EDI 852: The Weekly Velocity Data Retailers Already Send You

Fifty-two files a year. Each one lists, store by store, how many units of each Cinderhaven Provisions SKU sold last week and how many sit on hand, and nobody at the brand opens any of them. Inside this year's files: one SKU dark at 40 stores of a regional chain for six straight weeks, at $960 a week, part of $96,000 a year the unread file would have caught in week one.

Cinderhaven is a brand invented for this purpose and its weekly numbers are a synthetic dataset, but the file is the EDI 852 and every property of it described here is real. For the half of a specialty food business that has no Retail Link login, the 852 is the store-level view. Its value is measured in detection weeks.

The file already lists the zero

The EDI 852, Product Activity Data, is the transaction retailers and distributors send suppliers to report what the product did: quantity sold, quantity on hand, and quantity on order, by item and by location, weekly or daily. It is the same species of data a Walmart supplier pulls from Retail Link, delivered as a feed instead of a portal. Kroger publishes its own 852 map for suppliers, and mid-market grocers and drug chains do the same. Walmart handles the job differently, pushing store-level POS through Retail Link, which covers Walmart and nothing else.

That difference matters at Cinderhaven's scale, where half of $25M in wholesale moves outside Walmart: UNFI and KeHE trucks, Whole Foods, Costco, and regional chains. None of those channels comes with a free store-level portal. Store-level truth arrives as an 852 in the brand's EDI mailbox, or it does not arrive at all.

Most brands receive the file and archive it. The EDI provider confirms receipt, the inbox moves on, and the one document that says which shelves are empty this week goes unread, because reading it requires a parser nobody has built.

Everything downstream is a slower copy of this file.

A dark store bleeds by the week, and the rollup hides it

What a dark store costs is short arithmetic. One SKU authorized at 40 stores, moving 3 units per store per week at an $8 wholesale price: 40 × 3 × $8 = $960 a week in sales that stop when a shelf reset drops the item and replenishment never restarts.

The shopper does not wait out the fix. In the European out-of-stock data the ECR Retail Loss group compiles from the Corsten and Gruen research, 32% of shoppers facing an empty facing buy a different brand and 27% buy the item at another store, while 17% delay the purchase and 9% buy nothing; only 16% substitute within the same brand. That last figure is the one to read twice. Roughly one shopper in six stays with the brand. A dark week is not deferred revenue, and the largest single share of it walks a repeat buyer onto a competitor's label.

What keeps the store dark is the shape of the data that does get read. The distributor's monthly report rolls 40 zeros into a chain-level total, where the SKU still shows movement because 200 other stores are selling. The zero survives every summary it passes through. It finally surfaces when the velocity number a buyer uses at the delisting review has already absorbed six weeks of suppressed scans.

The chain average looked fine. Forty stores were zero.

Detection cadence is the entire cost

Read weekly, the 852 shows the incident in its first file: 40 stores, zero sold, zero on hand. Read never, the incident waits for a human to notice a soft month. The same event, priced at three detection cadences:

| How the zero gets found | Week found | Dark weeks | Cost per event | |---|---|---|---| | Weekly 852 scan | 1 | 1 | $960 | | Monthly distributor report | 6 | 6 | $5,760 | | Quarterly line review | 13 | 13 | $12,480 |

Cinderhaven's ledger logs 20 dark-item events a year across its non-Walmart channels: resets that dropped the item, replenishment loops that never restarted, on-hands frozen over a back-room pallet. Caught at week 6 instead of week 1, each event costs an extra $4,800, and twenty of them cost $96,000 a year in revenue that never had to be lost. The ledger is synthetic; the cadence math works the same on anyone's.

The on-hand column earns its own paragraph. Zero sold with zero on hand is a replenishment problem. Zero sold with 12 on hand, frozen for four weeks, is a ghost: product in a back room or on a mis-set shelf, scanning as available, selling as nothing. Scan-only views cannot tell the two apart. The 852 carries both columns, and it is the earliest place to interrupt the loop where a stockout teaches the forecast to order less, before the bad weeks enter the training history.

Week one and week six hold the same information, five dark weeks apart.

The parsing cost is real, small, and paid once

The honest objection to the 852 is formatting drift, and it is not a small objection. Every trading partner maps the transaction differently, and the X12 standard invites it, defining the 852 for distribution centers, warehouses, and retail outlets alike. Kroger's published map runs daily off a distribution center and reports availability codes rather than units sold at the shelf; another partner sends store-level detail; one reports units where the next reports cases; on-hand accuracy carries known caveats everywhere. Normalizing three partners' files into one weekly exception table, in NetSuite, a database, or the spreadsheet layer many brands this size run on, is an analyst-week to build and an hour a week to run.

Price both sides. At the same $50 an hour, the build is about $2,000 and the running cost about $2,600 a year, against $96,000 of avoidable dark-store revenue. Now price the alternative source. Syndicated data is sold in tiers: a single category in a single panel runs a few thousand dollars a year, a working multi-category build lands in the low five figures, and store-level access is a paid upgrade with every provider. Buy the upgrade and the panel still will not report the frozen on-hand at store 4417, because that store's chain may not sit in the panel at all. The 852 is already flowing into the EDI mailbox the brand pays for. The step it needs is the one Lailara's void finder performs on scan files: rank every dark store-SKU by the weekly revenue it suppresses, so the worst zero gets the first phone call.

Store-level data is a paid upgrade everywhere except your own mailbox.

Send me four weeks of 852s

Go into the EDI mailbox, take the last four weeks of 852 files from any retailer or distributor sending them, and forward them unedited. Back comes the dark rows, the frozen on-hands, and a weekly dollar figure on each one, ranked. Pull them straight from the mailbox. Nobody needs to build the parser to find out whether it would pay.