AI does not replace an employee. It replaces a set of tasks, and the honest question is which tasks and what they were costing you. Framing it as replacement is how businesses end up disappointed: they remove a person and discover that most of what that person did was judgement, exceptions and knowing which rules to break.
What does AI actually take off a person's desk?
| Task | Automatable | Why |
|---|---|---|
| Answering the same eight questions | Almost entirely | The answer does not change and does not need judgement |
| Taking a routine booking | Yes | A diary lookup with rules |
| First-line triage of an inbox | Mostly | Sorting and drafting; a person still approves |
| Sending reminders | Entirely | The only failure mode is forgetting, which software does not do |
| Handling an exception | No | Someone has to decide, and be accountable for deciding |
| Calming an angry customer | No | Every attempt makes it worse, quickly |
| Knowing that this client always pays late but always pays | No | Institutional memory is not in any document |
Read the bottom three rows again before making a headcount decision. They are the reason the "replaced an employee" framing usually costs more than it saves.
How do you calculate it honestly?
Not by comparing a monthly subscription with a monthly salary. That comparison is designed to flatter the subscription. Do it this way instead:
- Log a week. Have the person write down what they did in half-hour blocks. A week is enough and it is genuinely revealing.
- Split the list into routine and judgement.
- Total the routine hours and price them at the fully-loaded hourly cost of that role — salary plus levies plus the seat, not the gross salary divided by 160.
- Compare that number with the monthly cost of automating it.
- Then ask the more useful question: what would that person do with the routine hours back? If the answer is "sell", "follow up quotes" or "look after existing customers", the return is bigger than the saving — and you keep the person.
What is the realistic outcome?
In most small businesses, not a smaller team. A team that stops doing the parts of the job nobody wanted to do, and a business that stops losing the enquiries that arrived after hours. The saving is real but it is usually opportunity, not payroll — and opportunity is harder to put in a spreadsheet, which is why vendors quote payroll.
When is replacement actually the right call?
When the role is genuinely and entirely routine, and everybody involved knows it — a data-entry seat, a copy-paste job between two systems, an out-of-hours answering service whose only function is to take a message. Those exist. They are rarer than the pitch decks suggest, and treating a customer-facing role as one is the mistake this whole page is about.
What does it cost to automate the routine part?
Less than most people expect, which is exactly why the comparison to a salary is misleading in both directions:
- Website chatbot — R299 a month plus R250 once-off
- AI email assistant — R349 a month plus R500 setup
- Booking system — R1 299 a month plus R2 000 setup
- Quoting and invoicing — R80 a month, no setup fee
All month to month, no contract. The cost is small enough that the interesting question is not whether it pays for itself, but whether the job you pointed it at was the right one.
Most of what you are paying a good front-desk person for is the four things a machine cannot do. The other six are what you should be automating, and they are also the six they liked least.
What is the shortest version?
Log a week. Split routine from judgement. Automate the routine and give the hours back rather than removing the person. If the numbers only work by removing somebody, check the list again — the judgement column is usually longer than it first looked.