The second pallet arrives on a Thursday and nobody ordered it. Except somebody did, twice, six weeks apart, because the first purchase order never showed up as anything. The stock sat at the supplier, the incoming column read zero, the low stock report put the SKU back on the buy list, and a reasonable person bought it again. Now there is four thousand dollars of inventory in the aisle, the supplier is confused, and the person who placed both orders did nothing wrong.
This is the most expensive Shopify gap I know of that almost nobody writes about, and it comes down to one word being defined more narrowly than everyone assumes.
The short answer
Incoming inventory on Shopify counts units in transit, not units you have bought, so a purchase order the supplier has not shipped reads as zero and the low stock report tells you to buy it again. The purchase order and its linked inventory transfer operate independently by design, Shopify Flow cannot read incoming quantities or purchase order data at all, and the purchase order API that landed in the 2026-10 Admin GraphQL version carries exactly two statuses, DRAFT and ORDERED, with no received state and no expected arrival date. The playbook builds the on order number Shopify does not give you, dates every open purchase order, flags the late ones, and warns the buyer before the same pallet is ordered twice.
Incoming does not mean on order
Shopify's Help Center says incoming inventory is inventory that is on its way to your location from inventory transfers or apps. Units that have not been received are excluded from on hand and cannot be sold. Available, the number your storefront reads, explicitly excludes incoming as well.
Read that definition again and notice what it is about. It is about movement. A merchant who tested this on a development store in September 2026 put it plainly: the number shown as incoming on a product is narrower than it looks, because incoming counts units that are in transit. Transfers still in draft add nothing. Transfers marked ready to ship but not yet dispatched add nothing. A purchase order you placed three weeks ago that the supplier has not shipped shows as zero.
So incoming is a transit number. The number you need for a buying decision is an on order number: what you have committed to buy and not yet received, regardless of whether it has moved. Shopify has no field for that, no column for it, and no report that shows it.
And then the reports tell you to buy it again
On 24 July 2026 a merchant posted the consequence exactly. His custom purchasing reports depend on incoming quantities, and after the new purchase order system launched, incoming stopped reflecting open purchase orders. His line was: my reports keep telling me to procure what is already in process. He asked for the old purchase order system back and flagged the obvious risk to capital management. No staff reply.
Thirty minutes later he posted his own workaround, which is the part worth sitting with. Create a transfer from the purchase order and mark it in transit before it has shipped. In other words, tell Shopify the goods are moving when they are not, so the number comes out right. He called the extra steps frustrating. They are also now standard practice in a lot of stores.
Another member explained the mechanics the same evening. Incoming is derived from transfers and shipments, not from purchase orders, so the transfer step is mandatory. Then the warning that matters most: editing the purchase order afterwards, or receiving a short shipment, will not update the transfer, which can silently re-create overcounting. His suggestion was to store the purchase order number in a transfer metafield so reports can join the two records back together.
That is the whole trap in three sentences. Undercounting makes you double order. The fix for undercounting creates overcounting. And the join between the two records is something you have to build yourself out of a metafield.
The purchase order and its transfer stop talking
On 16 September 2026 a merchant asked for something that sounds like it should already be true: when you add products to a purchase order, the linked inventory transfer should update too. His words were that the current workflow requires duplicate manual edits. The reply quoted Shopify's own documentation back at him. Once a purchase order and an inventory transfer are linked, they operate independently.
So you keep two records of the same delivery. One is the commercial agreement and one is the count, they are linked in name only, and every edit after the link has to be made twice. Nothing tells you when they drift.
Two years, two systems, the same hole
This is not new and it did not arrive with the new system. In April 2024 the same merchant asked why Stocky purchase orders were not syncing to Shopify's incoming inventory. A week later another merchant replied that he had at least five active or partially received purchase orders and that Stocky was completely unaware of these incoming inventories, which made the low stock report useless. Support told him to re-enter the data manually. In July 2024 a third merchant asked whether anyone had solved it. No staff reply, no resolution.
Stocky shut down on 31 August 2026 and the native purchase order screen inherited the job. In the thread where merchants worked out what to do about the sunset, one of them wrote in May 2026 that the native purchase order system is fundamentally incomplete. Another reported that the only option support offered was to pay for an ERP. A third shared a support table listing minimum and maximum replenishment as not planned for 2026, and bulk CSV import as not currently available.
Flow cannot see any of this
On 26 January 2026 a merchant asked for a workflow that lists products with incoming inventory above zero, sorted by purchase order and vendor. He had already asked Sidekick, Shopify's own assistant, and quoted the answer: Shopify Flow cannot access incoming inventory quantities for reporting and automation. The thread was revived on 5 October 2026 and two developers confirmed nothing had changed, pointing at a custom app or a scheduled job against the Admin API instead.
Flow's inventory bench is short and entirely event-driven: inventory item created, inventory item deleted, inventory transfer completed, inventory transfer ready to ship, product variant back in stock, product variant out of stock, product variant inventory quantity changed. Every one of them needs something to happen.
A late purchase order is the absence of something happening. There is no trigger for a delivery date that passed, because Shopify does not hold a delivery date to pass. The only native shape left is the scheduled time trigger driving a sweep that you build and maintain. It is the right shape. It is not a feature.
The API finally got purchase orders, two weeks ago
Here is the genuinely new part, and most of the advice
online has not caught up with it yet. The 2026-10 Admin
GraphQL version exposes inventoryPurchaseOrders
and inventoryPurchaseOrder, behind a
read_inventory_purchase_orders scope. A
Shopify engineer confirmed on 2 October 2026 that the
scope is available to all apps now, outside the preview.
Getting there was messy enough to explain why nobody has
written about it. On 11 August a developer hit
ACCESS_DENIED on the query, found the scope was absent
from both the public access scopes reference and the
dashboard selector, and was told by a Shopify engineer
that declaring it in shopify.app.toml passes
validation and then fails at deploy. Shopify patched it
server side on 14 August so existing read_inventory
access would work on preview stores. Before that, every
feature request since June 2026 had gone unanswered,
including one on 30 July asking for a firm yes or no on
whether the API would arrive inside twelve months.
So it exists. Now read what it actually gives you. The object is defined as an agreement from a merchant to buy goods or services from a supplier, and it carries the name, the supplier as origin, the destination location, dateCreated, orderedAt, the line items, which can be searched by unit cost, variant id, supplier SKU and incoming quantity, and a transfers connection.
And the status enum has exactly two values.
DRAFT, open and not marked as ordered yet.
ORDERED, open and marked as ordered. That is
the entire lifecycle. There is no received, no partially
received, no closed and no late. There is an orderedAt
timestamp and no expected arrival anywhere on the object.
Which means the two questions you actually have, how much of this is still outstanding and which of these should have landed by now, are both things you compute from the linked transfers. The same transfers that, per Shopify's own documentation, operate independently of the purchase order.
What it costs while nobody counts
Take a store carrying four hundred SKUs across a dozen suppliers, with an average purchase order around four thousand dollars. Two duplicate orders a month is eight thousand dollars of working capital converted into stock you already owned, plus the storage, plus the markdown on whatever does not sell at full price. That is the loud half.
The quiet half is worse. Purchase orders that are simply late, with nothing watching the date. A thirty five day lead time slips by three weeks and nobody finds out until the shelf is empty and the product page says sold out. Then you are paying for air freight, or losing the sale, or both.
Both halves come from the same missing object: a list of what is on order, with a date on it, that something checks every night. Everything else here is a symptom.
The playbook worth building
As a Dugong playbook, in plain language:
# trigger
Nightly, and on every purchase
order status change, over all
open purchase orders
# steps
1. Pull every PO marked ordered
and not fully received, with
lines, supplier, destination
2. Reconcile: sum the linked
transfers and shipments per
line, keep the remainder
3. Date it: expected arrival
from supplier lead time when
the PO carries none
4. Flag late: past expected,
remainder still open, aged
in days not listed by date
5. Publish on order per variant:
ordered minus received, per
location, beside available
6. Block the duplicate: warn at
the moment someone drafts a
PO for a SKU already on order
7. Chase the supplier with the
PO number, the open lines
and the days late
8. Score lead time actual vs
promised, and short ship
rate, per supplier
Step five is the one everyone skips. Everything else is housekeeping. This one changes the decision. Write a real on order quantity per variant per location somewhere your reorder logic can read it, a metafield is fine, and the low stock report stops lying the same week. Until that number exists, every buying decision in the store is made against a count that is missing the orders you already placed.
Step six is where the money is. A warning at the moment someone drafts the order is worth more than any dashboard, because the duplicate order is not a reporting failure, it is a decision made with the wrong number in front of a person who is in a hurry.
Three numbers to write down
The build is a day. The decisions are what stall people, and all three are small. How many days past expected arrival makes a purchase order late, which is usually seven for domestic and fourteen for ocean freight. What short ship percentage you let through without chasing, which for most stores is about two percent. And the purchase order value above which a human reviews it before it goes to the supplier. Pick those three and the rest is configuration.
Where AI earns its place
Three jobs here, and none of them is running a query.
Matching the delivery to the order. Pulling open purchase orders is mechanical. Working out what arrived is not. The supplier's packing slip uses their part numbers, their case packs and their own wording for each line, and real deliveries come short, split across two shipments, or substituted. The purchase order line does carry a supplier SKU field, which helps on the clean ones and not at all on the rest. A model reads the supplier's document against your lines and proposes the match, which is what makes the remainder number trustworthy.
Finding the real date. Expected arrival almost never lives in Shopify. It lives in an email saying shipping Friday, a second one moving it to the week of the tenth, and a PDF with a carrier reference buried in it. A model reads that thread and keeps one honest date per purchase order, which is the input every other step depends on.
Writing the chase. The purchase order number, the lines still outstanding, the days past the date the supplier themselves gave you, in a tone that keeps the relationship. Most late deliveries move when somebody asks on day three instead of day thirty.
Keep three things human. Sending a purchase order above whatever value you set. Accepting a substitution. And any change to agreed cost.
Do this before you close the laptop
Open Products, then Purchase orders, filter to ordered, and sort oldest first. Count how many were created more than one supplier lead time ago and are still sitting there. Then take three of them, pick one variant from each, and look at its incoming count. In most stores at least one of the three reads zero, which means your reorder report has been treating that stock as if it does not exist since the day you bought it.
Sitting on purchase orders nobody has looked at since
they were sent, a supplier who is three weeks late and
does not know you noticed, or a low stock report you have
stopped trusting? The inbox is open:
field-notes@dugong.live. We are collecting
case studies for the next issue.