BOM Automation

The Problem

A misread character doesn't announce itself. A '1' that looks like an 'I.' An 'O' that looks like a '0.' It doesn't stop the work. It just becomes the wrong part on an order, weeks later, at the loading dock.

For one switchgear and panel building firm, that was the whole quality system: one engineer's eye, reading a dense drawing, once.

What We Understood First

Watched the Real Workflow

Multi-page PDFs, scans, photos: we mapped where mistakes crept in before building anything.

Three Reads, Not One

Every drawing read three times, independently. Only agreement gets confirmed.

OCR Cross-Check

Each reference located on the drawing itself. A line either checks out or gets flagged.

Sorted Like the Trade

Components sort into HVAC, electrical, plumbing, and fire protection, the way estimators already think.

Nothing Replaced

Fit into the tools already in use. Manual analysis dropped 90%.

Independent reads per drawing

Trade categories sorted automatically

“Right first. Then fast: 70% faster, in the end.”

Switchgear and panel building firm

What We Didn't Automate

The three reads settle what a line says. They don't settle what to do about a line no one's sure of, that call stays with a person.

Every line comes back marked Verified or Review, so attention goes to the handful that actually need a human eye.

Tensions Worth Naming

Inconsistent formats.

No two drawings looked the same: different scales, hand-annotated scans of varying quality. We read the raw drawing without pre-cleaning that could lose detail. Now it reads directly, whatever shape it arrives in.

Trust in judgment.

Engineers who'd built their reputation on catching errors by eye were wary of software doing that work instead. We positioned it as a first pass, not a replacement. Adoption grew once they saw it catching what they'd have caught.

If You're Facing This Too

If your team is still re-typing parts off drawings by hand, the real bottleneck is usually format, not typing speed: no two drawings are laid out the same, and one misread reference becomes a wrong quote.

Whatever you automate, don't trust a single pass to catch its own mistakes. Three reads catch what one confident guess won't.

What's Next

More drawing types beyond the first set, and further down the pipeline, straight into catalog matching and pricing.

None of this needs more estimators as volume grows, the same three reads scale to more drawings without more people reading them. And most tools stop at OCR; catching a '1' that looks like an 'I' before it becomes a wrong order is what actually sets this apart, not just faster typing.

IN PERSPECTIVE

This platform needed something sturdier than a faster reader: confidence that a fast BOM was still a right one, before anyone downstream had to trust it.