060 279 5587 info@sitect.co.za 139 Davies Street, Doornfontein, Johannesburg, 2001 Gauteng, SA
Services / AI Consultation & Integration / Document & Workflow Automation
Reads · Classifies · Drafts · Routes · Audits

Stop typing what a robot can read. Automate the paperwork.

AI that ingests your invoices, contracts, FICA docs, payslips, claims and POs — extracts the fields, validates the data, drafts the response, and routes to the right human for approval. With a complete audit log your auditor will love.

Live in 4 weeks 20+ doc types POPIA + FICA-aware From R14,000

Every business has a person who types things off PDFs.

And every one of those people would rather be doing the actual job they were hired for. Manual document processing is the most expensive bug in your business — and you've never seen it on a P&L.

!

Hours typing the same fields

Vendor name. VAT number. Line items. Totals. Re-keyed from a PDF into your accounting software, every single day. Auditable only if someone remembers to scan.

17 hrsper week, per junior
!

The "where's that contract?" hunt

Filing cabinets. Shared Drives. Outlook inboxes. WhatsApp PDFs. Five places a contract could be — none of them indexed. The day you need it is the day you can't find it.

38 minavg search per document
!

Audit-trail by screenshot

Who approved this? When? Did they actually read it, or just click? Your auditor wants the answer in writing. Currently it's a sticky note on Sarah's monitor.

62%of approvals lack proof

Nine document types. All read fluently.

If it's structured-ish and your team currently re-types it, we can teach the model to handle it. These nine cover ~80% of SA back-office paperwork.

Most common

Tax invoices & POs

Vendor, VAT number, line items, totals, due dates — extracted and matched against the original PO. Flags discrepancies before they hit AP.

vendorvat_noline_items[]totaldue_date

Contracts & NDAs

Parties, term, governing law, termination clauses, value, signature dates — pulled out and indexed. Find every contract with a 30-day exit in one query.

parties[]termvaluerenewalsignatories[]

FICA & KYC docs

SA ID, passports, proof of residence, company CIPC docs — verified, expiry-checked, age-against-document-date, full POPIA-compliant trail.

id_nodoc_ageexpirycipc_no

CVs & HR docs

Candidate name, contact, experience, skills, qualifications — structured for your ATS. Payslips and SARS forms also handled, with PII redaction.

candidateyears_expskills[]education[]

Insurance claims

Claim type, policy number, incident date, parties, damage estimate, supporting docs — assembled into a clean claim packet, fraud-flagged.

policyincident_dateclaim_valueparties[]

Loan & credit applications

Personal details, employment, income, bank statements, affordability — pre-scored against your credit policy. Decline / refer / approve in seconds.

incomeemployerdebt_ratiocredit_score

Bank statements

Multi-page statement parsing — transactions, balances, ending balance, salary deposits, recurring expenses. CSV or JSON output for affordability rules.

accountperiodtxns[]income_recurring

Quotes & proposals

Outgoing AI-drafted quotes (from briefs) + incoming quote comparisons. Auto-summarise three competing quotes into a single comparable table.

scopeline_items[]totalterms

Delivery & courier docs

Waybills, PODs, signed delivery notes, customs forms — captured, OCR'd and linked to the order. Drives faster invoice-to-cash cycles.

trackingsigned_bydelivered_atitems
The workflow engine

Trigger → condition → action.

Every flow is built from the same three building blocks. You drag-and-drop in our admin UI, or edit as YAML if your team prefers code-as-config. Re-deploy in 1 click — no scheduled maintenance windows.

Invoice ingestionAP automation
Email arrives
Extract fields
Match PO?
Amount > R10k?
Route to CFO
FICA onboardingCompliance
Customer signs up
SMS doc request
Verify ID + PoR
Doc age <3mo?
Activate account
Contract renewalSales · Legal
90 days to expiry
Draft renewal
AE reviews + edits
Send for e-sign
File + notify CRM
Insurance claimClaims processing
WhatsApp claim
Assemble packet
Fraud score
Value > R50k?
Route to assessor

21 integrations out of the box.

The tools you already pay for. We don't replace them — we wire them together with intelligence on top.

E-signature
Sign & route
DDocuSeal DSDocuSign AAdobe Sign BDropbox Sign
Document storage
Read & archive
SSharePoint GGoogle Drive DDropbox AAWS S3 MOneDrive
Accounting & ERP
Sync to ledger
QQuickBooks XXero SSage DDynamics
CRM & sales
Update records
HHubSpot PPipedrive SSalesforce
Workflow & triggers
Plug into existing automations
ZZapier MMake Nn8n PPower Automate TMS Teams

Document in, decision out.

Five stages between a PDF landing in your inbox and the right action happening. Every stage is observable, replayable and tunable — no black boxes.

01 · INGEST

Capture

From wherever it lives.
  • Email mailbox
  • WhatsApp upload
  • SharePoint folder
  • API webhook
02 · READ

OCR + LLM

Read text + vision.
  • AWS Textract
  • Google Doc AI
  • GPT-4o vision
  • Field extraction
03 · VALIDATE

Check

Rules + confidence.
  • VAT format check
  • Date sanity
  • PO matching
  • Anomaly detect
04 · ROUTE

Decide

Workflow engine.
  • Per-amount rules
  • Per-vendor rules
  • Approval chain
  • Escalation
05 · ACT

Execute

Sync, send, sign, file.
  • Post to Xero
  • Notify Slack
  • Send for e-sign
  • Audit log
Median end-to-end: 1.4 seconds per page · 95%+ field-level confidence on common doc types. Every step is logged + replayable · POPIA-compliant retention · DSAR-ready exports.

Six artefacts. All yours, all auditable.

You don't pay a per-document fee. You don't pay a SaaS subscription. The pipeline runs in your infrastructure, calling provider APIs at cost.

01 · Doc type schemas

JSON-Schema per document

The shape of every doc the pipeline handles — vendor, fields, validation rules, expected ranges. Maintainable by your team.

JSON-Schema + samples
02 · Extraction prompts

Tuned prompts + few-shot examples

Versioned, A/B-able prompts for each doc type. Trained against 20–50 of your real documents for high accuracy on day one.

Git-backed prompt registry
03 · Workflow engine

YAML-defined workflows

Visual editor for non-technical team + YAML/JSON export for source control. Triggers, conditions, actions all configurable.

Drag-drop + YAML
04 · Integrations

3–6 systems wired up

Whichever tools matter most: Xero/Sage, DocuSeal/DocuSign, HubSpot/Salesforce, SharePoint, Slack. We do the wiring + handoff docs.

OAuth + webhook patterns
05 · Admin dashboard

Run, retry, audit

Web UI for your operations team. See every document, retry failed extractions, search the audit log, export for compliance reviews.

Self-hosted Next.js
06 · Runbook + training

Hand-over your team can run

Markdown runbook for the top 10 incident types. 2-hour live training session for your ops + IT teams. 60 days of Sitect support included.

Live training + runbook

4 weeks. 5 phases. Zero "what does this field mean?" emails.

Most engagement happens in week 1 — we collect your real documents and map the rules. By week 4 your back-office team is processing automatically.

01
Week 1 · 3 hrs

Document audit

You send us 50 real samples per doc type. We define schemas + extraction rules.

02
Week 2 · 1 hr

Tune extraction

Train prompts on your samples. Hit 95%+ accuracy before moving forward.

03
Week 2–3 · 2 hrs

Wire workflows

Build the routing rules. Connect Xero/Sage/HubSpot/DocuSeal — whichever you use.

04
Week 3 · 2 hrs

Pilot

Run live for 5 business days. Tune from real edge cases. Train your team on the dashboard.

05
Week 4 · 1 hr

Go live + handover

Full cut-over. 60 days of Sitect monitoring + weekly tuning. Then your team owns it.

The numbers behind a back-office automation done right.

Median results across our deployed document pipelines. Your numbers will be yours — these are the order-of-magnitude bracket.

82 hrs
Saved per month
Median admin time freed per 1,000 documents processed.
97%
Field accuracy
After 50-sample tuning. Hits 99%+ on well-structured forms.
14×
Faster processing
Median speedup from manual baseline. Same-day vs same-week.
100%
Audit-trail coverage
Every decision logged, replayable, exportable for your auditor.

Pick the scope. We deliver in 3–4 weeks.

Build fee is fixed in writing. Variable cost = OCR + LLM tokens, typically R3–R12 per document all-in. You set monthly caps in the admin dashboard.

Starter

Single doc type

R14,000
/project · ex 15% VAT
  • 1 document type (e.g. invoices)
  • 3 integrations
  • 1 workflow
  • Email + Slack notifications
  • Basic dashboard
  • 30 days post-launch tuning
Choose Starter →
Enterprise

Back-office platform

R60,000+
scoped per engagement · ex VAT
  • Unlimited doc types
  • Unlimited integrations + workflows
  • SSO + multi-tenant + roles
  • Custom ML for niche docs
  • SLA + dedicated PM
  • 90 days hyper-care + QBR
  • On-prem deployment option
Talk to us →

Questions teams ask before signing.

If yours isn't here, send us a sample document — we'll respond with extracted JSON within 24h. No sales call required.

Will it work on our messy real documents?
Almost certainly. We tune extraction on 50 of your actual documents before going live, including the bad ones — scans, photos taken in bad light, handwritten annotations, multi-page PDFs. If after tuning we can't hit 90% field accuracy, the engagement refunds. We've never had to do that, but the guarantee is on paper.
POPIA & FICA — how is data handled?
Three layers: (1) Documents are processed in your tenant — never in shared storage. (2) PII fields are isolated and can be encrypted-at-rest with your own KMS key. (3) Every read is audit-logged with the user, timestamp and purpose. We sign a POPIA-compliant DPA before any document hits the pipeline.
What happens when the AI gets it wrong?
Three safeguards: Confidence thresholds on every field — below X% it routes to a human. Anomaly rules — invoice 10× larger than your usual? Goes to human review. Audit trail — every decision is replayable, so when you find a mistake, you fix the rule and re-run.
Can it learn from corrections?
Yes. When a human corrects an extraction in the dashboard, the correction is stored as a labelled example. Every 30 days we re-tune the prompts on the cumulative correction set. Accuracy generally climbs from 90% on day one to 98%+ by month three.
Do you handle SA-specific document formats?
Yes — SARS tax invoices (with the 7 mandatory fields), SA ID documents (incl. the smart card and the green book), CIPC company docs, FICA proof-of-residence formats, SA bank statements (FNB / ABSA / Standard Bank / Nedbank / Capitec / Discovery / TymeBank). We've seen them all.
What's the monthly running cost?
Variable cost per document: typically R3–R12 all-in (OCR + LLM tokens). A team processing 1,000 invoices/month lands at ~R6,000/mo in usage. You set a monthly cap; the pipeline pauses + alerts before exceeding it. Sitect takes nothing on usage — it's billed direct from providers.
Can we self-host this?
Yes — every Sitect document pipeline ships as a Docker image you run in your own AWS / Azure / on-prem infra. The admin dashboard is a standard Next.js app. The only "external" calls are to OCR + LLM providers, and even those can be swapped for on-prem (Llama 3 / Mistral) if data residency is strict.
What if our process changes?
Workflows are configuration, not code. Your ops team edits them in the admin UI — no developer needed. For bigger changes (new doc type, new integration) we offer a retainer (from R5k/mo) or one-off tuning engagements.

Send us one document.
We'll send back the extracted JSON.

Drop a sample invoice, contract or claim into our secure upload form. Within 24 hours you'll receive the structured extraction back — plus an honest assessment of whether automation is worth it for your volume. No sales pitch.

You'll get back within 24h

  • Extracted JSON of every field
  • Confidence score per field
  • ROI estimate at your volume
  • Recommended workflow shape

More AI Consultation services

AI Strategy & Roadmap

Where does AI fit in your business?

AI Chatbot Integration

Smart chatbots that capture, qualify and convert.

AI Voice & Phone Agents

24/7 AI agents that handle calls.