The Afterlife of a Drawing: Bridging Design and Production
A designer tightens the position tolerance on a machined face from 0.25 to 0.05. It gets checked, then approved and the revision releases on a Tuesday afternoon.
Go look at what the company knows on Wednesday morning. Pull the bill of materials and you’ll find the same file you had on Monday, byte for byte, because tolerance was never a BOM attribute. MRP runs that night against a demand picture nothing has disturbed. Over in purchasing, the open order for 400 pieces reads the way it read in March, when a buyer attached a drawing to an RFQ and emailed it to a job shop, which is where that drawing still sits and what those parts are still being cut to.
Nobody skipped a step. Every signature is where it belongs, the notification went out to the distribution list and the ECO is closed. Six weeks later a receiving inspector puts a part on a CMM and the company finds out.
This case is more interesting than the ones people usually write about, because it is not a story about a sloppy plant. It is what happens when configuration discipline works.
Why the ERP Cannot See it
Configuration management has a rule everyone learns in their first year. If form, fit and function survive the change, it is a revision, and if any of the three moves, it is a new part number. The rule is right, for a reason that has nothing to do with engineering: a part number is an inventory location before it is anything else, so two parts that cannot substitute for each other must never end up in a bin where a picker takes whichever is on top.
ERP starts from the same premise and arrives somewhere awkward. Most systems can only track, buy, consume and plan by part number, and where a revision field exists on the item master, it tends to be free text that neither the MRP run nor the purchase requisition has any reason to read. So the change that leaves form, fit and function intact, which is the exact change your own rule sends down the revision path, is the change your ERP has no mechanism to act on. Neither side is broken. The blind spot has a policy behind it, which is why it survives every process improvement anyone throws at it.
There is a wider version of this that we run into constantly and have never seen written down anywhere. The attributes most likely to change on a revision are the ones no bill of materials carries.
What moves cleanly between systems is structure: part numbers, quantities, the parent-child relationships that make up an assembly. Integrations were built for exactly that and they do it well. Tolerance isn't structure, and neither is surface finish, heat treat, weld symbol, material callout or a note at the bottom of the sheet reading “ALL MACHINED SURFACES FREE OF BURRS.”
All of it lives in the geometry and the note block, which means it lives in a PDF. The drawing may document the requirement, but it cannot tell you where else that requirement is used.
Nobody is Waiting on the Work
Terwiesch and Loch measured this in 1999 on the climate control system of a vehicle program, and their method was the useful part. They started the clock when someone noticed a problem and stopped it when the fix was physically in place, rather than when the paperwork closed.
The work in between, meaning every meeting and simulation and approval that constitutes the engineering content of a change, came to about 10 days. Elapsed time ran to weeks, routinely to months and, in the worst cases, past a year. Ninety percent of it was a document sitting somewhere with nobody touching it.
Twenty-seven years later Tacton put a version of the question to 280 manufacturing leaders across eight countries and 13 sectors: Do engineering changes propagate automatically into your sales and production systems? Twenty-one percent said yes, which leaves four out of five companies where somebody is retyping. Engineers own the 10 days and get blamed for the year, and that is more or less the entire political problem of the ECO process.
READ MORE: Team Collaboration Across CAD/CAM Workflows
The retyping deserves a closer look, because it is not laziness or a tooling gap. A change is authored as a document, whether that is a drawing, a redline, an ECO form or a PDF attached to an email, while every system downstream of engineering consumes records instead: material masters, BOM lines, requisitions, routings, inspection plans. Documents and records are different kinds of objects and nothing in the stack converts between them, so a person does.
That person is already at capacity, which is where the arithmetic turns ugly. No single change is difficult, but a hundred of them landing on a department running at 85% to 95% utilization is a queue, and a queue is indifferent to how motivated anyone is. Terwiesch and Loch modeled it and got lead times more than 10 times the actual work content. Add a review step to a queue that saturated and what you have mostly added is queue.
Effectivity
Effectivity fails so quietly that most people never trace the resulting scrap back to it. PLM handles date, serial number and lot effectivity. ERP mostly handles date effectivity. A change written to take effect at unit 1,240 therefore reaches the shop floor as a change taking effect on the 14th of next month. Whatever gets built between those two points becomes somebody’s rework.
Below that sits the remap from the engineering bill to the manufacturing bill, and the manufacturing bill carries phantoms, packaging levels, consumables and routing links that engineering has never seen and has no reason to see. At a lot of plants one planner holds that mapping together with a spreadsheet, and the mapping is not itself under revision control, so when the planner retires the company loses a translation layer it never knew it had.
The Money is Out the Door Before Approval
By the time a change clears approval the company has bought the old design several times over. There are open purchase orders against it, material on the receiving dock, material in a bin, material sitting halfway through a supplier’s queue two states away. Every well-formed ECO has a disposition field for precisely this situation, offering use as is, rework, scrap, or use and exhaust.
Whether that instruction reaches the buyer usually comes down to software the company already owns and never switched on. Dynamics 365 Supply Chain ships a change management module that will inspect open transactions and notify whoever owns them, and Costpoint carries a disposition code on the ECN. In mid-market deployments these tend to be unlicensed, unconfigured or set up once at go-live by an implementation partner nobody has spoken to since, so what lands in front of the buyer is a part number, a quantity and a promise date, with nothing to indicate that all three are now wrong.
Long-lead items lose this race for reasons unrelated to anyone’s diligence. A casting on a 14-week lead went to the supplier before the change existed, so if the ECO cycle runs eight weeks, the tool gets cut and the first articles poured while the change is still collecting signatures. No amount of rigor moves that, because it’s arithmetic.
Ask Your Buyer What the Supplier is Building to
It’s worth doing this week. When a purchase order goes out, what actually tells the supplier which revision to make? At most companies the honest answer is the last drawing anybody emailed them, probably attached to an RFQ, possibly last year. Portals, EDI and PDX packages solve the problem and the large OEMs run on them, but below that tier the transport is email, and email has no revision control.
Then read the two documents that constitute the agreement. A purchase order line carries part number, quantity, price and delivery date and the acknowledgment confirms those four. Neither has a field for drawing revision, so both companies end up holding a signed record that is silent on the only thing they are going to fight about.
READ MORE: How Engineering Leaders Can Avoid AI Pilot Purgatory
Most of what comes out of that gets filed as supplier quality, and most of it isn’t. A part shows up conforming to a print, correctly made, inspected and passed at the supplier, and wrong. Three quotes come back that cannot be compared because two priced Rev C and one priced Rev E.
Tooling gets cut to superseded geometry, which is the same failure with a comma moved into it. And a dimensional change on a ballooned print ought to trigger a new first article or a PPAP resubmission, which is quality’s call to make, except that quality is frequently not on the ECO distribution list.
The Three Usual Fixes
Companies land on one of three responses, and each buys something real at a price that never shows up on the invoice.
Buying the connector and syncing PLM to ERP nightly does what integration is built to do. Part numbers and quantities arrive reliably; tolerance and finish and note blocks do not, because they are not fields in either schema. The project transports data that was already structured and leaves the rest precisely where it was. This is the fix companies feel best about and it covers the narrowest part of the problem.
Tightening the process runs into the queueing math from the other side. At 85% to 95% utilization each additional review step is another position in line and the waiting rises nonlinearly, so it is entirely possible to improve the quality of your decisions and lengthen your lead time in the same quarter.
That’s the first-order effect. The second-order effect is that engineers start routing around the process, at which point you have neither better decisions nor faster ones.
Batching is the one we would defend. Where retooling setup costs dominate, holding changes and releasing them together is correct, and Terwiesch and Loch found it was usually a deliberate economic decision rather than a failure of will. What it does is convert a lead time problem into an obsolescence problem, and the material scrapped at the end of a use-and-exhaust window is a real number that nobody charges back to the batching decision that produced it.
What Can Actually be Automated
None of these gaps is an integration failure. They are the same data structure problem three times over, in that the change exists as a document while everything downstream needs records, and the operations that turn one into the other are extraction, comparison and routing.
A released drawing can be read, with dimensions, tolerances, notes and material callouts pulled into a structured record, and that record compared against the previous revision to produce a difference at the level of the feature rather than the file. The useful output is not a list of documents that changed. It is a line saying the position tolerance on feature four went from 0.25 to 0.05 and the finish callout went from 63 to 32, which can then be tested against open purchase orders, on-hand quantity, the process capability of the routing, and the revision on the supplier's last acknowledged print.
Reading and comparing require nothing at all from the ERP and only writing back does, which matters more than it sounds because it means the useful half can start long before the integration project everybody has been putting off.
None of the judgment moves, and none of it should. Whether a change is neutral to form, fit and function is an engineering call with liability attached to it. Use as is, rework or scrap is a decision about money and risk belonging to a person whose name goes on it, and PPAP resubmission belongs to quality. The value is in putting those decisions in front of the right people on the day of release rather than in week six, when a receiving inspector measures a part and finds out on everyone’s behalf.
The honest limit is ambiguity. A note reading “MATERIAL PER SPEC X UNLESS OTHERWISE APPROVED” cannot be turned into a clean record, because it is not clean information and any system that hands back a confident, structured answer from it is asserting something it doesn’t know. Flagging it and stopping is the correct behavior, and it is the first thing to test before trusting any tool with revision data, because plenty of them will fail it in front of you.
Start by Measuring the Lag
Before buying anything, measure it, since almost nobody has. One number will do, running from ECO release to the date the last affected purchase order or work order reflects the change. You can pull 20 changes from last quarter and look at the distribution rather than the average, because the mean hides the tail and the tail is where the money goes.
Then sort the elapsed time on each of the 20 into three buckets. The buckets are time sitting in PLM before release, time between release and the ERP update, and time between the ERP update and the supplier building the right thing. The result is usually lopsided, and it usually points somewhere the team didn’t expect. That’s the whole diagnostic and it costs an afternoon.
The Upstream Problem is Still Unsolved
Everything above accepts the drawing as the source of truth and tries to move it faster, which is a concession worth being honest about.
A drawing is a projection of a model, and a model can carry tolerance and annotation directly, as semantic PMI. A change to a toleranced feature in a model-based definition is machine-readable where it’s made, with no extraction, no PDF in the middle and no essay like this one.
That capability has been available for well over a decade, and mid-market manufacturers still release flat drawings, not out of conservatism but because the job shop at the other end can’t consume anything else. Your supply base sets your data format, not your PLM vendor and not the roadmap on the wall.
Until a mid-market manufacturer can send a model to a supplier and get a part back, everyone downstream will keep rebuilding, imperfectly, information that was fully structured before somebody hit print.
Have a question, comment or idea? Get in touch with the Machine Design editors at [email protected].
Naman Mukerji and Tanmay Neema are co-founders of Polymr, which builds operational software for manufacturers.
Sources
- Terwiesch, C. and Loch, C.H. (1999). Managing the Process of Engineering Change Orders: The Case of the Climate Control System in Automobile Development. Journal of Product Innovation Management 16(2), 160-172.
- Tacton, 2026 State of Manufacturing. Survey of 280 manufacturing leaders across 8 countries and 13 industry sectors.
About the Author
Naman Mukerji
Co-Founder, Polymr
Tanmay Neema
Co-Founder, Polymr
Related
dreamstime_l_182029592Voice Your Opinion!
To join the conversation, and become an exclusive member of Machine Design, create an account today!
Leaders LogoLeaders relevant to this article:



