Insights

What does AI automation cost in South Africa?

There is no list price, and any provider who quotes one before seeing your process is guessing. What you pay is set by six things: how many systems the automation touches, whether they have usable APIs, how clean your data is, the volume, how bespoke the rules are, and what support you want afterwards.

Why won't anyone quote you up front?

Two businesses can ask for the same thing in the same words and get quotes ten times apart, and both can be fair. One has a cloud accounting package with a documented API and forty invoices a month. The other has an on-premise system from 2011, three spreadsheets that disagree, four thousand invoices a month, and six exception rules living in one person's head. Same brief. Different job.

That is why reputable providers scope before quoting. Be wary of a fixed price offered off a single email, and equally wary of anyone who will not give you a range after walking the process with you.

What actually drives the price?

  • How many systems it touches. Each extra system is another set of credentials, another data format, another vendor whose changes you do not control. Five systems is not five times the work of one; the ways it can break multiply. Connecting systems that were never designed to talk to each other is usually the biggest line in a quote.
  • Whether an API exists. With a documented API the connection is largely a solved problem. Without one the alternatives get expensive: scheduled file exports, screen-level automation that breaks when a vendor moves a button, or database workarounds. Ask early which systems have APIs; it is the answer that moves a price most.
  • How clean the data is. Automation is unforgiving about mess people tolerate: duplicate customers, phone numbers stored four ways, a status column with eleven values meaning three things. Somebody has to fix that, and it is the most underestimated line in most budgets. Doing the cleanup yourself is a legitimate way to pay less.
  • Volume. High volume justifies monitoring, retries and error handling that a ten-a-week process does not, and platforms then charge per task, per record, per message or per AI request.
  • How bespoke the rules are. "Remind the customer three days after invoice date" is cheap. Adding "unless they are on 60-day terms, or it is a government account, or the rep flagged a dispute, or it is December" is not. Every exception you retire before the build is money back.
  • What happens after go-live. APIs change, staff leave, the business adds a product line. Budget for monitoring, fixes and training new people, or budget for the automation failing quietly in month eight.

What shape does the cost take?

Even without amounts you can know what the invoice looks like. Costs arrive in four shapes, and a quote that hides them in one figure is worth questioning.

  1. A once-off build fee

    Scoping, build, testing, handover and training, charged fixed-price or on time and materials. Fixed price protects you when scope is genuinely clear; when it is not, the provider has priced their uncertainty and you paid for it.

  2. Monthly platform and licence fees

    The tools it runs on: an automation platform, an extra CRM seat, WhatsApp Business messaging, an AI subscription. These are yours, and they continue whether or not the provider does.

  3. Usage-based charges

    Per run, per message, per document, per AI request. Invisible at low volume, dominant at high volume. Ask for a projection at your real monthly numbers.

  4. Support, retained or ad hoc

    A retainer, or an hourly rate when something needs attention. Both are fine. What is not fine is discovering after go-live that neither was discussed.

Plan for one thing specifically: most automation platforms and AI models bill in US dollars, so your running cost moves with the rand in months when nothing about your business changed. Ask which items are dollar-denominated, and budget headroom.

Can you use the tools you already pay for?

Very often, and it is the cheapest honest answer here. Many Microsoft 365 plans include a level of Power Automate, Google Workspace includes Apps Script, most modern CRMs ship with workflow rules, and most cloud accounting packages already schedule reports and issue recurring invoices. Entitlements differ by tier, so check your plan. Businesses routinely buy an automation platform to do what their CRM already did.

Custom work earns its place when the logic is specific to how you make money, when volume means reliability matters more than convenience, or when the systems were never designed to speak to each other. That is where a properly scoped workflow automation build is worth paying for, and where a no-code flow assembled by a willing staff member becomes a liability by month four.

How do you work out whether it pays back?

Work out your own ceiling before speaking to any provider. The method is hours, not feelings. Count them by watching the process for a week rather than asking people to estimate, because small recurring tasks are always undercounted. Use the fully loaded cost of the person doing the work, so leave, UIF and overhead are included. Multiply out to a year, then subtract the running cost.

Here is the sum with placeholder inputs. They are illustrative, not benchmarks, so use your own. Suppose two people each spend five hours a week re-capturing orders between systems: ten hours a week. Take 52 weeks, less three weeks' statutory leave and twelve public holidays, and call it 46 working weeks — so about 460 hours a year. At a fully loaded cost of R110 an hour, the process consumes roughly R50,000 of paid time annually. That is your ceiling: to pay back inside a year, the build must come in under R50,000 less twelve months of running cost.

Hours are the honest starting point, not the whole picture. Same-day quotes and enquiries answered in minutes are real gains, but never let them carry a case that fails on hours alone. Automation also rarely removes a whole person; it removes the worst part of several people's weeks, which is why automating a process and hiring for it are not the straight swap they look like.

When is automating not worth the money?

  • The volume is too low. Twenty minutes a week is about fifteen hours a year, on the same 46 working weeks. Very little is worth building for that, however irritating those twenty minutes are.
  • The process is about to change. A new system is coming, the team is restructuring, the regulation is under review. Automating what you are about to replace means paying twice.
  • The step should not exist. Reports nobody reads, approvals never once declined, a form that duplicates another form. Deleting a step is free and takes effect this afternoon.
  • Nobody will own it. Every automation needs someone internal who notices when it stops. Without that person it fails silently, and a customer tells you.
  • It is your biggest, messiest process. Understandable instinct, poor first project, and the subject of deciding what to automate first.

A provider who tells you a process is not worth automating has shown you what their advice is worth on the rest.

What should you ask before signing?

Take these to any provider, including us. The answers tell you more than the price.

  • What is in scope, and what is explicitly out?
  • Which parts of the number are once-off, and which recur every month?
  • Which line items are billed in US dollars, and who carries the currency movement?
  • Which of our systems have proper APIs, and where are you relying on a workaround?
  • Who owns the accounts, the logic and the code if we stop working together?
  • Can we see it running on our own data before the final payment?
  • What is the smallest useful version of this, and what would that cost?
  • How many hours should this save, and how will we measure that together?
  • Are you reselling any tool you are recommending?

How big should a first project be?

Small enough that failure would be an inconvenience, not a story you tell for years. A first engagement buys two things: a working automation, and reliable information about whether the provider deserves a second one. Both are cheaper to learn on something modest, which is why automating a single process end to end beats a programme of work.

Common questions

Can I get a ballpark price before a scoping call?

A range, yes, once someone knows how many systems are involved, whether they have APIs, and roughly what volume runs through each month. A firm number without that is either padded for the unknowns or comes back later as variation orders.

Will we pay a monthly fee forever?

Almost always something, because automations run on platforms that bill monthly. A provider retainer on top is optional. Ask for running cost quoted separately from build cost before you sign.

Why do automation quotes vary so much between providers?

Because they are quietly pricing different jobs. One provider has assumed your systems have usable APIs and your data is clean; another has walked the process and priced the exception rules, the cleanup and the support. Compare what each quote puts out of scope, not only the number at the bottom, and ask both to separate build cost from monthly running cost.

Want a real number for your own process?

Bring one process to a free 30-minute session. We will walk it with you, say honestly whether it is worth automating, and where we know enough, give you a realistic cost shape and range. No quote is written before we understand the work. You can also read how we scope and build a process automation.