Almadar A finance team near Olaya processing invoices by hand is doing exactly the high-volume, rules-bound work that a KAFD-based fintech long ago automated.
The process being replaced is familiar: invoices arrive by email or paper, someone reads them, keys the header and line data into the ERP, matches to a purchase order and receipt, routes for approval and files the document. Automation handles extraction, matching and routing, leaving people to handle exceptions and judgment.
How extraction actually works
Modern extraction handles varied layouts without templates, reading supplier name, VAT registration number, invoice number, dates, line items, VAT amounts and totals from PDFs, scans and images in Arabic and English. Confidence scoring routes uncertain fields to human review rather than guessing, which is what makes the output trustworthy enough to post.
Matching and exception handling
Extracted data is matched three-way against purchase order and goods receipt. Clean matches post automatically. Exceptions, price variance, quantity difference, missing receipt, no purchase order, route to the right person with the discrepancy highlighted. Most of the value is here rather than in extraction, because exception routing is what actually compresses the cycle time.
Saudi-specific requirements
Extraction must capture the buyer and seller VAT registration numbers and validate their format, handle Arabic invoice layouts, and reconcile against ZATCA-cleared invoice data where the supplier is within Phase 2 scope. Where a supplier's invoice does not match what was cleared with ZATCA, that is a discrepancy worth flagging rather than posting.
A common Saudi scenario
A Riyadh distributor processes roughly two thousand supplier invoices monthly across three staff, with an error rate that produces regular duplicate payments and supplier disputes. Automation handles extraction and three-way matching, with around seventy percent posting without intervention. The team shifts to exception handling and supplier query resolution, and the duplicate payment problem largely disappears because matching is systematic rather than visual.
Measuring it honestly
We baseline before deployment: invoices per month, average processing time, error rate, duplicate payments, early payment discounts missed through slow approval. Measuring these after gives a real return figure rather than an assumed one, and it connects to broader process improvement, since automation applied to a badly designed approval workflow simply makes a poor process faster.
Approval workflow and payment timing
Extraction and matching only compress part of the cycle. If an approved invoice then waits days for a payment run, the cash and supplier relationship benefit is lost. We look at the whole path from receipt to payment, including approval routing, payment run frequency and early settlement discount capture, because the constraint is frequently downstream of the automation. This connects to procurement automation upstream and to cash forecasting, since accurate payables timing materially improves forecast quality.
High-volume distribution and trading businesses in Riyadh see the fastest payback given invoice volumes, while project-based businesses with fewer but more complex invoices gain more from exception routing than from extraction speed.