From clerks' heads to thousands of documents a day.

An automotive parts manufacturer takes multi-customer quotes with per-OEM schemas, validated against different rules by major automotive OEMs. Every OEM structures their documents differently and validates them against different rules. What the plant needed was a platform that could standardize the work without standardizing the systems.

The plant had invested in OEM-specific portals and an ERP built around years of process work. Consolidation wasn't on the table: the ERP reflected too much institutional logic to rip out without stopping production.

Off-the-shelf extraction tools promised a fix, but every demo broke on the same reality: accurate on the page, blind to the plant, and wrong on every rule that actually decided the PO. The clerks knew the logic. The systems didn't.

Pedestal’s analyst sat down with the plant’s clerks and learned their document rules directly — per-OEM schemas, the validation logic, the edge cases that had never been written down. In parallel, it connected to the existing ERP and email channels, capturing each rule as a versioned entry in Pedestal Atlas.

Once Atlas reflected how the plant actually worked, the Document Intelligence agent came online on top of it: ingest inbound documents, extract against the right customer schema, validate against the plant’s rules in Atlas, write straight into the ERP. The first few weeks were a tight loop with the clerks — the agent would flag anything it wasn’t sure about, a clerk would resolve it, and the resolution became a new rule in Atlas that every future run inherited. Accuracy crossed the bar set by the senior clerks, and the loop went quiet.

Today, the Document Intelligence agent runs thousands of inbound documents a day through the plant on Atlas the analyst built with the clerks. Business rules that had lived in clerks’ heads now live as versioned, attributed entries in Atlas. Onboarding a new OEM customer is a config change, not a project.

Atlas opens the door to the next set of agents — procurement exceptions, supplier performance, claims — each inheriting the document context already captured and adding to it. What began as a document-extraction project is becoming the knowledge layer the plant runs on.