Production order intake and validation for co-packers, built custom (PackSpec)

A food and beverage co-packer runs production for dozens of CPG brands, and every brand sends its purchase orders, specs, and forecasts in a different format: PDFs, emailed spreadsheets, Word docs, portal exports, and the occasional scanned page. Someone re-keys all of it into the ERP by hand, then checks it against what they know, whether the SKU is still active, whether the spec matches the current formula, whether the quantity fits agreed capacity, whether the dates are feasible.

That manual pass is slow and error prone, and the mistakes are expensive. A wrong case pack or a misread quantity turns into a mis-run, a short ship, or in the worst case a recall. The ERP holds the clean order, but nothing reads the messy inbound document. That gap is where the errors live.

PackSpec reads any incoming order or spec, whatever the format, and turns it into a structured production order. It extracts every field it can find, marks the confidence on each one, flags what is missing instead of guessing, and runs a second pass that surfaces conflicts and gaps before the order reaches the floor.

• •Extraction from PDF, image, Word, spreadsheet, or pasted text, including scans and photos
• •A structured production order with a confidence indicator on every extracted field
• •Missing fields marked clearly rather than filled with a guess, and every value editable inline
• •A conflict and gap pass that flags internal contradictions, implausible dates, and quantities that do not add up, each with a severity
• •Clean export as JSON or CSV, ready to feed the ERP

An order arrives in whatever format the brand uses. The document goes to the model, which returns the fields it can find and a confidence level for each. Anything it cannot find is left blank and marked missing, so no one mistakes a guess for a fact.

A second review pass reads the extracted order back and looks for problems it can prove from the document itself: required fields that are absent, a quantity that does not match case pack times case count, dates in the past or out of sequence, a line item an order of magnitude off from the rest, or a spec that is referenced but not attached. It flags these with a plain explanation and a severity, and stays out of business rules that would need a specific co-packer’s master data.

Inbound orders become clean, checked production records instead of a re-keying chore.

Contradictions and missing information surface before a run is scheduled, not after.

The order that reaches the ERP is one a planner has reviewed, not one they typed from a PDF.