Document AI & extraction
How can we automatically read invoices and documents into our systems?
The short answer
TEVEL (תבל) builds document AI workflows that read invoices, orders, forms and other documents, extract the relevant fields and turn them into structured data that flows into the ERP, CRM or another system. The data goes through validation and human approval at points defined up front, so manual data entry shrinks without giving up control.
The problem
Documents arrive by email, as scans or as photos, and someone types them into the system by hand. It's slow, repetitive and error-prone — and human hours go into a technical task instead of work that needs judgment.
What the workflow looks like
A typical document flow, adapted to the business's document types:
- Intake — from email, a folder, an upload or a photo
- OCR — recognizing the text in the document
- Classification — identifying the document type
- Extraction — pulling fields such as supplier, date, amounts and line items
- Validation — checking against rules and existing data in the system
- Human approval for exceptions or by policy (approval flow)
- Writing to the ERP or CRM via API and storing the original document
Control and accountability
AI-based extraction isn't perfect, so we plan up front what happens when the system isn't confident: the document is flagged for review rather than entered automatically. Every action is logged, permissions are defined, and extraction quality is tested on the business's real documents before usage expands.
More questions
How accurate is document reading?+
It depends on document quality, type and the variety of formats. That's why we test on a real sample and define when human approval is required — instead of promising a number.
Can it connect to the ERP we already use?+
In many cases yes, subject to the APIs, permissions and capabilities of that system.
Is it only for invoices?+
No. The same approach fits orders, delivery notes, forms, contracts and other documents whose data needs to reach a system.