084 227 4264 [email protected] 39 Modderfontein Road, President Park, Midrand, 1685, Gauteng, South Africa
Services / AI Consultation & Integration / AI Voice & Phone Agents
24/7 · Multi-language · Live in 5 weeks

AI that picks up the phone so your team doesn't have to.

Real voice agents — not pre-recorded IVR menus — that answer inbound calls, qualify outbound leads and book meetings 24/7. Speaking SA English, Afrikaans and isiZulu in your brand voice. Hands off cleanly to a human when it matters.

Live in 5 weeks Inbound + outbound POPIA + RICA-aware From R18,000

Hold music is the sound of revenue walking away.

Most SA businesses lose customers to "please hold" — not because the team is bad, but because there aren't enough humans on the line at 19:30 on a Friday.

!

"Press 1 for sales…"

Old-school IVR trees frustrate callers within 20 seconds. They were designed for a 1995 PBX, not a 2026 customer. Most callers hang up before reaching a human.

47%IVR abandonment rate
!

After-hours dead air

If a customer calls at 18:30 and gets voicemail, they call your competitor at 18:31. The first 6 hours after a missed call are when 80% of those leads convert elsewhere.

3 in 4lost leads call once
!

Tier-1 burnout

Your best support agents spend their day on "what's my order status?" calls. By 14:00 they're exhausted, and the complex calls — the ones that need humans — get the worst version of your team.

68%of calls are repetitive

Two superpowers. One agent.

The same trained voice — inbound and outbound. One personality, one knowledge base, one source of truth for what your customers heard.

Inbound
Picks up · answers · resolves
  • Order & delivery status — looks up the order in your system and reads the courier's latest tracking event.
  • Account & balance questions — pulls live data from your finance API, reads back the latest invoice or balance.
  • Appointment booking — checks staff availability, books the slot, sends the email + SMS confirmation.
  • Password / OTP help — authenticates the caller, triggers a one-time reset, walks them through the next step.
  • Smart routing — recognises complaint intent, escalates to a real human with full context in < 5 seconds.
  • After-hours coverage — handles calls 24/7, captures messages, urgent ones page the on-call team.
Outbound
Calls · qualifies · books
  • Lead qualification — calls inbound web leads within 2 minutes, qualifies, books a meeting with sales.
  • Appointment reminders — calls 24h before bookings, confirms, reschedules if needed, updates calendar.
  • Invoice reminders — friendly payment-due nudges with payment-link follow-up via SMS.
  • Post-call NPS — calls customers after service interactions, captures CSAT in < 90 seconds.
  • Re-engagement — calls dormant leads or churned customers with an offer, books a winback meeting.
  • Survey campaigns — runs structured-question campaigns, exports clean answers to your BI tool.

South African voices. South African accents. South African slang.

Generic US/UK voice agents stumble over "howzit", "now-now", "sharp", isiZulu names and Afrikaans street addresses. We train every voice on local data — and you can hear the difference within five seconds.

SA EN
South African English
Local accent, SA expressions, place names.
Production
AFR
Afrikaans
Western Cape + Highveld accent options.
Production
ZUL
isiZulu
Native-speaker quality, KZN-tuned.
Production
XHO
isiXhosa
Eastern Cape + Gauteng accents.
Beta
SES
Sesotho
Free State + Gauteng tuned.
Beta
SET
Setswana
North West + Botswana cross-border.
Beta

Works with the lines you already have.

We don't replace your PBX — we plug the agent in alongside it. Most SA telephony stacks are live within a week of contract signing.

Primary
Twilio Voice
SA numbers · SIP
Generic SIP
Most PBX / VoIP
Teams Phone
Direct Routing
Zoom Phone
via Local Bridge
Native
3CX
Popular in SA
Asterisk / FreePBX
On-prem PBX
Vonage / Plivo
SIP trunks
Under the hood

The voice pipeline, end-to-end.

From the caller's first "hello" to the action that's taken — every stage is observable, replayable and tunable. Sub-second latency at every step.

1 · Phone in

Twilio / SIP / PBX
Audio stream

2 · Speech-to-text

Deepgram · Whisper
~250ms

3 · LLM brain

GPT-4o · Claude · Gemini
~600ms

4 · Tools / RAG

CRM · Calendar · APIs
Parallel calls

5 · Text-to-speech

ElevenLabs · OpenAI · Azure
~180ms

6 · Caller hears reply

Or warm-transfer to human
Full audit log
Total round-trip: ~1.2s caller speaks → caller hears reply · Sub-second on a warm pipeline. Every call recorded, transcribed and retained 90 days · POPIA + RICA compliant.

Six artefacts. All yours.

Every Sitect voice agent ships with the same set of artefacts — so your team can run, tune and audit the agent without us in the loop.

01 · Voice persona

Brand voice & script library

One persona, one cloned voice, one library of brand-tone phrases — including the SA-specific greetings, apologies and confirmations.

1 cloned voice + prompt library
02 · Telephony setup

Numbers, routing & SIP wiring

SA numbers provisioned, calls routed through your PBX or our Twilio bridge, warm-transfer rules configured, DR fallback to your existing team.

Twilio · SIP · 3CX wiring
03 · Knowledge base

RAG on your business

Product catalogue, pricing, policies, FAQs, courier instructions — indexed and retrieved in real-time during the call. No hallucinations on facts.

Pinecone · Qdrant
04 · Tool integrations

Actions the voice can take

CRM lookup, calendar booking, payment-link send, order-status check, OTP resend — 3 integrations on the Inbound tier, 6 on Inbound + outbound, unlimited on Enterprise.

HubSpot · Pipedrive · Calendly
05 · Recording & audit

Compliance-grade audit trail

Every call recorded, transcribed, stored with metadata. Searchable by phone number, agent name, intent or sentiment. POPIA-compliant retention.

90-day default retention
06 · Analytics & tuning

Call dashboard + weekly tune

Volume, talk-time, resolution rate, transfer rate, top intents, sentiment trends. Weekly tuning included for the first 30–90 days, depending on tier.

Live dashboard + weekly report

5 weeks. 5 phases. One agent picking up the phone.

You'll spend ~10 hours of stakeholder time across the whole build. Most of that is in week 1 — after that we work in the background and you sign-off at the gates.

01
Week 1 · 4 hrs

Discovery

Map the use cases, the call flows, the integrations, the escalation rules. Choose the voice.

02
Week 2 · 2 hrs

Voice clone

Record voice samples, fine-tune the synthesised voice, validate accents and pronunciation.

03
Week 3 · 1 hr

Wire telephony

Twilio / SIP / PBX connections, number porting if needed, warm-transfer rules tested.

04
Week 4 · 2 hrs

Pilot calls

Run 50–100 real calls against pilot users. Tune from real transcripts daily.

05
Week 5 · 1 hr

Go live + tune

Full traffic switched over. 30–90 days of weekly tuning + monitoring included, depending on tier.

The numbers behind a voice agent done right.

Median results across our deployed voice agents. Your numbers will be yours — these are the bracket to plan against.

92%
Answer rate
Of calls picked up — including outside business hours.
<2s
Pick-up time
Median ring-to-answer. No hold music, no menus.
41
Concurrent calls
Median capacity per deployment — scales to 1,000+.
R74k
Monthly savings
Median Tier-1 phone-support cost displaced per month.

Recorded, encrypted, disclosed.

Voice AI in SA touches POPIA (Section 11 processing, Section 71 monitoring) and RICA (call interception rules). We don't gloss over it — we build it in. Every call opens with a disclosure, every recording is encrypted, every transcript is auditable.

POPIA Section 11 consent RICA-compliant recording AES-256 encryption SA data residency option Configurable retention DSAR-ready exports

Every call captured for audit:

Audio recordingWAV · AES-256
Full transcriptSearchable JSON
Tool calls logReplayable
Sentiment tracePer-utterance
Consent timestampISO-8601 UTC
Default retention90 days

Pick the call coverage that fits.

Build fee is fixed in writing. Ongoing telephony + STT + TTS + LLM costs are billed at provider cost (~R1.40–R2.80 per call minute) with a monthly cap you set.

Inbound

Inbound-only agent

R18,000
/project
  • 1 phone number + 1 voice
  • Up to 4 use cases
  • 3 tool integrations
  • English-only
  • Warm transfer to one team
  • 30 days post-launch tuning
Choose Inbound →
Enterprise

Contact-centre replacement

R72,000+
scoped per engagement
  • Unlimited numbers + multi-voice
  • Unlimited use cases
  • Unlimited integrations
  • All 6 SA languages
  • Skills-based routing
  • SLA + dedicated PM
  • 90 days tuning + QBR
Talk to us →

Common voice-agent questions.

If yours isn't here, hit the chat in the bottom-right or call us on 084 227 4264.

Will my customers know it's an AI?
Yes. We always open the call with a clear disclosure — "Hi, I'm Aria, a virtual assistant at Sitect". POPIA and basic ethics both require it. Callers who ask for a human are transferred straight away.
How real does the voice actually sound?
Indistinguishable from human on a normal phone line for most callers. We use ElevenLabs / OpenAI TTS / Azure Neural — modern voice synthesis is dramatically better than even 12 months ago. You'll record the demo call yourself before signing.
What's the latency? Is there an awkward pause?
Replies come back quickly enough for a natural conversation — we measure the response time with you before launch. We tune for "barge-in" (the caller can interrupt the agent mid-sentence, just like a human).
POPIA & RICA — what do we actually need to do?
Three things: (1) Caller hears a disclosure that the call is recorded and that an AI is involved — we hard-code this; (2) Recordings stored encrypted with a configurable retention — default 90 days; (3) Your Information Officer signs a one-page record of processing — we draft it. We do not use call audio to train any vendor model.
What happens if the agent can't help?
Three escalation paths: (a) Warm transfer to a live human with the transcript spoken aloud, (b) Schedule a callback for next business hour, (c) Send a follow-up SMS/WhatsApp with a payment link or self-service link. You pick the rules per intent.
Can it handle complex authentication?
Yes — we integrate with your existing auth (OTP via SMS, knowledge-based questions, voice biometrics for high-security). For account changes or financial transactions we always recommend a second human-in-the-loop confirmation, regardless of how confident the model is.
What does this cost to run per month?
Voice usage is the variable cost. Average per-minute cost lands at R1.40–R2.80 (STT + LLM + TTS + telephony). A typical inbound deployment handling 2,000 calls/month at 3min avg costs ~R12k/mo all-in. You set a monthly cap on the dashboard.
Can we keep our existing phone numbers?
Yes — we port your existing SA numbers (082/083/0860/etc.) onto the voice agent stack. Porting typically takes 5–10 business days; during port we run a temporary number in parallel.

We'll call you with a voice agent.
You decide what happens next.

Drop your number in the form and within 24 hours a Sitect-built voice agent calls you, demos a real use case relevant to your industry, and answers your questions live. No human pitching — just the agent doing its thing.

What you'll hear in the demo call

  • SA-accented voice in real time
  • Tool call (calendar booking)
  • Warm transfer to a human
  • Full transcript by email

More AI Consultation services

AI Strategy & Roadmap

Where does AI fit in your business?

AI Chatbot Integration

Smart chatbots that capture, qualify and convert.

LLM API Integration

OpenAI, Claude, Gemini — embedded in your stack.