Blog
When a single delivery document references multiple SAP purchase orders, split the quantity accurately. Learn how to rank evidence, manage account risk, and eliminate invoice blocks.
A pallet arrives at dock 3 a little after two in the afternoon. The delivery note is clean, typed, and completely legitimate. Up in the header, under "Your Order Ref," the supplier has printed two purchase order numbers, 4500071203 and 4500071389. Below that sit six lines of material. Five of them are easy. Line 4 is not.
Line 4 reads "M10 hex bolt, zinc plated, 2,500 EA." Both purchase orders include that bolt. PO 4500071203 has 1,500 still open on item 30 at $0.42 each. PO 4500071389 has 2,000 open on item 10 at $0.39, because purchasing renegotiated the price six weeks later and then expedited the order. The delivery note does not say how the 2,500 bolts divide between the two.
The receiving clerk makes a sensible call. Oldest order first. She posts 1,500 against the older PO in MIGO, puts the remaining 1,000 against the newer one, and moves on to the next truck. Three weeks later, accounts payable finds a blocked invoice, a GR/IR balance that will not clear, and a buyer who swears the supplier shipped the expedited order in full. Nobody did anything careless. The document simply asked a question that nobody on the dock had enough information to answer.
This post is narrow on purpose. It does not walk through procure-to-pay or the full goods receipt workflow. It looks at one decision point. When a single delivery document references more than one purchase order, how do you decide which PO item each unit belongs to, and how sure do you need to be before you post?
A standard goods receipt in SAP has a tidy shape. Every item on the material document points at exactly one purchase order and one PO item. The delivery note number sits once, in the header reference field. SAP assumes that whoever posts the receipt already knows which PO line every quantity belongs to.
Suppliers do not think that way. Their warehouses pick by material and ship by truck, not by customer purchase order. If a customer has three open orders for the same bolt, the picker grabs 2,500 bolts from one bin and the shipping system prints one line. From the supplier's side, this is efficient and perfectly correct. From the receiving side, it breaks the one-item, one-PO-line assumption that MIGO depends on.
EDI partners usually avoid the problem because the message structure forces clarity. An ANSI X12 856 advance ship notice nests orders inside the shipment, so each PO gets its own hierarchical level with its own items underneath. A DESADV message carries order references at the line level in a similar way. Paper and PDF delivery notes follow no such rules. Some print the PO number on every line. Some print it once at the top. Some print the supplier's own sales order number and leave the customer to work it out. Most mid-size manufacturers receive a mix of all three every single day.
That mix is why a multi-PO delivery deserves its own decision logic instead of being treated as a messy version of a normal receipt.
Matching logic only makes sense once you recognize which shape of document you are holding. Each one calls for a different amount of inference.
The supplier prints the PO number, and often the PO item number, next to each line. This is the easy version. The work is mostly verification, confirming that the referenced item is still open, belongs to this vendor and plant, and has enough open quantity to absorb the delivery.
The header lists two or three orders, and the lines underneath carry material descriptions and quantities with no order reference at all. If each material appears on only one of the listed POs, the mapping resolves itself. If a material appears on more than one, you are into inference mode.
This is the bolt problem from the opening. The supplier consolidated demand from several orders into a single line with a summed quantity. Now the receipt has to split one quantity across multiple PO items, and the document gives no split.
The delivery closes out the last 400 units of an older PO and starts filling a newer one. On paper it often looks identical to Shape 3, but the intent is different. The supplier is finishing one order, not dividing a shipment evenly, and the correct allocation usually closes the older item completely.
The document cites a PO already marked delivery completed, a number with two digits transposed, a scheduling agreement instead of a standard PO, or the supplier's internal sales order number. The goods are real and expected. The reference is not usable as printed, so the match has to come from somewhere else.
A single delivery note can contain several of these shapes at once. Line 1 might be Shape 1, lines 2 and 3 Shape 2, and line 4 Shape 3. Good matching logic works line by line, not document by document.
It is tempting to treat a misallocated receipt as a paperwork problem. The stock is in the building, after all. Follow the bolt example through SAP and the cost becomes much clearer.
The clerk posted 1,500 units against PO 4500071203 item 30 and 1,000 against PO 4500071389 item 10. The supplier's billing system, which knew exactly which order it was filling, invoiced 500 units on the older PO and 2,000 on the newer one. The supplier had shipped the expedited order in full and added 500 from the older order to fill out the pallet.
When the invoice comes in through MIRO, the match runs per PO item. On the newer PO, the invoice claims 2,000 units against a goods receipt of 1,000. That quantity variance sits far outside tolerance, so the invoice is blocked for payment. On the older PO, the invoice claims 500 against 1,500 received. Nothing blocks, but 1,000 units now sit on the GR/IR clearing account at $0.42 each, a $420 balance with no invoice coming to meet it.
At least not yet. The supplier still owes 1,000 bolts on the older order and will ship them eventually. When that delivery arrives, SAP shows the older order as fully received and the newer order with 1,000 open. The clerk, quite reasonably, posts against the newer PO. The supplier invoices against the older one. The same mismatch appears again in mirror image, and one bad call on a Thursday afternoon has now spread to a second delivery.
The inventory side takes a hit too. Those extra 1,000 bolts on the older order were valued at $0.42 instead of $0.39. If the material runs on moving average price, the stock value is off by $30. If it runs on standard price, the difference lands in the price variance account against the wrong order. The amounts are small for one line. A plant receiving a few hundred multi-PO deliveries a month turns them into a steady trickle of blocked invoices, month-end GR/IR clean-up in MR11, and variance postings nobody can explain.
Purchasing and planning feel it as well. The total open quantity across both orders is still correct, but it sits on the wrong order with the wrong dates. The buyer watching the expedited order sees 1,000 units outstanding and calls the supplier to chase bolts that are already on the shelf. The older order, which the supplier genuinely still owes, looks finished, so nobody follows up on it. MRP plans against supply that is misplaced in time.
The fix is manual and slow. Someone reverses the receipts with movement type 102, reposts them against the correct items, and hopes nobody consumed the stock in between. Almost all of that pain traces back to a single moment on the dock when the allocation was a guess.
The better approach separates two questions that usually get collapsed into one.
The first question is about allocation. Given everything we know, what is the most likely split of this delivery across the open PO items? The second question is about confidence. Is the evidence strong enough to post that split directly, or does it need a safer holding position or a human answer first?
Keeping those questions apart matters because the answer to the second one depends on more than evidence. It also depends on how much damage a wrong answer would do. A split between two identical PO lines with the same price and account assignment carries little risk. A split between a stock PO and a PO charged to a capital project carries a lot.
Every allocation starts with the list of PO items that could legitimately receive the goods. Get this list wrong and nothing downstream can rescue the match.
The candidate set begins with open purchase order items for the same vendor and receiving plant. Material matching comes next. If the delivery note shows your internal material number, the match is direct. More often it shows the supplier's own part number, so the lookup runs through the purchasing info record, where the vendor material number is stored against your material. Free-text descriptions need fuzzier matching, and those matches should carry less weight later on.
Then come the exclusions. Items flagged for deletion drop out, and so do blocked items. Items marked delivery completed drop out as well, though they deserve a second look when the supplier explicitly references them. Items for a different plant come out entirely, since a receipt has to post to the plant on the PO item. Items with a different storage location stay in but rank lower, because the storage location can be changed at receipt.
For each surviving item, the open quantity is the ordered quantity minus everything already received, taken from the PO history. The over-delivery tolerance on the item defines how far above that open quantity a receipt can go before SAP pushes back. That tolerance becomes a hard ceiling in the split calculation. It is not something to lean on for convenience.
What remains is usually two to four candidate items per contested line. That is a small enough set to reason about carefully.
Not all clues are equal. A reliable matching approach ranks evidence by how directly it reflects what the supplier intended, and it lets stronger evidence override weaker evidence rather than averaging everything together.
Explicit line-level references sit at the top. If the supplier printed the PO number and item next to the line, the supplier is telling you what it shipped. The main job is validation.
Carton and pallet labels come next, and they get overlooked constantly. Many suppliers print the customer PO number on every carton label even when the delivery note consolidates lines. A photo of the labels, taken at the dock, can turn a Shape 3 line back into a Shape 1 line. If 16 cartons carry the newer PO number and 4 carry the older one, and each carton holds 125 bolts, the split is 2,000 and 500. No guessing required.
Supplier confirmations are strong evidence as well. When a PO uses confirmation control, the supplier's order acknowledgment or shipping notification is stored against the PO item with a confirmed quantity and date. If the supplier confirmed 2,000 units on the newer PO for delivery this week and confirmed nothing on the older one, the allocation leans heavily toward the newer order.
Exact quantity closure is a useful middle-strength signal. If a consolidated quantity exactly closes one item and partly fills another, it points to Shape 4 behavior. Suppliers like to finish orders.
Price printed on the delivery note helps when the candidate POs carry different prices. A note that shows $0.39 per unit is quietly telling you which order it was filling.
Schedule line dates provide an ordering rule when stronger evidence is missing. Filling the earliest due schedule line first is a reasonable default. It is still a default and not evidence, and it is exactly the rule that went wrong in the opening example.
Supplier history comes last. Some suppliers always ship expedited orders first. Others always fill oldest first. Some consolidate everything, while others never do. A pattern observed across hundreds of past receipts and their matching invoices is real evidence, but it should only break ties among weaker clues. It should never outrank a label or a confirmation.
With candidates listed and evidence ranked, the split itself follows a short, disciplined sequence.
Quantities with direct evidence get assigned first. Label counts, explicit references, and confirmations each claim their share before any inference happens. In the bolt example, if the confirmation shows 2,000 on the newer PO, that item receives 2,000 immediately.
The remainder gets distributed next. The leftover 500 goes to the next best candidate, the older PO with 1,500 open. That remainder sits well within the open quantity, so no tolerance question comes up.
Every item then gets checked against its ceiling. No PO item should receive more than its open quantity plus its over-delivery tolerance. If the math only works by pushing one item past its ceiling, the split is probably wrong or the supplier over-shipped. Either case needs a human decision.
Last comes the comparison with the runner-up. What would the second most likely split look like, and how different is it? That comparison feeds straight into the confidence decision.
Run the same logic without labels or confirmations and you get a very different picture. With only schedule dates to go on, the older PO fills first, exactly as the clerk did it. Good allocation logic recognizes that this split rests on a weak default, and the posting decision reflects that weakness instead of hiding it.
This is where the confidence bar should move. A practical way to frame it is the cost of being wrong, meaning the difference in financial and operational outcome between the best split and the second best.
When candidate PO items share the same price, the same account assignment, the same plant, and the same storage location, a wrong split costs very little. Invoices may need a nudge, but valuation, cost objects, and stock locations all stay correct. Modest confidence is enough to post.
Different prices raise the stakes, since every misallocated unit carries a valuation or variance error. The larger the price gap and the larger the quantity, the stronger the evidence needs to be.
Different account assignments raise them further. If one PO item is a standard stock item and another carries account assignment category K for a cost center or P for a project, the split decides which budget absorbs the cost. Moving 500 units of spend onto a capital project by mistake is not a GR/IR clean-up issue. It is a controlling issue, and it may surface in a project review months later. These lines should rarely post automatically without explicit evidence.
Batch-managed materials and QM-relevant receipts add another layer. If the delivery carries batch numbers or certificates of analysis tied to specific orders, putting the wrong batch on the wrong PO creates traceability gaps that matter far more than the dollar amounts.
Setting thresholds for this does not need to be complicated. Many teams start with a simple rule. Auto-post when the evidence is strong or the cost of being wrong is trivial. Hold when the evidence is moderate and the difference is material. Escalate when the evidence is weak and the difference touches account assignment, batches, or large sums.
Once the split and the confidence level are settled, SAP gives you three practical landing spots.
For confident allocations, post one material document per physical delivery. MIGO accepts items from more than one purchase order in the same document, and so does BAPI_GOODSMVT_CREATE when an integration posts the receipt, with each item carrying its own PO number and PO item. The delivery note number goes into the header reference field and the bill of lading into its own header field.
That header discipline pays off later. MIRO can pull goods receipt items by delivery note instead of by purchase order, so an AP clerk or an automated invoice match can find every receipt tied to one delivery in a single step, no matter how many orders it touched.
One detail catches people out. The BAPI posts all items or none. If one contested line fails validation, the whole document fails with it. The cleaner pattern posts the confident lines together and handles the contested line separately, with the same delivery note number in the header of both documents so the link survives.
For moderate confidence with material consequences, movement type 103 records the goods as physically received against the best-supported PO item without creating valuated stock or a GR/IR posting. The receipt is visible in SAP and the dock is not sitting on a mystery pallet, but no accounting document exists yet.
When the buyer confirms the allocation, or the supplier's invoice arrives and settles the question, a 105 release posts the valuated receipt. If the allocation turns out to be wrong, reversing a 103 with movement type 104 is far cleaner than reversing a 101, because there is no financial document to unwind. This option still needs a PO reference, so it works best when the uncertainty is about quantities between known items rather than about which orders the goods belong to at all.
For weak evidence on high-risk lines, the right answer is a person. What separates a good escalation from a bad one is what the buyer actually receives. A forwarded PDF with "please advise" creates work. A short message showing both candidate PO items, their open quantities, prices, account assignments, the evidence found, and a proposed split turns a fifteen-minute investigation into a ten-second confirmation.
Plants that run inbound deliveries through confirmation control have one more option. The allocation can be settled at the inbound delivery stage, when the shipping notification comes in and before the truck even arrives. The same evidence ranking applies. It simply runs earlier.
Artificio's AI agents treat a multi-PO delivery as a reasoning task rather than a template extraction job. The document agent reads the delivery note, packing list, and any carton label photos captured at the dock, and it recognizes which of the five shapes each line follows. It does not need a template per supplier, which matters when the same supplier formats notes differently for different plants.
The matching agent pulls open PO items, PO history, info records, and supplier confirmations from SAP. It builds the candidate set, ranks the evidence in the order described above, and calculates the split. Every item gets checked against its tolerance ceiling, and the best allocation gets compared with the runner-up to estimate the cost of being wrong.
Posting follows the three landing spots. Confident lines post as a single material document with the delivery note carried in the header. Moderate-confidence lines with a material difference go to GR blocked stock. Weak-evidence, high-risk lines reach the buyer as a ready-to-approve proposal with the evidence laid out. Every decision keeps its reasoning attached, so when AP opens a variance three weeks later, the answer to "why was it split this way" is already on record.
The agents also learn from outcomes. When supplier invoices arrive, Artificio compares invoiced quantities per PO item with the receipt allocations. Agreement reinforces the supplier pattern. Disagreement gets flagged and changes how that supplier's future consolidated lines are weighted. Over a few months, the share of lines needing human review shrinks for exactly the suppliers that cause the most trouble today.
None of this changes your SAP configuration. Tolerance keys, confirmation control, account determination, and release strategies stay as they are. What changes is that the allocation decision stops depending on who happens to be working the dock that afternoon.
Teams that want to tackle this problem, with or without automation, can learn a lot from their own data before changing anything.
Start with blocked invoices. Pull the last quarter of MIRO invoices blocked for quantity variance and check how many of the related goods receipts shared a delivery note number with receipts on other purchase orders. That number is your baseline, and it is usually larger than people expect.
Check header discipline next. If receipts get posted without the delivery note number in the reference field, or with inconsistent formatting, MIRO cannot use the delivery note as a reference and every downstream match gets harder. This is a training fix that costs nothing.
Then talk to your top ten consolidating suppliers. Many of them can print PO references at line level or on carton labels with a small change to their shipping templates. They rarely do it unless a customer asks.
Finally, agree on thresholds with purchasing and controlling. Decide together which differences in price, account assignment, or batch handling justify holding a receipt. Writing that rule down takes the pressure off receiving clerks, who should not be making judgment calls that belong to someone else.
A delivery note that references two purchase orders is not a rare edge case. It happens every day in plants that buy the same materials across multiple orders from the same suppliers. The dock team usually handles it with a reasonable rule of thumb, and that rule of thumb is right often enough that nobody notices the cost until month-end.
The shift is small in concept. Treat allocation as a question of evidence, not habit. Let the cost of being wrong set how much evidence you need. Give yourself a middle option between posting and waiting. Those three changes turn a guess on the dock into a decision that holds up at invoice verification, in the GR/IR reconciliation, and in the next MRP run.
Split shipments are one exception pattern among many in SAP receiving. The same narrow treatment works for other single decision points, like a delivered material that does not match the ordered one, or a delivery that shows up against a purchase order someone already closed. Each deserves its own rule, made once and applied every time.
Lal Singh, SAP AI Automation Expert
CEO & Founder of Artificio
See it in your SAP environment
Bring us a document, a process, or a bottleneck. We'll show how Artificio captures, validates, and posts into SAP — then scale from there.
Security & compliance
ISO 27001:2013 certified, SOC 2 Type 2 compliant, GDPR and HIPAA ready. Every agent action is logged, auditable, and runs in isolated environments.
