Invoices and POs
Straight into the ERP
Line items, totals and PO numbers land as fields your ERP can post.
Product of
Turn paper into data.
Any document, any length, into structured data your systems can use.
93.2OmniDocBench v1.5
6 to 15xCheaper than basic cloud OCR
On-premiseRuns on your own servers
InvoicesPurchase ordersContractsCertificates of analysisDelivery notesBills of ladingScanned faxesPhone photos of receiptsMulti-column reportsFormulas
P. 02 The problem
Your data is in here. Somewhere.
Invoices, contracts, certificates and reports hold the numbers your business runs on. Most of that data is stuck inside PDFs, scans and photos.
80%
of enterprise data is unstructured.Source: Gartner
Read 0 of 96 documents
P. 03 The film
Fifty seconds, from paper to data.
Invoices, contracts, scans and phone photos. Watch Sift-OCR read them page by page and turn them into structured data your systems can use.
P. 04 Read anything
Any document. Any length. Tables and formulas intact.
PDFs, scans and phone photos, including tables, formulas and multi-column pages. Sift-OCR keeps the layout and reads everything in the right order.
Northfield Foods Accounts Payable
Running on-premise
Markdown
Key and value
Ready
Sample documents are fictional.
P. 05 No page limit
No page limit.
One-page invoices to 500-page reports, processed page by page, in order.
Page
Reading in order
Pages are counted from 1 to 500.
of 500
InvoiceInvoice INV-32305 · Corvane Cold Chain
Rows 0
Markdown · JSON · CSV
P. 06 Data you can use
Clean, structured data. Ready for your systems.
Out comes clean text, tables and layout as Markdown, JSON or CSV, ready for your ERP, your search and your AI.
Line items
Waiting for the pageFields
- ERPFields and line items
- SearchFull text and layout
- AI assistantClean Markdown
Supplier certificates
Checked against specs
Each result on a certificate comes out as data, ready to check against your limits.
Contracts
Dates, terms and obligations
Pull the dates, terms and obligations out of agreements of any length.
Archives
Paper made searchable
Scans and faxes from the filing room become text you can search.
P. 07 Accuracy
More accurate than leading AI models.
On OmniDocBench v1.5, a published benchmark for document parsing, Sift-OCR scores higher than Gemini 2.5 Pro and GPT-4o.
Sift-OCR · overall score
93.2
- +5.2 vs Gemini 2.5 Pro
- +18.2 vs GPT-4o
OmniDocBench v1.5 · document parsing
Overall score, higher is better
- Sift-OCR 93.2
- Gemini 2.5 Pro 88.0
- GPT-4o 75.0
P. 08 Cost
At a fraction of the cost.
Self-hosted on your own servers, Sift-OCR reads the same pages for a fraction of the cost of cloud OCR and cloud parsers.
Cost compared with cloud services
Self-hosted
cheaper than basic cloud OCR
cheaper than cloud layout and table parsers
About $0.25 per 1,000 pages, self-hosted.
Estimate your monthly cost
Sift-OCR · self-hosted
about
$250
per month
1,000,000 pages ÷ 1,000 × $0.25 = $250
Estimate of self-hosted processing at about $0.25 per 1,000 pages.
Same budget, more pages
Basic cloud OCR1 page
Sift-OCR6 to 15 pages
For the same spend, basic cloud OCR reads 1 page while Sift-OCR reads 6 to 15 pages.
Pages read for the same spend, from the multiples above.
P. 09 Private by design
Your documents never leave your network.
Sift-OCR runs on your own GPU server, alongside LoopLM, Aether AI's private on-premise LLM. Documents come in from scanners, email and shared drives, and leave only as structured data for your internal systems.
- Runs on your own GPU server
- Works alongside LoopLM
- Documents never leave your network
P. 10 Paper to data
From paper, to data.
Every sample document on this page, read and turned into structured data your systems can use.
P. 11 Get started
Turn paper into data.
Talk to Aether AI about putting Sift-OCR to work on your documents, on your own servers.