AI for Parts & BOM Management
A released drawing and the PDM bill of materials can quietly disagree on part numbers, quantities, or revision. Here is what an AI agent can catch on its own, and what still needs an engineer.
·
⏱
8 min read
Michelle Ben-David
Michelle Ben-David is a mechanical engineer and Technion graduate. She served in an IDF elite technology and intelligence unit, where she developed multidisciplinary systems integrating mechanics, electronics, and advanced algorithms. Her engineering background spans robotics, medical devices, and automotive systems.
BOTTOM LINE
A drawing and its PDM bill of materials disagree more often than most release processes assume, usually on quantity, part number, missing lines, or revision. Quantity and part-number mismatches, and missing or extra lines, are objective facts an agent can flag immediately with a citation back to the exact drawing and BOM location. Revision mismatches are different: they often depend on effectivity dates and change-order status that only an engineer can weigh. The fix is not replacing the reviewer. It is making sure the objective mismatches never depend on someone happening to notice them, so the reviewer's time goes to the calls that actually need judgment.
A drawing gets released. Somewhere between the model, the balloon callouts, and the bill of materials sitting in PDM, one line stops matching another. Maybe a quantity was updated on the model but not the parts list. Maybe a part number was retyped by hand and a digit landed wrong. The drawing looks finished. It prints clean. And it is wrong in a way that nobody catches until a buyer orders the wrong quantity or an operator on the floor finds a bracket that does not fit.
This is not a rare failure mode. It is a routine one, because a drawing's parts list and a PDM bill of materials are maintained by different processes on different schedules, and nothing forces them to stay in sync automatically. The question worth asking is not whether this happens. It is which of these mismatches a piece of software can catch reliably, and which ones still need a person to decide what they mean.
The four places a drawing and a BOM can disagree
Most mismatches between a released drawing and its bill of materials fall into a short list. Knowing the list matters, because each type needs a different kind of check and carries a different kind of risk.
Part number mismatch. The balloon on the drawing points to a callout number that does not match the part number carried in the BOM for that same position, usually from a retype, a copy-paste from an old revision, or a part that was superseded after the drawing was last touched.
Quantity mismatch. The model has four of an item at a given position, the drawing's parts list says four, but the BOM in PDM says three, or the reverse. Quantity errors are the most common type and the most expensive, since they drive purchasing and kitting directly.
Missing or extra line item. A fastener, a seal, or a sub-assembly appears on the model and the drawing but never made it into the BOM, or a line survives in the BOM after the part was designed out of the assembly.
Revision mismatch. The drawing references a part at a revision that is no longer the one carried as current in PDM, so the drawing is technically accurate for a version of the part that nobody should be building anymore.
Any one of these can sit undetected for months if nothing in the review process is built to compare the drawing against the live BOM rather than against itself.
IN PRACTICE
With Leo, our team improves design quality, reduces mistakes, and shortens time-to-market.
- Uriel B., Field Warfare and Survivability Specialist
What checking a drawing against the BOM actually requires
A useful check needs three things in front of it at the same time, not three things read separately and compared from memory.
The first is the drawing's own parts list: the balloon numbers, the quantities called out against each balloon, and the part numbers or descriptions printed in the table. The second is the model's actual structure: how many instances of each component the assembly really contains, independent of what any table claims. The third is the bill of materials as it currently stands in PDM or PLM, not a BOM exported last week and left on someone's desktop. A stale export defeats the entire point of the check, since it compares the drawing against a snapshot that may already be out of date itself.
Reading all three together is what makes the comparison trustworthy. A tool that only reads the drawing's own table can confirm the table is internally consistent, but it cannot tell you whether that table still agrees with reality. The check has to reach into the model and into the live PDM record, not just the sheet in front of the reviewer.
This is also where the check runs into a real constraint: it can only be as current as its access to those three sources. An agent that reads a cached BOM or an older assembly state will produce false confidence, which is worse than producing no answer at all.
Which mismatches an agent can flag without a human
Some of the four mismatch types above are objective counting problems. A quantity of four on the model against a quantity of three on the BOM is not open to interpretation, and neither is a balloon that points to a part number string that does not appear anywhere in the current BOM. These are the mismatches an agent can flag on its own, with a citation back to the exact drawing sheet, balloon number, and BOM line involved, so a reviewer can verify the flag in seconds rather than re-deriving it.
The same is true for a missing line: if the model contains an instance of a part that has no corresponding row anywhere in the BOM, that is a fact, not a judgment call. An agent reading the assembly structure and the BOM side by side can surface it immediately, before the drawing goes anywhere near a release stamp.
What makes this reliable rather than noisy is scope. The agent is not deciding whether the drawing is correct. It is reporting where two records that are both supposed to describe the same assembly stop agreeing with each other, and pointing at exactly where. That is a narrow, checkable claim, which is why it can be automated with confidence.
Which mismatches still need an engineer's judgment
Revision mismatches are a different animal. A drawing pointing to a part revision that is not the current one in PDM is sometimes an error, and sometimes exactly correct, because an effectivity date has not arrived yet, or an engineering change order covering that part is still working through approval and has not been released. In either case, the drawing may be intentionally referencing the revision that is valid as of today, even though a newer revision already exists in the system.
An agent can and should surface every case where a referenced revision is not the current one. What it should not do is decide on its own that this means the drawing is wrong, because that decision depends on context the BOM and the model do not carry: an ECO's approval state, a customer's specific effectivity requirement, or a planned cutover date that has not been reached yet. That context lives with the people running the change, not in the geometry.
The same caution applies to a part number mismatch that turns out to be a legitimate substitution rather than a typo. An agent should flag the disagreement either way. Whether it is a mistake or an approved substitute that has not been formally re-balloted onto the drawing is a call for an engineer who knows what changed and why.
What a mismatch caught a day late actually costs
The cost of catching a drawing-versus-BOM mismatch depends entirely on when it is caught. Before release, it is a five-minute correction. After release, it is a purchase order for the wrong quantity, a kit built with the wrong part sitting in inventory, or a line stopped while someone traces the discrepancy back to its source. None of those outcomes are hypothetical. They are the ordinary failure mode of drawings and BOMs maintained on separate schedules by separate tools.
This is the value driver behind connecting a drawing check to the live PDM and model data rather than treating it as a manual proofreading pass. Leo reads across the drawing, the CAD assembly, and the current PDM bill of materials at once, so a mismatch between them shows up as a specific, cited discrepancy rather than something a reviewer has to notice by eye on a busy release day. The judgment calls, effectivity, approvals, intentional substitutions, still belong to the engineer. What changes is that the objective mismatches no longer depend on someone happening to notice them before the drawing ships.
Related reading on catching drawing errors before release: AI engineering drawing review covers the broader five-layer check a drawing needs, and AI-powered BOM validation looks at catching errors inside the BOM itself. For the three-system version of this problem, see BOM discrepancies across CAD, PDM, and ERP. If the EBOM-to-MBOM handoff is the more common source of drift on your line, EBOM vs MBOM explains where that split usually breaks. And for the drawing side of consistency more generally, engineering drawing standards covers what a compliant drawing needs beyond the parts list.
FAQ
See where your drawings disagree
Check drawings against your live PDM BOM automatically.
Leo reads your drawings, CAD assemblies, and PDM bill of materials together, and flags exactly where they disagree before a release goes to the shop.
Schedule a Demo →
#1 New AI Software Globally - G2 2026
Enterprise-grade security
Trusted by world-class engineering teams
