Use case · Intelligent Document Agent, scoped to quoting"Which of these tables is the bill of materials?"

Bill-of-Materials Extraction

Quote requests arrive as spec sheets in each customer's format, and the bill of materials inside them has to reach your own quoting template before anything can be priced. The document agent finds the BOM, reads it cell by cell, learns each customer's format once, and exports - with a person reviewing every step that matters.

The problem

Why generic extraction fails on spec sheets.

A spec sheet is not one table - it is a page of them. Revision histories, title blocks, notes, and somewhere among them the bill of materials. The PDFs are often scans with no text layer, so there is nothing to copy. And every customer lays the same information out differently, under different column names.

The standard trick - screenshot the table, ask a model to extract it - produces confident output with swapped columns and merged rows, and nothing flags the error. So the work stays manual: someone opens each file, finds the BOM, and re-types it into the quoting template, cell by cell, for every file in every enquiry.

Read the demand entry on the AI ROI Map
The approach

Find the table. Read it cell by cell. Land it in your template.

The agent scans the document for every table, then looks for the one that is semantically a bill of materials - and when two candidates look alike, it shows both and lets a person choose rather than guessing silently. The chosen table is read from the document's own geometry: each cell located, then transcribed individually, which is where the accuracy comes from. Column names are mapped onto your fields as a reviewed profile, and the result exports into your quoting template.

The parsing engine underneath is the same one that runs in production foranalytical lab reports - this is that engine, scoped to quote requests. Not a third product: the same document agent, pointed at a different document estate.

Spec sheetBOM tableCellYour template

Each step is inspectable. The value in your template can always be traced back to the cell it came from.

The review flow

A person stays in charge of what ships.

Extraction is automatic; judgment is not. This is the flow as it runs in the demo environment - the human decides what counts, corrects what is off, and owns the export.

  • Upload the enquiry - the PDFs and spreadsheets exactly as they arrived.
  • Review every table flagged as a possible bill of materials.
  • Mark the one that counts - or several, when the spec sheet splits it.
  • Check the extracted line items beside their source cells; edit, add or delete rows.
  • Map their column names to your fields once - it holds for every future file in this format.
  • Export into your quoting template.
Can this read our spec packs and fill our template?
What it unlocks

Quoting at the speed of judgment, not typing.

Who it's for

Built for the people who turn enquiries into quotes.

Teams

  • Quoting and proposal teams
  • Application engineering
  • Sales operations
  • IT, as the control owner

Where it runs

Entirely inside your infrastructure. The spec sheets are your customers' confidential designs - they never leave your boundary, and neither does your pricing logic.

See it on your own spec pack.

Bring one enquiry - the files exactly as they arrived - and watch the line items land in your template. Or run the public exercise first and try to prove us wrong. No form, no gate.

On-prem. Your data never leaves your boundary.