OCR isn’t enough
Extracting fields on its own doesn’t tell what’s right and what isn’t.
Use case 04
Invoices without manual
Invoices without manual
retyping and blind checking
The workflow doesn’t just extract fields from an invoice. It compares them against the order, the delivery note and internal rules - and hands the accountant mainly the discrepancies.
Where the problem starts
What slows the process down today
Five typical points that together steal time, quality and trust in the outputs. If you run into any of these, a workflow has somewhere to start.
Manual matching against the order
Accountants spend an hour a day clicking between systems.
Items outside the order
Price, quantity, discounts - mismatches are spotted by eye.
Rules in people’s heads
VAT limits, cost categories, approval matrices.
Manual export to the ERP
Copying data from one system into another.
Step by step in a single flow
The workflow doesn’t just read documents or just show numbers. It connects data, rules and decision points into a single flow with a clear output.
Invoice intake
Email, web upload, EDI, batch. Multi-invoice PDFs are split.
Header and line item extraction
Supplier, numbers, due dates, lines with prices and quantities.
Match against the order
Item by item, prices, quantities, VAT, delivery note.
Applying the rules
Limits, categories, approval matrices, threshold values.
Flagging exceptions
Whatever doesn’t match the order, a rule or the historical average.
Export to the ERP
A structured format, validated data, an audit trail.
What changes in operations
The concrete impact on the work
The goal isn’t to deploy AI. The goal is to give people back time to decide and free the process from its dependence on a few experts.
Accountants handle discrepancies, not routine
Clean invoices go straight into the ERP.
Fewer data errors
Validation against the order eliminates typos and duplicates.
Faster processing
From days to hours or minutes for standard cases.
An audit trail
For every invoice it’s traceable what was checked and what it passed.
When it makes sense to start
Three typical situations
A workflow doesn’t deliver the same value everywhere. Here are three scenarios where it pays off to start.
Hundreds of invoices a month
A high volume where manual processing eats up capacity.
An order / delivery note exists
The workflow needs data to compare against - without it, it’s just OCR.
Defined approval rules
Limits, roles, cost categories - rules that can be described.
What to watch out for
Limits we’re upfront about
Three risks we recommend addressing at the design stage - not later in operations, where they cost time and trust.
Corners and stamps
Scanned invoices with handwritten edits require more tolerance.
Suppliers without an order
For ad-hoc invoices there’s no reference data - the workflow flags them, but doesn’t auto-approve.
Non-standard items
Discounts, bonuses, returns - these need special rules or human validation.