Building the system takes thirty minutes. The seven days are yours, and they are the part that decides whether it works. That is worth stating plainly, because most "implement AI in a week" plans quietly describe a vendor's build queue. The build is not the constraint. Deciding what your business says, and then checking that the thing says it, is.
Why seven days if the build is thirty minutes?
Because three things cannot be compressed, and none of them is technical:
- Writing down what your business actually says about prices, policies and the things you do not do. Most businesses have never written this down anywhere.
- Testing it against real questions rather than the ones you hope people ask.
- Letting it run for a few days and reading the log before you trust it with everything.
Skip those and you have a live system nobody has checked, which is worse than no system — it answers wrongly at scale, politely, while you assume it is fine.
A week that actually works
| Day | What you do | Time |
|---|---|---|
| Monday | Pick ONE job with a number attached. The enquiry nobody answers, the quote that sits, the diary nobody manages. Write down what that job currently costs you | 1 hour |
| Tuesday | Pull your last 50 enquiries and tally the literal questions. Publish the answers to the top five on your website — half the problem often disappears here | 2 hours |
| Wednesday | Write the material: prices, hours, areas, what you do NOT do, escalation rules, three replies you were happy with | 2 hours |
| Thursday | Build. Thirty minutes on our side. Then spend an hour trying to break it | 1.5 hours |
| Friday | Go live, quietly. Tell nobody. Watch it handle real traffic | 30 minutes |
| Weekend | It works the hours you do not. This is the point of the exercise | — |
| Monday | Read every conversation. Fix what it got wrong. Now measure against Monday's number | 1 hour |
Total: about eight hours of your time across a week, most of it writing.
Why start with one job rather than a strategy?
Because a strategy cannot be checked and a job can. "We are becoming an AI-enabled business" has no failure condition; "we were not answering 40% of our enquiries and now we are" does. Pick something you can count before and after, and the decision about whether to do more is made for you by the result rather than by enthusiasm.
What should you deliberately not do in week one?
- Do not automate four things at once. When something goes wrong you will not know which one, and something will go wrong.
- Do not announce it. A quiet launch gives you a few days to fix the embarrassing answers before anyone important sees them.
- Do not let it handle complaints. Configure that as a hard escalation from day one.
- Do not skip Tuesday. Publishing the five most-asked answers on your own website is free and it is often the biggest single improvement of the week.
What does the first job cost?
It depends which one you pick, and all of these are month to month with no contract:
- Website chatbot — R299 a month plus R250 once-off (R549 at checkout)
- Quoting and invoicing — R80 a month, no setup fee
- AI email assistant — R349 a month plus R500 setup
- Booking system — R1 299 a month plus R2 000 setup
There is a free trial on the chatbot, which means Thursday and Friday can happen before you have paid anything.
The build is thirty minutes. The week is for finding out what your business actually says, which turns out to be the hard part and the valuable one.
What is the shortest version?
One job, chosen because you can count it. Two hours writing down what you say. Thirty minutes building. An hour trying to break it. A quiet weekend live. Then read every conversation on Monday and compare against the number you wrote down.