A pallet of carbon fiber prepreg comes off the truck at 7:40 on a Tuesday morning. The receiving clerk scans the packing slip, confirms the roll count, and posts the goods receipt in SAP. Eleven minutes of work, start to finish. The material moves to the cold store and the clerk goes back to the next truck.
Two documents came in on that shipment. One was a certificate of analysis from the supplier, listing resin content, volatile content, gel time, tack, and the batch number tied to those results. The other was the supplier invoice, referencing the purchase order, the line item, the quantity shipped, and a freight charge that was not on the original PO.
Those two documents are about the same pallet, the same supplier, the same purchase order line, and the same batch. And in most AS9100 plants, they will never be in the same system, the same queue, or the same conversation again.
The COA goes to quality. Somebody keys the characteristic results into an inspection lot, or files the PDF in a network folder and marks the lot as passed based on a visual scan of the numbers. The invoice goes to accounts payable, where it sits in a shared mailbox until somebody works the exception queue and figures out why it will not match. Two departments, two backlogs, two clocks, two sets of people asking the same supplier the same questions about the same shipment.
That split is not a quality problem or a finance problem. It is a document architecture problem, and aerospace manufacturers pay for it twice.
The two backlogs nobody connects
Walk into a mid-sized aerospace or composites manufacturer and ask about document automation. You will usually hear about one of two projects.
The finance side has an accounts payable initiative. The goal is to stop keying invoices, cut the exception queue, and get invoices posted before the discount window closes. Success is measured in touchless post rate and days to post.
The quality side has a different initiative. The goal is to stop retyping certificates of analysis into inspection lots, get usage decisions made faster, and make sure the record holds up when a customer or a registrar comes looking. Success is measured in lot cycle time and audit findings.
Both projects are real. Both are worth doing. Artificio has written at length about each of them, covering the invoice matching cascade, MRBR block resolution, goods receipt automation on the AP side, and COA extraction, characteristic result posting, and usage decision workflows on the QM side. This piece is not going to repeat any of that mechanical detail.
What this piece is about is the thing that gets missed when those two projects run in separate rooms. They are automating two halves of the same document event.
Both streams start at the same supplier. Both reference the same purchase order line. Both are triggered by the same physical shipment. Both wait on the same goods receipt to become actionable. And both, when they fail, fail for overlapping reasons. The quantity on the packing slip does not match the quantity on the invoice. The batch on the COA does not match the batch recorded at receiving. The supplier shipped against a PO line that was already closed. The material came in under a revision that quality has not qualified yet.Â
Run those as two projects and you build two ingestion pipelines, two extraction models, two exception queues, two supplier master mappings, and two audit trails covering the same event. Then you spend the next two years explaining to auditors why the AP record and the QM record disagree about what arrived.
Why aerospace makes this harder than most industries
In a general manufacturing environment, an invoice mismatch is an annoyance. Somebody calls the supplier, gets a credit note, and the line clears.
AS9100 changes the stakes because the quality record is not a parallel nice-to-have. It is the thing that makes the part sellable.
Clause 8.4 puts the burden of control over externally provided processes, products, and services squarely on the manufacturer. Clause 8.5.2 requires identification and traceability that survives the entire lifecycle of the part. Counterfeit part prevention requirements push documentary evidence of origin down through every tier. And the flow-down clauses in most primes' terms mean the manufacturer is contractually obligated to hold, retrieve, and produce supplier documentation on demand, sometimes for decades.
Composites make this sharper still. Prepreg has an out-life clock and a freezer-life clock that start ticking from dates printed on the supplier's paperwork. Adhesive films, honeycomb core, and resin systems all carry shelf-life constraints where the governing date lives on a supplier document rather than in a system field. If the COA sits in an inbox for four days before somebody transcribes the manufacture date, four days of usable out-life are gone, and no one can prove exactly when the clock started.
Now consider the same shipment on the finance side. The invoice cannot pass three-way matching until the goods receipt is posted. If the material is sitting in quality inspection stock, the receipt exists but the stock is restricted. If a usage decision has not been made, the receiving inspection is technically incomplete. Some AP teams block payment on that basis, which is defensible. Others release payment and then discover the batch was rejected two weeks later, after the cash has left.
The point is not that one department is right. The point is that the decision requires information from both sides, and in most plants neither side can see the other's queue.
The purchase order line is the join key
Here is the thing that makes unification practical rather than theoretical. The two document streams already share a spine.
Every supplier invoice line in an aerospace supply chain references a purchase order and a line item. Every incoming inspection lot is generated against a material, a batch, a vendor, and, in a properly configured SAP environment, the goods receipt document that created it. That goods receipt carries the purchase order and line item. The batch carries the vendor batch number that appears on the COA.
So the join already exists in the data model. Purchase order line, goods receipt document, material, batch, vendor. Five fields that both streams touch.
What has been missing is a document layer that reads both kinds of paper against that same spine, resolves them to the same SAP objects, and raises a single exception when either half does not line up.
That is what an end-to-end aerospace document workflow actually means. Not one giant system that does everything. One ingestion and understanding layer that recognises what arrived, extracts what matters, resolves it against the same set of SAP keys, and then hands each result to the module that owns it. Invoices to FI and MM. Characteristic results and usage decisions to QM. Both writing back to the same purchase order line, both leaving evidence in the same audit record.
What the AP half looks like when it works
On the accounts payable side, the target state is well understood. The invoice arrives by email, EDI, portal, or paper. It gets read without a template. Header and line values are pulled out, including the ones suppliers place inconsistently, like freight, tooling amortisation, certification charges, and minimum order surcharges.
Then the matching cascade runs. PO line, goods receipt, price, quantity, tolerance. Where everything agrees, the invoice posts without a human touching it. Where it does not, the system identifies which specific tolerance failed and routes it to the person who can actually decide, with the underlying documents attached rather than referenced.
Blocked invoices become a managed queue instead of a mystery. Price variances go to purchasing. Quantity variances go to receiving. Missing receipts trigger a receipt check rather than an email chain. The MRBR release process stops being a monthly fire drill.
Artificio's AP Studio work covers the mechanics of that cascade in detail, including how no-PO invoices get resolved, how partial receipts and over-deliveries are handled, and how GL coding gets predicted from historical posting patterns. There is no need to restate it here. What matters for this discussion is a single fact about that workflow.
The AP engine already knows, for every invoice line, exactly which goods receipt it matched against and which batch that receipt created.
That is a quality-relevant fact sitting inside a finance system, and almost nobody uses it.
What the QM half looks like when it works
On the quality side, the target state is equally well defined. A supplier certificate of analysis arrives, usually as a PDF, sometimes as a scan of a scan, occasionally as a photograph taken on a phone at a loading dock.
The document gets read and the values that matter get pulled. Vendor batch number. Manufacture date. Expiry or retest date. The measured characteristics, with their units, their specification limits, and their pass or fail state. Any nonconformance notes. Any deviation or concession references. The authorised signatory.
Those values get resolved against the inspection lot that the goods receipt created, posted as characteristic results, and evaluated against the inspection plan. Where every characteristic sits inside specification and the document itself is complete and authentic, the usage decision can be proposed automatically and the stock moved from restricted to unrestricted. Where something is out of tolerance, incomplete, or inconsistent, the lot holds and a quality engineer gets a queue item that already contains the comparison.
Artificio's SAP QM content covers usage decision automation, characteristic result posting, batch record digitisation, and the specific handling of aerospace COAs in depth. Again, no need to repeat it.
The fact that matters here is the mirror image of the AP fact.
The QM engine already knows, for every inspection lot, which purchase order line and which vendor the material came from, and whether the batch has been released or rejected.
That is a payment-relevant fact sitting inside a quality system, and almost nobody uses it either.
The four handshakes
Once both halves run against the same spine, four specific exchanges become possible. Each one removes a category of work that currently exists only because the two systems cannot see each other.
The receipt handshake. Both streams stall when a goods receipt is missing, late, or wrong. Today, AP chases it as a matching failure and QM chases it as a missing inspection lot, separately, often on the same day, sometimes to the same receiving supervisor. With one workflow, a single receipt exception is raised once, worked once, and clears both queues.
The quantity handshake. The invoice says forty rolls. The packing slip said forty. The COA covers two batches with an implied split. Receiving posted thirty-eight because two rolls were damaged. In a split world, AP discovers this as a quantity variance and QM discovers it as an inspection lot with the wrong lot size, and the two findings never get connected to the same root cause. In a unified world, the discrepancy surfaces once, against the receipt, with all three documents attached.
The hold handshake. This is the one with real money attached. If a batch is rejected on incoming inspection, that fact should be visible to AP before payment runs. If an invoice is on price block, quality does not need to know, but if quality has rejected the material, finance absolutely does. Linking the usage decision state to the invoice block state gives AP a defensible reason to hold, and a documented trigger to release once the material is accepted or the credit note arrives. Chasing a supplier for money already paid on material already scrapped is one of the more avoidable losses in this industry.
The supplier handshake. Supplier performance in aerospace is usually measured on quality and delivery. Invoice accuracy rarely makes it into the scorecard, because it lives in a different system owned by a different function. When both streams write to the same record, the supplier profile gets richer. A supplier whose COAs are consistently incomplete and whose invoices consistently carry unauthorised freight charges is telling you something about their internal controls. That is exactly the kind of signal clause 8.4 expects a manufacturer to act on.
None of these four requires a new master data project or a rebuilt SAP landscape. They require the document layer to hold both streams against the same keys.
What the auditor actually sees
Auditors do not grade automation. They grade evidence.
The question in an AS9100 surveillance audit is rarely whether a process is fast. It is whether the manufacturer can demonstrate control, show the record, and explain the decision. Pull a part number, trace back to the batch, trace back to the incoming inspection, trace back to the supplier documentation, and see whether the story holds together.
A split architecture makes that story harder to tell than it needs to be. The finance record lives in SAP. The quality record lives partly in SAP and partly in a folder structure. The original supplier PDF lives in a mailbox or a scanned archive with a filename convention that made sense to somebody who left in 2019. Connecting them takes a person with institutional memory, and that person is usually in a meeting.
A unified document workflow produces something different. For a given batch, one query returns the original supplier COA as received, the extracted values and who or what posted them, the inspection results against the plan, the usage decision and its author, the goods receipt, the invoice that matched that receipt, and the timestamp on every one of those events. Nothing reconstructed. Nothing inferred.
That record answers the traceability question and the control-of-suppliers question with the same click. It also answers a question auditors have started asking more often, which is how the manufacturer knows the supplier document is authentic and unaltered. Capturing the source file alongside the extracted values, with a hash and a timestamp, is a much better answer than a printout in a binder.
The audit trail work Artificio has published goes deeper into the specific evidence structures. The relevant point for this discussion is that a unified trail is not an extra deliverable. It falls out of the architecture for free, provided both document streams were built against the same spine in the first place.
The composites case in practice
The Gill Corporation is the kind of operation this pattern was built for. Aerospace composites, honeycomb core, bonded panels, floor structures, a supply base that spans resin systems, adhesives, aluminium, and specialty chemistry, and an AS9100 quality system that has to hold up under the scrutiny of every prime it ships to.
An operation like that receives thousands of shipments a year, and a meaningful share of them carry both a certificate and an invoice. Every one of those shipments is a small fork in the road where the paperwork splits and the effort doubles.
Run rough numbers on a plant receiving two hundred qualifying shipments a month. If the COA takes twelve minutes of quality technician time to transcribe and verify, and the invoice takes eight minutes of AP time to key and clear, that is roughly sixty-seven hours a month of pure document handling before anyone has made a single judgement call. Add the exception work, which typically consumes more time than the clean cases, and the real figure climbs considerably higher.
Compress that with a unified workflow and the time saving is the least interesting part of the outcome. The interesting part is what happens to shelf life. Material that used to sit in restricted stock for three or four days waiting on a transcription now clears in hours, and the out-life clock that started at the supplier's plant does not burn down in your own receiving area. For prepreg with a limited freezer life and a much shorter room-temperature out-life, days matter to production planning in a way that finance metrics never capture.
The second interesting outcome is the disappearance of a specific recurring argument. When production cannot start because material is still restricted, the conversation between planning, quality, and receiving usually turns into a search for whose queue the document is stuck in. That conversation ends when there is one queue.
How manufacturers actually get there
Nobody sequences this as a single programme, and they should not try.
The realistic path starts with whichever stream is currently causing more pain. For most manufacturers that is accounts payable, because the cost is visible on a report every month and the discount leakage is easy to quantify. Get the invoice stream reading cleanly, matching against the PO and receipt, and routing exceptions to the right owner. That work establishes the connection to SAP, the supplier master mapping, and the exception handling patterns.
The second phase adds the quality stream on top of the same foundation. The extraction models are different because a COA is a different animal from an invoice, but the resolution logic, the SAP write path, and the exception routing are largely already built. Adding COA processing to an existing document layer is a fraction of the work of standing one up from scratch.
The third phase turns on the handshakes. This is the smallest technical step and the largest operational one, because it changes who sees what. Quality holds become visible to AP. Receipt discrepancies become a shared exception. Supplier scorecards start carrying document accuracy alongside on-time delivery and first-pass yield. Expect this phase to require more conversation than configuration.
Some manufacturers run the sequence in reverse, starting with quality because a customer audit or a shelf-life incident forced the issue. That works equally well. What does not work is running both as independent procurement decisions with different vendors, different data models, and different integration patterns, because then the third phase never happens. The handshakes require a shared spine, and you cannot bolt a shared spine onto two systems that were never designed to have one.
The shipment, revisited
Go back to that pallet of prepreg on the Tuesday morning dock.
In the unified version, the receiving clerk still scans the packing slip and still posts the goods receipt. Eleven minutes, same as before. What changes is everything downstream.
The supplier's COA, which arrived by email the previous evening, has already been read. Its vendor batch number is waiting to be matched to whatever batch the receipt creates. The moment the receipt posts, the inspection lot generates, the characteristic results land against it, and the values get compared to the inspection plan. Everything sits inside specification, so a usage decision is proposed and the material moves to unrestricted stock before the clerk has finished with the next truck. The out-life clock started at the supplier and has lost hours rather than days.
The invoice arrives Thursday. It matches the PO line and the receipt on price and quantity. The freight line does not match, because it was never on the PO, so it goes to purchasing with the original document attached and a note showing exactly which value failed. Purchasing approves it in the afternoon. The invoice posts, inside the discount window.
Somewhere in the middle of all that, an audit trail was written that connects a customer part number to a batch, to an inspection result, to a supplier certificate, to a goods receipt, to an invoice, to a payment. Nobody assembled it. It exists because the two document streams were built to run against the same spine.
That is what end-to-end means in an aerospace context. Not the removal of judgement, because quality engineers and AP leads still make the calls that require experience. The removal of the gap between two document streams that were always describing the same shipment, and the recovery of the hours that gap has been quietly consuming.
Artificio builds document automation for SAP-based manufacturers, with AP Studio for supplier invoice processing and SAP QM integration for supplier certificates and batch release. To see how the unified workflow maps to your receiving process, get in touch at support@artificio.ai.
