Document intelligence
Documents read, checked and posted, with people on the exceptions.
Document intelligence reads the invoices, contracts, applications, IDs and delivery notes your teams re-key today — scanned or digital, Arabic or English. It classifies each document, extracts the fields you need, checks them against your records and posts clean results to your ERP. Anything it is unsure of goes to a person, with the source on the page highlighted.
The problem
Where manual processing costs you
- 01Staff re-key the same fields from PDFs and scans into SAP or Oracle, and mistakes surface weeks later in reconciliation.
- 02Documents arrive by e-mail, portal and scanner, in mixed layouts, languages and quality.
- 03Template-based OCR breaks whenever a supplier changes its invoice layout.
- 04When an auditor asks where a figure came from, someone has to dig through folders for the original page.
How it is built
How the pipeline is built
A pipeline, not a single model. Each stage can be tuned or replaced on its own, and every extracted value keeps a link to the page and position it came from.
Layer 1
Inputs and people
- E-mail inboxes and portals
- Scanners and mobile capture
- Review queue for staff
Layer 2
Reading
- Classification and splitting
- OCR and layout analysis
- Field and table extraction
- Confidence scoring
Layer 3
Validation and posting
- Checks against master data
- PO, GRN and invoice matching
- Business rules and duplicate checks
- Posting through ERP interfaces
Layer 4
Infrastructure
- Private model serving
- Document and audit store
- Queue and workflow engine
- Monitoring
Capabilities
Document intelligence
Classification and splitting
A scanned batch is split into individual documents, and each is classified — invoice, credit note, contract, ID, delivery note — before extraction starts. Unknown types go to a person instead of being forced into the wrong template.
Extraction without fixed templates
Layout-aware models read fields and line-item tables from layouts they have not seen before, in Arabic, English or both on one page. Hijri and Gregorian dates, Arabic-Indic digits and VAT numbers are normalised to the formats your systems expect.
Validation against your records
Extracted values are checked against supplier master data, open purchase orders and goods receipts, and three-way matching of PO, GRN and invoice catches quantity and price differences before anything posts. QR codes on ZATCA e-invoices are read and compared with the printed totals.
Confidence thresholds and review
Each field carries a confidence score. Above the threshold you set, it passes; below it, the document goes to a review queue where staff see the page with the field highlighted and correct it in place.
Posting to your ERP
Approved documents post to SAP S/4HANA, Oracle or Microsoft Dynamics 365 through their standard interfaces — as parked or draft entries if you want a final human check.
A field-level audit trail
Every value records the page and position it came from, the model version that read it, the rules that checked it and anyone who changed it. An auditor can go from a posted entry back to the original page in one step.
How it works
From inbox to ledger
- Document received
- Classified and split
- Fields extracted
- Checked against records
- Reviewed if uncertain, posted
Connects to
Works with
- SAP S/4HANA and SAP ECC
- Oracle E-Business Suite and Oracle Fusion
- Microsoft Dynamics 365
- Microsoft 365 and Exchange mailboxes
- SharePoint and document management systems
- Supplier portals and SFTP drops
- Network scanners and multifunction printers
- QR codes on ZATCA e-invoices
- Workflow and BPM platforms
- Data warehouses for reporting
Where it runs
On-premises
Models and data run on hardware in your own facilities. Nothing leaves your network unless you allow it.
Private cloud in the Kingdom
Dedicated, isolated capacity in a Saudi data-centre region of a provider you select — the elasticity of cloud without a shared public AI service.
Hybrid
Sensitive workloads stay on-premises; less sensitive ones use private cloud capacity. You set the policy for what goes where.
What we measure
Measures we agree with you
We agree these measures with you during discovery, against a baseline of today’s process, and report against them once the system is live. They are targets we track together, not results we promise in advance.
01
Straight-through rate
Share of documents posted without a person touching them, tracked by document type and supplier.
02
Field accuracy
Share of extracted fields that are correct on an audited sample, reported per field so weak spots are visible.
03
Cycle time
Time from a document arriving to its being posted or rejected, against the manual baseline measured in discovery.
04
Exceptions caught before posting
Duplicates, price and quantity mismatches and missing approvals flagged before posting, compared with what reconciliation used to find afterwards.
Indicative delivery path
Indicative delivery path
- 01
2–3 weeks
Sample and baseline
We collect a representative sample of each document type, label the fields that matter and measure today’s effort, error rate and cycle time.
- 02
4–8 weeks
Build and tune
Classification, extraction and validation rules are built and tuned against the labelled sample, and confidence thresholds are set with your finance or operations owners.
- 03
3–4 weeks
Parallel run
The system runs alongside the manual process. Staff compare results, and we adjust thresholds and rules until the agreed accuracy is reached.
- 04
Ongoing
Go live and extend
Posting is switched on one document type at a time, and further types or entities are added once the first ones are stable.
Durations are indicative and depend on scope, data and integrations.
Questions
Document intelligence
How accurate is the extraction?
It varies by document type, scan quality and field, which is why we measure it on your own sample before agreeing any target. Clean digital invoices from regular suppliers behave very differently from faded handwritten forms. Confidence thresholds mean uncertain fields go to a person rather than straight into your ERP.
What happens when a supplier changes its invoice layout?
Extraction does not rely on fixed templates, so a new layout does not break the pipeline the way it breaks template-based OCR. If accuracy drops for a supplier, monitoring shows it, and corrections made in the review queue feed the next round of tuning.
Can it process national IDs and other personal documents?
Yes, with extra care. Personal documents are processed inside your environment, access to the review queue is limited by role, and retention follows the rules you set. This supports your obligations under the Personal Data Protection Law, alongside your legal and compliance teams; it does not by itself make you compliant.
Will it post to our ERP automatically?
Only when you decide. We usually start by posting documents as parked or draft entries for a final check, then allow direct posting for the document types and suppliers that have proven reliable. The thresholds and rules stay under your control.
Other solutions
All solutionsPrivate AI platform
Open-weight and commercial models served inside your boundary, behind one governed gateway.
View solutionKnowledge assistant
Cited answers from your documents and systems, in Arabic or English — limited to what each user may open.
View solutionCustomer service agent
Arabic and English service on web chat, WhatsApp and voice — approved answers, lookups, hand-over to people.
View solutionAgentic operations
AI agents that handle procurement, invoice, HR and IT requests end to end — within limits you set.
View solution
Have a process that should run itself?
Tell us about it. We will help you work out whether AI is the right answer, what it would take, and what it should return.

