulls the shipment, follows each line back to its heat, and collects the matching MTRs. It fills in the certificate of conformance from the job and inspection records, and the certificate of origin from the receiving data. It checks the values against each other: does the grade on the MTR match the grade on the PO, does the heat on the MTR match the heat on the receiving form.
It also builds a source index that links every value in the packet back to the document it came from. If the CoC says a part passed inspection, the index points to the inspection form. If the heat number is 4471823, the index points to the page of the MTR where that number appears. Then the software stops and hands the packet to a person.
Your quality manager reviews the packet instead of rebuilding it. Anything the software flagged as a mismatch sits at the top, and every other value is one click from its source. She checks what needs checking and signs, having spent her time on judgment instead of the filing cabinet.
AI can only assemble what your team already recorded. The forms come first.
No AI tool can trace a heat number that nobody wrote down in a field. It can guess from a photo, but a packet built on guesses is worse than one built by hand, because it looks more trustworthy than it is.
Three common objections, answered
Expect three worries, and they are fair.
The first is "Our guys won't fill in tablets." Some won't at first, often because someone once handed them software that made their day harder. Make it easier this time. Use the same fields they fill in on paper, in the same order, and nothing more. Use pick lists instead of typing where you can, and put the tablet where the clipboard used to hang. If a digital form takes longer than the paper one, fix the form and ask the operators to help design it.
The second is "This is an ERP project in disguise." You are giving the forms you already use fixed fields, and you are leaving your ERP and the way the shop runs alone. The ERP keeps doing its job, and the forms fill the gaps it never covered, like the heat number on the cut step. You can connect the two later.
The worry is earned. In Panorama Consulting's last three ERP reports, about 1 in 3 projects ran over budget. The median project in the 2024 and 2025 reports cost $450,000. One digital receiving log is a much smaller bet.
Source: Panorama Consulting ERP Reports 2024 to 2026. Includes projects slightly and significantly over budget.
The third is "It'll take a year." Start small and it won't. Digitize one form on one line, watch people use it for a couple of weeks, fix what annoys them, and then do the next one. The receiving log is often the right first form, because that is where the heat number enters the building.
For scale, the median ERP project in Panorama's reports took 15.5 months in 2024 and 9 months in 2025 and 2026. The one-form plan above is measured in weeks.
Check your forms before you pick an AI tool
If you are thinking about AI for quoting, compliance, or scheduling, first ask whether your company records the information those tools need, in fields, connected to the job. Which AI tool to buy comes later.
If it does not, every AI project will spend its first months fighting your paper. If it does, the same projects get simpler and easier to check.
What to do this week
- Pick one typical job that shipped recently.
- Collect one copy of every form used on it, including the receiving log, traveler, inspection sheet, cert request, packing list, and any spreadsheet or email that carried information. Lay them out on a table in order.
- Circle every place someone wrote the heat number by hand, count them, and check whether they all match.
- Ask your quality manager how long the packet for that job took and where she had to go looking.
- Pick the one form that would save the most chasing if it were digital tomorrow, and make it your first form.
If that form is the receiving log, a first version needs only a few fields: the PO (picked from a list), heat number, grade, quantity, the scanned MTR, and who received it and when. Copy the paper form's order, and add nothing else until people are using it.
None of this costs anything, and it will tell you more than any vendor demo.
Questions people ask
What is structured data in manufacturing?
Structured data is information stored in labeled fields, the same way every time, like a heat number in a box marked Heat Number. A photo of a handwritten receiving log is unstructured, so a computer has to guess what each mark means. AI is far more accurate and easier to check when data is structured from the moment someone enters it.
Do I need to digitize paper forms before using AI?
For work like cert packets, quoting, and scheduling, yes. No AI tool can trace a heat number that nobody wrote down in a field, and a packet built on guesses from photos looks more trustworthy than it is. Start with the forms you already use, such as the receiving log, traveler, and inspection sheet.
Does digitizing shop floor forms mean replacing the ERP?
No. Digitizing gives the forms you already use fixed fields, and the ERP keeps doing its job. The digital forms fill gaps the ERP never covered, like the heat number on the cut step, and the two can be connected later.
Terms in this piece
- Structured data
- information stored in labeled fields, the same way every time, like a heat number in a box marked Heat Number.
- Unstructured data
- information without fixed fields, like a photo of a handwritten page or the body of an email.
- Digital form
- a form on a tablet or computer with fixed fields.
- Traceability
- the ability to follow any part back through every step of the job to the material and documents it came from.
- Source index
- a list that links every value in a document to the exact source it was taken from, so a reviewer can check it in one click.
- ERP
- the system that runs orders, inventory, and accounting. Digital forms work alongside it.
