The email arrives with a photo attached. Blue, size medium. The order said black, size large.
You open the order. Black, size large. You open the fulfillment. Black, size large. Everything in Shopify agrees with everything else in Shopify, and none of it was in the room when the bag was sealed.
So you do the only thing left. You believe the customer, you post the right item, you apologize. By Friday you have done it four times and you could not name one of the four without scrolling back through your inbox.
The short answer
Shopify records what was ordered twice, once on the order and once on the fulfillment, and records what was picked exactly never. The shipment record is built from the order record, so it cannot contradict it, which means it can never be evidence. There is no second version of the truth to check the first one against.
Shopify does have a name for the failure. The return reasons enum carries a value for the customer receiving the wrong item. But a return reason is set after the customer opens a return, so the platform hears about the mis-pick from the one person whose account of it you cannot verify.
By then the order is frozen. Fulfilled line items cannot be removed or re-quantified, so the record of what you sold can never be corrected. Every fix is a second document: a return, a refund, or a new order that puts a sale in your numbers that never happened.
Two stock figures have also gone wrong, both in the direction that oversells. The item on the order was decremented and is still on your shelf. The item in the box was never decremented and is gone.
There is one piece of real evidence and you already pay for it. The order knows what the parcel should weigh. The carrier weighs what you actually handed over. Nothing in Shopify puts those two numbers next to each other, and that comparison is most of the automation.
So the job splits in two. Catch the light parcel on your own bench, before the customer finds out. And when a claim does land, read it, attach it to the order, settle it the same way every time, and count the pairs so next month is better.
The shipment record is a copy of the order, so it cannot disagree
Look at what a fulfillment actually stores.
FulfillmentLineItem
is documented as a line item from an Order that is included in a
Fulfillment, linking the fulfillment to specific items from the
original order and tracking how many units were fulfilled. Its fields
are an id, a created timestamp, two money totals, a quantity, and
lineItem, the pointer back to the order.
Read that list again for what is missing. No packer. No scan. No photo. No serial or lot. No weight. The fulfillment answers one question, how many of this ordered line went out, and it answers it by repeating the order back to you.
That is not an oversight. It is what a fulfillment is for. Shopify's job is to tell the carrier and the customer that an order has shipped. Whether the right thing is inside the box is a warehouse fact, and Shopify has never claimed to hold warehouse facts.
The consequence is the one that bites on a Tuesday morning. When a customer says they got the wrong item, there is no second record to check against the first. You are not short of a report. You are short of a witness.
Shopify names this failure, then only hears about it from the customer
The platform is not silent on mis-picks.
ReturnReason
in the Admin API has ten values, and one of them is
WRONG_ITEM, documented as the item is returned because the
customer received the wrong one and displayed to the customer as
Received the wrong item. Around it sit the ordinary ones: size too
large, size too small, style, color, unwanted, defective, not as
described, other and unknown.
So with self-serve returns switched on, a customer can tell you exactly this, in a structured field you can read back later. That is worth having, and switching it on is the ten-minute version of this article. Customers request returns on items that have been delivered, you customize the request form to collect return reasons and product condition, and you approve or decline each request against your own policy.
Two things to know before you plan around it.
The reason describes what arrived, not what left. A return tagged wrong item tells you a claim was made. It does not tell you a mis-pick happened. Those are different facts, and the difference between them is your margin.
Most of these claims never reach the portal. They arrive as a reply to the shipping confirmation, with a photo, asking for the right item rather than a refund. Flow's seven return triggers, return requested, approved, declined, canceled, closed, processed and reopened, all need a return object to exist first. Nothing in the trigger list fires on an inbound customer email. The event that starts this workflow is not an event Shopify can see.
One mis-pick breaks two stock counts, and both break the same way
This is the part that costs money quietly, long after the customer has been made happy.
The order was for the black large, so Shopify decremented the black large. The packer reached one bin over and sent the blue medium.
Your shelf now holds a black large that Shopify thinks is sold, and it does not hold a blue medium that Shopify thinks is available. Two SKUs, both overstated. The blue medium is live on your storefront and will oversell the moment somebody buys it.
The merchant who mailed a customer the wrong item in December 2022 worked this out on his own. He duplicated the order, removed the items he was not resending, marked it as paid and printed a label, and then said he would go and count inventory in case the wrong item had also been taken out of stock. He was right to worry, and he was doing it by hand, from memory, on a Wednesday.
Nothing automates it, because nothing knows. The refund path will restock whatever the customer sends back, which is the wrong item. Nobody ever tells Shopify to put the right one back on the shelf.
The replacement nobody has a clean way to send
You cannot repair the original. Shopify's considerations for editing orders are blunt about it: you can edit unfulfilled items in an order, but you cannot remove fulfilled items or adjust their quantity. Once the wrong thing has shipped, the sale is written in ink.
That leaves three documents and no obviously good one. A duplicate order marked as paid gets the label printed and books revenue that does not exist. A draft order marked as paid does the same thing one menu over. Both were suggested in that December 2022 thread, the draft order by a community voice and the duplicate by the merchant himself, and nothing on the thread is marked as a solution.
The reporting worry turned up later in the same place. In October 2023 another merchant asked whether draft orders affect gross and net sales and sales tax reports, which is the right question. The answer was that draft orders are stored separately and their tax should not appear in reports until the order is paid. Which it is, because marking it paid was the only way to print the label.
The cleaner native path is the one fewer merchants use. Open a return on the original order and add the correct product as an exchange item, where you can apply a discount to its price with a reason. One order, one customer record, one trail, and no invented revenue. It costs you a few more clicks in the admin, which is exactly why it is the step worth handing to a rule.
The evidence already exists and nothing compares it
The order knows its weight. The Order object carries a total weight in grams, and a second field for the weight after returns and refunds. Every variant with a weight entered against it contributes. That figure is a prediction of what the parcel should scale at.
The carrier measures the other half. Every parcel crosses a conveyor and gets weighed, and the scanned weight is the number they bill you on. You are already paying for a second opinion on every shipment you send.
Put the two side by side and a mis-pick becomes arithmetic. A blue medium where a black large should be is eighty grams light. A missing item is a whole item light. Two of something instead of one is an item heavy. One variance is not proof, but a variance on the parcel a customer later complains about is as close to proof as this job gets, and you can have it before the complaint rather than after.
The merchant in the November 2023 short shipment thread reached the same conclusion from the wrong end. Running the store alone, he had combined two same-day orders into one bag, moved up from a 12 by 15 to a 14 by 19 mailer, photographed it, and could say the new weight is 3 lbs. When the customer reported an item missing and filed a carrier claim at the same time, his plan was to ask UPS for the scanned weight from processing. Nobody replied to him. No staff answer, nothing marked as solved.
He was right that the weight was the evidence. He was two weeks late to use it, and he was about to go and ask a carrier for a number that was already sitting in his own shipping account.
What this costs on an ordinary month
Three thousand orders a month at 99.5 percent pick accuracy, which is a good warehouse, is fifteen wrong boxes.
Say twelve of the fifteen customers write in. Each costs a replacement unit at cost, a second label, twenty minutes of somebody's attention and one email you had to word carefully. Call it forty dollars and twenty minutes, so roughly five hundred dollars and four hours. The three who do not write in are the expensive ones: one leaves a review, one never comes back, and one raises a dispute you find out about from your payment provider.
Then the stock. Thirty SKU-level figures are now wrong, fifteen of them in each direction. You will find them at the next count, or you will find them when something oversells, and when it oversells you will blame the channel sync.
None of that is the real cost. The real cost is that nobody knows which fifteen, so nobody knows that eight of them were the same two variants sitting in adjacent bins with near-identical labels. That is a half-hour fix nobody got to make, repeating every month.
What the automation actually has to do
As a Dugong playbook, in plain prose:
# trigger
Fulfillment created, plus
inbound customer message
# steps
1. Capture the parcel weight
when you buy the label
2. Compare it to the order's
expected weight
3. Hold and recheck the parcels
outside tolerance
4. Pull every claim into one
queue with the order attached
5. Replace as an exchange, not
as a new sale
6. Correct both stock counts,
not just one
7. Count mis-picks by SKU pair,
not by customer
Steps one to three are the only part of this that happens before the customer finds out. Read the order's expected weight, read the weight on the label you just bought, and flag the gap while the parcel is still on the bench. Set the tolerance in grams rather than as a percentage, because a percentage is generous on heavy orders and paranoid on light ones. Most stores land near the weight of their smallest sellable item.
Step four is unglamorous and pays the same week. A wrong item claim arrives as prose with a photo, so somebody has to read it, work out which order it belongs to, work out which line is wrong, and decide. Reading an email, matching it to an order and pulling out which SKU the customer says they received is exactly what a language model is good at, and it is the step that currently eats the twenty minutes.
Step five is a policy decision you make once. An exchange on the original order keeps the trail and invents no revenue. A duplicate order is faster to click and lies to your profit and loss. Write down which one you use, and let the rule do it the same way every time, including at 6pm on the last day of the month.
Step six is the one everybody forgets, and it has to be two writes. Decrement the SKU that actually left and restock the SKU that never did, at the location that shipped it. Do it as part of settling the claim, not as a note for the next stocktake, because the next stocktake is in March.
Step seven is the only step that makes next month cheaper. Count the pairs: shipped X when Y was ordered. Fraud clusters on customers and mis-picks cluster on pairs, which is why counting by customer finds you one bad actor and counting by pair finds you the shelf. When the same pair appears three times, the cause is usually a bin, a label or two variant names nobody can tell apart at arm's length, and the fix takes less time than the three replacements did.
Why this is a workflow and not a checklist
Every step here is a two-minute job. Weigh the parcel, read the email, send the right thing, fix the stock, write it down. The reason it does not get done is that it has to get done during the twenty minutes you are already behind on picking, for an order that has already gone out, in front of a customer who is already annoyed.
The judgment in it is also smaller than it feels. You decide two things once: how many grams count as a gap, and whether a replacement goes out as an exchange or as a new order. After that it is a subtraction, a lookup, two inventory writes and a tally.
The limits are honest ones. A weight check cannot catch a swap between two variants that weigh the same, which is most color swaps, so it narrows the problem rather than solving it. It needs variant weights filled in, which is the same prerequisite as accurate shipping rates and is missing more often than anyone admits. It cannot tell you whether a claim is true, only whether your own numbers support it. And the pattern counting only earns its keep if somebody actually walks over to the shelf and moves a bin.
Do this before you close the laptop
Three checks, ten minutes, no code.
Search your support inbox for wrong item and missing from order over the last ninety days and count the threads. That number is your mis-pick rate and your claim rate added together, and it is almost always bigger than the number you would have guessed.
Open your shipping account and compare the scanned weight on any three labels from last week against what those orders should have weighed. If you cannot do the comparison because your variants have no weights, you have just found the first job.
Then take your last ten wrong item claims and write down the pair for each: what shipped against what was ordered. If any pair shows up twice, you have found something worth more than the ten replacements put together.
If you sell on Shopify and have a version of this story, the photo of
the wrong color, the replacement you sent as a duplicate order, the
oversell you traced back to a mis-pick three weeks later, the inbox is
open: field-notes@dugong.live. We are collecting case
studies for the next issue.