What is the wrong comparison?
“This AI tool replaced a person.” It is the headline everyone writes and it is usually false, because the tool does a slice of the job and the person did the whole thing. A receptionist answers the phone, greets the visitor, calms the angry customer, notices the thing nobody wrote down. A chatbot answers the repeated questions and books the diary. Comparing their costs directly flatters the tool and insults the person.
What is the right comparison?
Hours nobody was covering. A full-time employee is at the desk roughly 45 hours a week. There are 168. The 123 hours in between are when the website’s enquiries went to a form that was read the next morning, which puts every one of them in the worst bucket of the best study on the subject: Harvard Business Review’s audit of 2,241 companies found that a lead contacted within an hour was nearly seven times as likely to be qualified as one contacted an hour later, and more than sixty times as likely as one left for a day; 23% of companies never replied at all 1. The tool is not replacing the receptionist. It is staffing the 123 hours she was never paid for.
How do you work out the ROI?
With one number you already know: what a customer is worth to you over a year. Then ask how many enquiries arrive after hours or go unanswered. The arithmetic is not clever.
| Step | Question | Where the number comes from |
|---|---|---|
| 1 | What is one new customer worth a year? | Your average sale × repeat rate |
| 2 | How many enquiries a month arrive after hours or go unanswered? | Your form, your call log, your inbox timestamps |
| 3 | What share of those would you have won with an instant answer? | Be conservative: use your normal close rate, not a hope |
| 4 | Recovered customers a month × step 1 | The tool’s return |
| 5 | Compare with the tool’s monthly cost | From R299 a month for a website chatbot |
For most service businesses, step 4 is larger than step 5 at one recovered customer a month. That is the whole case; there is no need to dress it up. It is also why adoption is broad and impact is narrow: McKinsey’s 2025 survey found 88% of organisations using AI in at least one function but only a minority able to point to an effect on profit 2. The ones that can are the ones that ran the table above on real numbers.
What does the tool not do?
Judgement, empathy and exceptions. It should hand the conversation to a person the moment it stalls, with the history attached, so the customer does not repeat themselves. A tool sold as replacing the person entirely is sold on a false premise, and the business that buys it on that premise will be disappointed by exactly the parts it should never have automated.
What does the person gain?
Her day back. The forty repeated questions, the diary admin and the after-hours backlog were never a good use of a person, and removing them leaves the work that needs one. In the businesses that do this well, the receptionist is not replaced; she stops being an answering machine.
What does it cost to find out?
Nothing, for the first fifty messages or seven days. The website chatbot has a free trial with no card; count the conversations it has in a week that would otherwise have been a form, and run the table above on your own numbers.
Questions people ask
Will a chatbot reduce my staff costs?
It reduces the hours spent on repeated questions and diary admin, and it covers the hours nobody was working. Whether that changes headcount is a decision about what your people should be doing, not something the tool decides.
How quickly does it pay back?
On the table above, usually at the first recovered customer. The free trial is the cheapest way to measure your own enquiry volume before deciding.
What if it answers wrongly?
It answers only from your own pages and price list, refuses what it was not trained on, and hands over to a person when it is unsure. You can edit any answer from the dashboard.