Skip to main content
010 447 2445Contact
Arke Technologies home
Arke Technologies home
An administrator reads a document at his desk beside a large monitor.

Document intelligence and OCR software

Document intelligence is OCR software plus AI: it reads scanned and digital documents, extracts the data you need, checks it and sends it to the right system or person. Arke Technologies, in Johannesburg, builds document processing AI for organisations across South Africa, with human review, audit trails and POPIA controls.

OCR software, intelligent document processing and document AI

OCR (optical character recognition) software turns an image of text, such as a scan, a photo or a fax, into text a computer can search and copy. On its own it does not know which number is the invoice total or whether a supplier's VAT number is valid. Document processing AI adds that understanding: it classifies the document, finds the fields and tables you need, checks them and passes them on.

OCR softwareTemplate-based captureAI document processing
What it doesConverts images of text into searchable textReads fixed fields from layouts set up in advanceClassifies documents, extracts fields and tables from varied layouts, validates and routes them
Copes withClean printed textOne layout per templateMany suppliers' layouts, tables, stamps and, with lower accuracy, handwriting
OutputA searchable PDF or plain textField valuesStructured data with a confidence score for each field, plus a workflow
People stillRead and key the dataMaintain templates as layouts changeReview the exceptions the system flags

Most organisations need the third column for their busiest documents and plain OCR for the archive. Document intelligence is part of our AI consulting and implementation practice.

How AI document automation works, step by step

  1. Capture

    Documents arrive from scanners, shared mailboxes, uploads and portals.

  2. Classify

    The system decides what each document is: an invoice, statement, application, contract, identity document or certificate.

  3. Extract

    OCR and language models that read images pull out the fields and tables you need.

  4. Validate

    Values are checked against your records: is the supplier on the vendor master, does the VAT number match, do the line totals add up, is there a matching purchase order?

  5. Review

    Fields below the agreed confidence level go to a person, who confirms or corrects them on screen.

  6. Route and file

    Clean data posts to your accounting or ERP system, and the document is filed with its metadata in SharePoint or Google Drive.

  7. Record

    Every document, extraction, correction and approval is logged, so an auditor can trace any figure back to its page.

Corrections improve the rules and instructions we give the system; we never train models on your documents without your explicit consent. When document steps are part of a longer process, our AI automation service designs the workflow from end to end.

Two colleagues discuss a trend line on a large wall screen.

Where document processing AI pays back

  • Accounts payable

    Supplier invoices, statements and credit notes captured, matched to purchase orders and posted to Sage, SAP Business One, Microsoft Dynamics 365 Business Central or Xero through our ERP and accounting integration work.

  • Supplier onboarding and procurement

    Company registration documents, B-BBEE certificates and affidavits, bank confirmation letters and tax compliance status checked for completeness and expiry dates before a supplier is approved.

  • Municipal revenue

    Paper account applications, meter records and valuation documents captured, so that billing, metering and property-valuation records can be reconciled and exceptions reported for follow-up, as part of municipal revenue enhancement.

  • HR and payroll

    Employment contracts, leave forms and supporting documents filed against the right employee, with access to special personal information restricted.

  • Education

    Admission forms, bursary applications and supporting documents checked and filed, with children's personal information protected under POPIA.

  • Contracts

    AI contract automation that extracts parties, values, renewal dates and notice periods into an obligations register, so renewals are not missed. It supports your legal team; it does not replace legal review.

  • Client onboarding

    Identity documents and proofs of address collected for FICA checks in financial and legal services, with every check recorded.

Digital document transformation and the digitalisation of data

Digital document transformation means paper stops driving the process: each document arrives, or is scanned, once; the data goes where it is needed; and the file can be found in seconds. We usually start day-forward, so new documents flow through the new process from go-live, and prioritise older records by how often they are needed, with the extent of any back-catalogue work scoped per project.

Before paper originals go, check two laws. The Electronic Communications and Transactions Act recognises electronic records, provided their integrity can be shown, which is why we keep the audit trail. For public bodies, the National Archives and Records Service of South Africa Act, and the provincial archives laws for provincial and municipal bodies, require written authorisation from the National Archivist or the provincial archivist before public records are destroyed, so a scanning project needs a disposal authority in place first. Where records move from an old system, our data migration service moves them with their history intact.

A manager presents results on a wall screen to colleagues in an open office.

Choosing OCR software or a document AI service

OptionSuitsWatch for
Desktop OCR softwareOccasional conversion of scans into searchable filesNo validation and no workflow; staff still key the data
OCR in your scanner or document-management systemMaking archives searchableProduces text only
Cloud document AI services from Microsoft Azure, Google Cloud or AWSHigh volumes of common documents, such as invoices and receiptsCharged per page; check where the data is processed
Language models that read documents, such as Claude and GeminiVaried layouts, long documents and summariesNeeds validation rules and a review step
A private deployment on open-weight modelsSensitive records that must stay in South AfricaHardware and operations are yours

To choose, answer five questions: how many documents arrive a month, how varied the layouts are, how accurate the data must be before it posts, where the data is allowed to go, and which systems must receive it. We test candidate approaches on a sample of your own documents and show you the measured accuracy before you commit. For records that may not leave your control, our private AI service runs the models on infrastructure you own.

Accuracy, risk and the controls we build in

  • Confidence thresholds. Nothing below the agreed confidence level posts without a person checking it.
  • Validation. Totals, dates, VAT numbers and suppliers checked against your records.
  • Poor scans and handwriting. Accuracy falls with image quality, so we measure it on your samples rather than quote a figure in advance.
  • Fraud signals. Duplicate invoices and changed bank details flagged for a person to verify before payment.
  • Personal information. Special personal information and children's information handled under POPIA's stricter conditions, retention periods applied, and processing outside South Africa only where section 72 allows it.
  • Security. Encryption in transit and at rest, access controls and audit logging.
  • Governance. Each document process is entered in your AI use register under our AI governance approach.

Governance, reporting, procurement and continuity

  • Reporting. A named service lead and a monthly service review covering documents processed, the share that passed straight through, exceptions by reason, turnaround times and corrections made. Changes to extraction rules and posting thresholds go through change control.
  • Compliance. POPIA, King IV Principle 12 for boards, PFMA and MFMA audit trails for public bodies, and records kept in line with your file plan.
  • Procurement. Arke Technologies (Pty) Ltd is a Level 1 B-BBEE contributor (sworn affidavit available on request), which gives buyers 135% procurement recognition. Company documents, our CSD supplier number and SARS tax compliance status are available on request. The pilot and build are bought against a fixed-scope proposal and the running service under an SLA; we complete standard bidding documents for tenders. See our company facts.
  • Continuity. You own the documents, the extracted data, the rules and the logs, in open formats, and they stay with you if the service moves.

What document intelligence costs

Last reviewed: September 2026.

We quote the build as a fixed-scope project after testing a sample of your documents, and the running service monthly. What drives the cost:

  • the number of document types, and how varied their layouts are;
  • monthly volume, which sets per-page service charges or the size of a private deployment;
  • the systems the data must post to;
  • the review workflow, and who approves;
  • handwriting, poor scans and the languages involved; and
  • where the data may be processed.

For context, published rates for IT project work in South Africa run from about R550 to R750 an hour ex VAT (matnet.co.za, September 2026). A pilot on one document type is the lowest-risk way to prove the return before you scale.

A finance clerk reviews a printed document beside her laptop.

Frequently asked questions

What is OCR software? Show answer

OCR software uses optical character recognition to turn images of text, such as scanned pages, photos or PDFs made from scans, into text a computer can search, copy and edit. It is the first step in document automation, but on its own it does not understand which value is a total, a date or a supplier, so the data still has to be keyed or validated.

What is the difference between OCR and intelligent document processing? Show answer

OCR reads the characters; intelligent document processing understands the document. It combines OCR with AI to classify each document, extract specific fields and tables, validate them against your records and route them to the right system or person, with confidence scores that tell reviewers which fields to check.

How accurate is AI document processing? Show answer

Accuracy depends on the documents: clean, printed and digitally created documents extract far more reliably than faded scans or handwriting. That is why we measure accuracy on a sample of your own documents before go-live, set a confidence threshold for each field, and send anything below it to a person. You see the measured figures before you commit.

Can document AI read handwriting? Show answer

Yes, within limits. Current models read neat handwriting reasonably well, but accuracy falls with poor handwriting, faint scans and forms filled in across boxes. For handwritten forms we test on your samples, route low-confidence fields to a reviewer and, where it makes sense, move the form online so the data is captured digitally at source.

Where is our data processed, and is it POPIA-compliant? Show answer

Your data is processed where you decide: in a cloud region you approve or on a private deployment in South Africa. POPIA requires security safeguards, limits on retention and conditions before personal information leaves the country, and we design for all three. We encrypt documents in transit and at rest, restrict access and log every action.

Can the extracted data go straight into our accounting or ERP system? Show answer

Yes. Validated data can post to systems such as Sage, SAP Business One, Microsoft Dynamics 365 Business Central and Xero through their interfaces, and documents can be filed in SharePoint or Google Drive with their metadata. We set rules for what posts automatically and what waits for approval, such as a new supplier or changed bank details.

Can we destroy paper originals once they are scanned? Show answer

Often, but not always. The Electronic Communications and Transactions Act recognises electronic records whose integrity can be shown, yet some documents carry their own retention rules, and public bodies need written authorisation from the National Archivist or the provincial archivist before destroying public records. Check your retention schedule and disposal authority first; we build the audit trail that supports it.

Test it on your own documents

Send us a sample of the documents your team keys by hand, anonymised where you can, and we will show you what OCR and document processing AI extract from them, and how accurately.