CPG EDI: The Seven Documents a Food Brand Exchanges

A KeHE purchase order arrives at 6:10 on a Tuesday morning as an EDI 850. By 6:11 the brand's provider has sent back a 997 saying it arrived. Everyone on the brand's side treats that second message as confirmation the order is understood. It is not. The 997 reports "the results of the syntactical analysis" and "does not cover the semantic meaning of the information encoded", meaning the envelope was well-formed. Nobody has checked whether the case pack on the order matches the case in the warehouse.

That gap, between a document that transmits and a document that is true, is the whole subject of EDI for a CPG brand, and it runs through every document in the set. Seven X12 transactions carry a food brand's retail business from order to cash, and each one left a mark somewhere in Cinderhaven Provisions' ledger.

Seven documents carry one purchase order to cash

Cinderhaven is a fictional brand and its ledger is a synthetic dataset, built so that every deduction traces back to a specific document and a specific field. The brand runs EDI with six of its nine trading partners: Walmart, Costco, UNFI, KeHE, Whole Foods, and its largest regional chain. Across those six, one purchase order produces the following paper.

| Document | Who sends it | What it carries | Where Cinderhaven's ledger felt it | |---|---|---|---| | 850 Purchase Order | Retailer to brand | Order details from a buyer to a supplier: items, quantities, dates, ship-to | Units ordered are cases times the case pack in the item file; a wrong pack misprices the order before it is picked | | 855 PO Acknowledgment | Brand to retailer | The seller's acknowledgment of the buyer's purchase order: accepted, changed, or rejected lines | Required by KeHE for every PO; without it the order is unconfirmed on the distributor's side | | 856 Advance Ship Notice | Brand to retailer | The contents of a shipment and information relating to it: cartons, SSCCs, lot codes | $17,000 a year in ASN-quantity deductions, the second-largest family in the ledger | | 810 Invoice | Brand to retailer | Billing for goods and services provided, sent in response to the 850 | $140,000 in invoice deductions at Walmart, half of them winnable with the 810 itself | | 820 Remittance Advice | Retailer to brand | Payment and the remittance detail needed for cash application: what was paid, what was withheld | The deduction ledger's raw feed: 5,600 lines a year | | 812 Credit/Debit Adjustment | Retailer to brand | The itemized deduction with its reason code and reference | Sorts those 5,600 lines into four families worth $67,000 | | 852 Product Activity Data | Retailer to brand | Weekly units sold, on hand, on order, per store | $96,000 a year in dark-store weeks nobody read |

Under all seven runs the 997, which trading partners return for each document received and which confirms nothing but syntax. There is an eighth, the 846 inventory advice, a seller's inventory information to a prospective purchaser, which matters for direct-ship programs and for UNFI; it is a supply signal rather than a money document, and it is the one most brands can defer.

The right-hand column, read top to bottom, is the pattern: the same wrong value, entered once in the item file, is carried by the 850, restated on the 856, billed on the 810, deducted on the 820, and itemized on the 812. Five documents transmitted it flawlessly.

The documents do not disagree with each other. They agree with each other about a number that is wrong.

KeHE, Walmart, and UNFI each demand a different set

The set is not optional and it is not uniform. KeHE's supplier policy, effective November 2025, requires EDI for purchase order transmission and confirmation, names the 855 specifically, and takes invoices by 810 as its preferred method. Walmart's store-supplier set is the 850, 855, 856, and 810 over AS2. UNFI's core, per a vendor's 2026 guide, is the 850 or its grocery cousin the 875, the 855, 856, 810 or 880, 846, and 997, with UNFI sending the 820 and 824 back.

Two of those documents are grocery variants most guides skip: the 875 and 880 are the grocery-industry purchase order and invoice, structurally different from the 850 and 810, and a brand moving from a natural distributor to a conventional grocer can find its "working" EDI setup does not speak the dialect. The provider handles the translation. The brand still owns whether the case pack inside it is right, and every partner's list, whatever else it contains, includes the 856.

The 856 is where the money leaves

Vendors and brokers say the same thing about the ASN: late or missing 856s are "the single biggest chargeback trigger", KeHE's most common chargebacks hit suppliers for ASN non-compliance, and at Walmart the 856 is the most chargeback-prone document. The one research source that ranks the causes is more measured. The 2018 Credit Research Foundation survey of 203 companies lists EDI and ASN errors fourth among compliance deduction reasons by count, behind concealed shortages, transportation and routing, and early or late delivery. Both can be true. The ASN is checked against physical cartons at receiving, so its errors surface at the dock rather than in a later audit, which is where the vendors are looking and the survey was not.

The mechanism matters more than the ranking. An 856 is the brand's own sworn statement of what is on the truck: carton counts, SSCC labels, lot and batch codes, quantities per line. If the item file says 12 per case and the carton holds 6, the 856 will state a quantity the receiving scan will contradict, and the retailer will deduct the difference as a shortage and fine the mismatch as a compliance defect, on the same shipment. Walmart's requirement that traceability lot codes ride the ASN adds a third way for the same document to be wrong.

An ASN that transmits cleanly and states the wrong quantity is an item-file failure that EDI delivered on time. Four error types cause most ASN chargebacks, and all four are data fixes; the seven exception classes that cover most content mismatches all start upstream of the provider.

The provider proves the document left. The dock proves whether it was right.

Read the 997 for what it is, and reconcile the rest

The practical consequence for a brand of any size is that the 997 is the wrong document to trust and the only one most brands read. It answers one question: did the message parse. The questions that cost money, does the 856 quantity equal the 850 quantity, does the 810 price equal the 850 price, does the 820 payment equal the 810 total less agreed allowances, are answered by no document at all. They are answered by a reconciliation that joins the four money documents on the PO number, which the provider does not run and the retailer runs against you.

That reconciliation is also where the choice of provider stops mattering. SPS Commerce and TrueCommerce both deliver clean documents; neither checks whether the documents are true. The item file feeding them is the brand's, the case pack in it is the brand's, and every one of the seven documents above will carry it faithfully to a receiving dock that measures the carton.

Send me one PO's full paper trail

Pick one purchase order from your largest EDI partner and pull every document it generated: the 850, your 855 and 856, the 810, and the 820 line that paid it. I will lay the five side by side and write back with every field that changed value between documents, and which of those changes is a deduction on the remittance. Pull one PO's paper. The field that changes is almost always one the item file put there.