Most businesses have now been told that AI will transform them, without being told what to actually do on Monday. The gap between those two things is where budgets go to die.
This is a practical view for owners and managers of growing South African businesses: what AI does well enough to depend on today, what it does not, and how to scope a first project so you can tell whether it worked.
What does AI genuinely do well today?
Reading things at volume
Summarising, classifying and extracting structure from documents, emails and enquiries. If a person currently reads a queue and decides where each item goes, that is a strong candidate. The work is repetitive, the criteria are usually explainable, and errors are visible.
Drafting a first version
Response drafts, product descriptions, proposals and summaries: anything where a person reviews before it goes out. The productivity gain comes from editing rather than starting from a blank page, and it is real. The failure mode is removing the reviewer.
Answering routine questions from your own information
Customer and staff questions about prices, policies, hours and procedures, answered from approved documents, with a handover to a person when needed. Our guide to AI customer service agents explains how to do this safely.
Finding patterns in structured data
Segmentation, anomaly detection, forecasting and scoring across data you already hold. This is the least discussed and often the highest return, because it surfaces things nobody had time to look for. Examples include predictive lead scoring and dynamic pricing.
Where does AI pay off first in a business?
| Area | Typical first project | What to measure |
|---|---|---|
| Sales | Scoring and routing leads; drafting follow-ups | Response time, conversion rate by score band |
| Customer service | Answering routine WhatsApp and web questions | Correct resolution rate, response time, satisfaction |
| Marketing | Drafting content variants; summarising campaign results | Time saved, output reviewed and approved |
| Operations | Extracting data from invoices, orders and forms | Hours saved, error rate against manual entry |
| Management | Weekly performance summaries and anomaly alerts | Time to decision, issues caught early |
What does AI not do reliably?
- Anything requiring certainty about facts it was not given: it will produce a confident answer regardless.
- Judgement calls with real consequences and no human review.
- Work where being wrong occasionally is unacceptable and errors are hard to spot.
- Replacing a process nobody has bothered to define: automation applied to chaos produces faster chaos.
How do you scope a first AI project?
The projects that succeed look almost boringly narrow. Pick one recurring task, with a measurable before, where a mistake is visible and recoverable.
- Name the task in one sentence, without using the word "AI".
- Measure what it costs today: hours per week, or delay in days.
- Decide what "working" means numerically before you build anything.
- Keep a person reviewing output until the error rate is known, not assumed.
- Compare against the baseline after a defined period, and be willing to stop.
What questions should you ask an AI vendor?
- Where does our data go, and is it used to train anything?
- What happens when the model is wrong: who sees it, and how quickly?
- What does this cost per unit of work, at volume, not in the pilot?
- What breaks if the provider changes their model or pricing?
- Can we see the source behind each answer or recommendation?
The cost question is where enthusiasm frequently meets arithmetic. Per-request costs that are trivial in a pilot can become significant at production volume, and the pilot rarely models that.
What about POPIA and company data?
Putting customer or staff information into an AI tool is processing personal information under POPIA. Use business accounts with terms that exclude your data from model training, give tools only the data a task needs, write down which tools staff may use for what, and put agreements in place with providers that process personal information on your behalf. A short internal AI policy prevents most problems before they start.
Does your data need fixing before AI?
For drafting and document work, usually not. For anything that scores, forecasts or reports, yes: AI built on disconnected, inconsistently defined data produces confident nonsense. Connecting your marketing, CRM and finance data is often the real first project. Our guide to what a growth intelligence layer does explains that groundwork, and AI growth intelligence describes how we approach it, starting with a diagnostic audit.
Frequently asked questions
Where should a small business start with AI?
With one recurring task that takes measurable time today, where mistakes are easy to see and fix, such as drafting replies, sorting enquiries or answering routine customer questions from approved information.
Will AI replace my staff?
For most growing businesses, AI changes jobs more than it removes them: it takes over repetitive reading, sorting and drafting so people can spend more time on customers, judgement and selling. Keep people reviewing anything with consequences.
Why do AI projects fail?
Most often because the process being automated was never clearly defined, success was not measured against a baseline, or the data underneath was disconnected and inconsistent. Narrow, measured projects succeed far more often.
Is it safe to put company data into ChatGPT or similar tools?
Use business plans whose terms exclude your data from model training, avoid putting sensitive personal information into consumer accounts, and set a simple internal policy on which tools staff may use for what. POPIA applies to personal information processed this way.
How much does an AI project cost?
It ranges widely, from monthly subscriptions for off-the-shelf tools to custom builds. At Lulonke, a WhatsApp AI agent costs R25,000 to set up and R6,500 a month, and data and growth intelligence work typically starts at R65,000 to R220,000 after a diagnostic audit.
