Short answer: in 2026 a website assistant trained on your own content costs roughly R15 000 to R60 000 to build properly in South Africa. A WhatsApp agent that answers customers, qualifies leads and hands over to a person costs R30 000 to R120 000. Running costs are usually R300 to R3 000 a month for a small or mid-sized business. The price is driven far less by the AI itself than by what the assistant has to connect to and how carefully it is tested.
Typical prices in 2026
| What you are building | Typical build price | Timeline |
|---|---|---|
| Website assistant that answers from your pages and documents and captures leads | R15 000 to R60 000 | 1 to 3 weeks |
| WhatsApp agent: answers, qualifies, books, hands over to a person | R30 000 to R120 000 | 2 to 6 weeks |
| Inbox or lead triage: reads incoming email or forms, sorts, drafts replies | R25 000 to R90 000 | 2 to 4 weeks |
| Document processing: invoices, forms or contracts read and captured automatically | R40 000 to R150 000 | 3 to 8 weeks |
| AI built into a product (a tutor, an examiner, call analysis) | Scoped per project | Weeks to months |
What it costs every month
- Model usage. You pay the AI provider per word processed. For a business handling a few hundred conversations a month this is often a few hundred rand; a few thousand conversations, a few thousand rand. A well-built assistant uses a cheaper model for simple questions and caches what it can.
- WhatsApp. The WhatsApp Business Platform is accessed through a provider that charges a monthly fee, and Meta charges for messages your business initiates. Replies to customers who messaged you first are generally free within the service window.
- Hosting and monitoring. Small: usually a few hundred rand a month.
- Upkeep. Prices change, your products change, customers ask new things. Budget an hour or two a month for somebody to review conversations and update what the assistant knows.
What moves the price
- Answering versus doing. An assistant that answers questions is simple. One that checks stock, books an appointment, takes a payment or updates your CRM needs an integration for each action.
- Where your knowledge lives. A tidy website and a price list are easy. Ten years of PDFs, emails and what only the owner knows is a content project first.
- Hand-over. A proper assistant knows when to stop and pass the conversation to a person, with the transcript, and lets that person reply in the same thread. That is real engineering and it is the part customers notice.
- Testing. The difference between a demo and something you can put in front of customers is a set of real questions it is tested against every time it changes.
- Languages and channels. Website only is cheapest. Add WhatsApp, Instagram and email, or several languages, and each adds work.
What an assistant can honestly do
It answers the same forty questions your team answers every day, instantly, at 22:00 on a Sunday. It collects a name, a number and what the person wants before a human gets involved. It reads a document and pulls out the fields. It does these well.
It should not quote prices you have not given it, promise delivery dates, give legal or medical advice, or argue with an upset customer. A well-built one is told this and hands over instead. The assistant on this website works that way: it answers from the same content as the pages, will not invent a price, and passes the chat to our team with the full transcript when someone asks for a person or a quote.
When it is not worth it
- 01You get fewer than a handful of enquiries a day. A good FAQ page and a fast human reply will do more for less.
- 02Your answers are not written down anywhere. Write them down first. That alone will speed up your team.
- 03You want it to replace the sales conversation. It should start that conversation and qualify it, not close it.
Three examples from our own work
GoFlight has an AI instructor and an AI oral examiner that student pilots pay a subscription for. ZADialer, our own product, records, transcribes and analyses every sales call so managers can coach. And beyond the case studies we build WhatsApp and social inbox agents that qualify leads for clients we cannot name. The common thread is that the AI sits inside a product people already use, not beside it.