The Ten Touches Between a PO and a Release

A purchase order lands in an inbox at 7:04 in the morning. Between that moment and a truck leaving the dock, a flat-rolled service center will handle that order about ten separate times. It gets read, entered, matched to a master coil, planned onto a slitter, turned into a work order, cut and tagged, skidded, freighted and released, invoiced, and finally used to work out what to buy next. Most of those touches are a person retyping something a computer already knew. Two of them are genuine decisions, and those two set the margin on the order.

None of the ten steps is broken on its own. Each costs a few minutes, and once in a while one of them costs a coil.

The order comes in twice

Orders arrive every way imaginable: email attachments, PDFs, faxes that became PDFs, phone calls written on a pad, a text with a photo of a spec sheet, and EDI from the larger accounts. An EDI order is structured data. A phone order is one person's transcription of another person's memory. Both land in the same order book, and only one of them can be trusted without a second read. The quieter cost is latency. An order that sits in someone's inbox for four hours misses today's plan and lands in tomorrow's, which is how a two-week lead time slowly becomes three.

Then inside sales types it into the ERP, line by line, while someone checks the customer against a credit hold and someone else confirms the price against a quote or last month's number. This is the most re-keyed step in the building. Everything being typed already exists in the customer's PO, and every field is a chance to move a decimal. A 4.250-inch width keyed as 4.520 does not fail at entry. It fails three days later, on a coil that has already been slit.

Both steps are transcription. Reading a PDF or a photo and pre-filling the order is now reliable enough that entry can become a review screen (how AI order entry works, and where it fails). What stays with a person is the exception, like the price that does not match the quote or the customer over their limit.

Two decisions that set the margin

The first hard question in the flow is which master coil to pull. An order line carries a gauge, a grade or coating class, a finished width, a weight and a due date. A master coil carries all of that plus an ID and OD, a PIW, an edge condition and a location in the building. Matching them turns into a search, because the right coil for one order depends on what else can be cut out of it. A planner working coil by coil will find a coil that works. Pulling a prime 60-inch master to cut one 4.250-inch order looks fine on the work order and terrible on the month-end yield report.

The second is the slitting plan, and on most floors it belongs to the most experienced scheduler in the building, often the only one who can do it. This is where the scrap rate gets set, days before any metal moves, and where a block of that person's morning goes every day. The plan has to respect knife setups, min and max mults, width tolerances, PIW, edge conditioning, multiple passes, slitter capability, skidding rules, must-ship orders and lead times at once. A plan that misses one of them gets overruled on the floor. A scheduler can weigh dozens of arrangements before the shift starts, while the real number of possibilities on a full order book runs to millions.

In practice the two decisions are one. Coil and pattern have to be chosen as a pair, across the whole open order book and the whole inventory at once, and that search is the part software does far better than a person with a spreadsheet. The levers it works over are the ones in our guide to reducing scrap on a slitting line. Approval should stay with the scheduler. Every override tells the system something about the floor that was never in the data.

Paper, trucks and the invoice

Once the plan is approved, the planner usually translates it into work orders in the ERP, prints them and walks them out. That is pure transcription, and a second chance at the same mistakes as order entry. Outside processing is worse. An order to a toll processor tends to be an email with a spreadsheet attached, followed by a week of asking where it is.

On the floor, the operator runs the setup and tags each finished coil, and the shipping office pulls the mill test report for the master. Genealogy is the expensive one to get wrong. If a finished coil cannot be traced to its master there is no cert, and in some markets no shipment, so the parent-to-child link should be recorded when the coil is cut rather than reconstructed later from a work order and someone's memory. Skidding rules tend to live in a note field on the customer master, and a plan built without them produces a load that gets rebuilt on the dock.

Freight is where being busy costs money. A traffic coordinator, or in a smaller shop the same person who did everything above, emails three carriers, waits about an hour, compares the quotes and sends a release. When the day is running behind, the coordinator calls the carrier they always call. That is defensible under time pressure and expensive a hundred times a year. Collecting and comparing quotes is a small search on a deadline and one of the easier wins in the flow, though choosing when the cheapest carrier and the best one differ is still a judgment about service history.

After the truck leaves, someone chases proof of delivery and accounting invoices, usually in a weekly batch. An order that shipped Tuesday and was invoiced the following Monday is five days of working capital the service center financed for free. Meanwhile the buyer is working from a shortage list the planner produced as a byproduct of the plan. That list is accurate the moment the plan is approved and stale as soon as the next order lands, and buying from a stale list is how aging inventory starts.

Where the margin goes

Service center margins are thin by construction, since the business is buying metal, changing its shape and selling it. Ryerson, one of the largest service centers in North America, reported an 18.4% gross margin in Q1 2026, before warehousing, delivery, selling, general and administrative expense of $265.2 million that quarter.

Scrap takes the largest bite and is the most controllable, because every point of yield was paid for at mill prices. At an operation running tens of thousands of tons a year, one point of scrap is a six-figure sum in material, and it is worth working out with your own tonnage and your own cost per ton. Aging inventory comes next, material bought against a plan that changed and then carried until it sells at a discount. US service centers held 45.6 shipping days of flat-rolled sheet supply in July 2026, per Steel Market Update, and that figure is an average. Re-keying errors and freight overpay are smaller and steadier. None of them appears as a line item, which is why they last.

The ten steps sort themselves fairly cleanly. Where a person is moving information the building already has, as with order entry, work orders, certs and the bill of lading, software should do the moving. Where the answer needs more combinations than anyone can hold in their head, as with the coil, the pattern and the carrier, software should do the searching. Where a step is a commitment you would have to walk back in front of a customer, such as a price, a due date or a substitution, it stays with the people who own the relationship.

LineSight AI is built around that split. Its Command Center runs the order from PO to release as one connected flow, with a person approving the decisions that need one. Elite Steel added $850K in annualized revenue and cut its scrap rate 2.9 points after going live.