The front door of the floor.
A customer PO lands as a PDF. Bulk reads it, matches it to your parts and routings, and turns it into a production-ready order with jobs — no re-keying between intake, production, QC, reporting and invoicing.
What is Bulk Inbound & OCR?
Bulk Inbound & OCR is the front door of the floor. A customer PO, schedule, or delivery note lands as a PDF, scan, or spreadsheet; Bulk reads every line, matches it to your parts and routings, and — once a person approves it — turns it into a production-ready order with jobs. No re-keying between intake, production, QC, reporting, and invoicing.
Paper in. Jobs out.
Every document takes the same three steps: Bulk reads it, a person confirms it, and approved lines land on the floor as numbered jobs — nothing skipped, nothing retyped, and the original file kept forever.
Every line, captured
Drop in a PDF, scan or spreadsheet and Bulk reads every field and line item — even messy tables. The original stays untouched, so you can always see exactly what the paper said.
A person signs off
Nothing becomes work by itself. Anything Bulk wasn't sure of is flagged for a quick fix — you correct, confirm and approve. You stay in control of every order.
Approved becomes jobs
One approval turns ready lines into numbered jobs on a production order — on the floor's screens the same minute, with nobody retyping a thing.
Reads every line. Matches every part.
Drop in a customer's PO, schedule or delivery note and Bulk reads it end to end — then matches every line to your own parts, routings and customers. What comes out is ready to become real work on the floor.
The document arrives
Any PO, schedule or delivery note — PDF, scan or spreadsheet. Drag it in and you're done.
Bulk reads it
Every field and line item is captured — even messy tables and unfamiliar layouts — and anything unclear is flagged, not guessed.
Matched to your data
Each line is matched to your own parts, routings and customers — ready to approve and send straight to the floor.
Anything Bulk isn't completely sure about is held back for a person to check — never pushed through silently. You stay in control of every order.
Onboard a supplier format in minutes.
Admins teach a config how to read one customer's documents through a five-step wizard — no engineering. Progress saves per config so you can resume, and every save snapshots a version you can roll back to.
Upload sample
Drop a representative document; the AI analyses it and detects fields.
Extracted data
Review detected fields, pick what to extract, mark the line-item driver.
Map document fields
Map to PO number, customer, dates, title and notes — with transforms.
Map line items
Map part number, routing, quantity, price and dates, with master-data lookups.
Preview
Live extraction against the sample; on save, mappings compile into rules.
Built for messy, real-world paper.
The things that make intake actually survive a Tuesday: ambiguous part numbers, duplicate emails, a flaky AI provider, and a format that changes the week after you onboard it.
Master-data matching
Extracted text resolves to real parts, processes, routings and customers by exact, fuzzy, manual or default method — each carrying a confidence score.
Driver grouping
A nominated line-item driver field groups raw rows into the right number of line items, with per-field aggregation across the group.
Duplicate detection
Every file is hashed (SHA-256) on upload and compared to prior files, flagging same-session and previously-uploaded duplicates before they double up.
Field-mapping pipeline
One source field can hit many columns, each with its own transform, pattern extraction, lookup, calculation and date-manipulation chain.
Automation levels
Run manual review, or set a config to auto-create jobs the moment valid items land — gated by the required fields you choose per customer.
Circuit breaker & retries
Retryable failures back off and retry on a schedule; an AI-health circuit breaker trips on repeated errors so a bad hour doesn't poison the queue.
Extraction proposes. People approve.
The AI never writes work straight to the floor. Its output is a proposal in an immutable layer; a human is always the final gate before anything becomes a job.
What the model does
Reads the document, extracts tables and fields, resolves them against your master data, and records confidence on every match — then stops.
- Raw extraction kept immutable
- Low-confidence matches flagged “needs review”
- Promotes into an editable document, then locks
What the reviewer does
Works the document side-by-side with the source PDF — corrects items, fills detail, marks ready, then submits. Submission is all-or-nothing: every item must be able to create a job.
- Edit, duplicate, mass-update line items
- Submit · approve · reject, gated by permission
- Approve creates one job per ready item
Track the batch. Process by scan.
A session tracks every file across the Upload → Inbound → Production pipeline; the scan kiosk takes a single item end to end on a shop-floor terminal.
See it run
Watch a customer PO become a production job and a printed label — in a single pass.
- Customer—Northwind Mfg.99%
- PO number—PO-4821399%
- Part—BRK-2207-A97%
- Routing—CNC-20495%
- Quantity—1,200 pcs98%
- Due date—Jul 30, 202696%
- Production order createdORD-48213
- Job createdJ-1042 · 1,200 pcs
- Material reservedBRK-2207-A · 1,200 pcs
- Routing scheduledCNC-204 · 3 ops
One item, start to finish.
Follow the document as it moves from intake to a finished job on the floor.
- 1 Read the documentThe AI reads the PO and pulls every field — part, routing, quantity, dates.
- 2 Review & approveA person checks the extracted values, then submits the order to production.
- 3 Create the orderThe production order, job and material reservation are created automatically.
- 4 Print the job labelA floor-ready label is generated and sent straight to the line printer.
Paperwork that types itself.
or spreadsheet
you finish
your data
region
Nothing gets retyped downstream.
Capture an order once and it moves through the whole platform on its own — no one re-types it anywhere.
Master Data
Every order lands matched to the right parts, processes and customer — automatically.
Production
Approve an order and it drops straight onto the floor, ready for the team to run.
Quality & Forms
The right checks travel with every job, so nothing moves on until it's complete.
Pricing
Every line is priced as it's captured — a clear total you can stand behind before work starts.
Exports
Send any order on to your other systems in a single click.
Search
Find any document or order in seconds, however you remember it.
Questions, answered.
What is Bulk Inbound & OCR?
It's the intake layer of Bulk: a customer's PO, schedule, or delivery note is read end to end, matched to your own parts, routings, and customers, and turned into a production-ready order with jobs.
Does the AI write orders straight to the floor?
No. Extraction proposes; people approve. The AI reads the document and records confidence on every match, then stops — a human is always the final gate before anything becomes a job.
What kinds of documents can it read?
Any PO, schedule, or delivery note, whether it arrives as a PDF, a scan, a photo of a paper order, or a spreadsheet. Bulk reads every field and line item, even messy tables and unfamiliar layouts.
How does it match text to my own parts?
Extracted text resolves to your real parts, processes, routings, and customers by exact, fuzzy, manual, or default method, and every match carries a confidence score so low-confidence ones are flagged for review.
Do I need engineering to onboard a new supplier format?
No. Admins teach a config how to read one customer’s documents through a five-step no-code wizard. Progress saves per config, and every save snapshots a version you can roll back to.
Where are my documents processed?
Your documents are processed inside your own region and never leave it, so your data stays your data.
Send us your worst delivery note.
We'll build a working extraction config against your real documents and show you the order it produces — then you decide.