Many small businesses already use AI. Usually, someone copies text into ChatGPT, Claude or Gemini, pastes the answer somewhere else, and repeats the process the next day. The tool is often a free one, and client details are pasted in with the question.
This is a start. The main benefits come from AI that works inside your systems on the tasks your team repeats most often. A person checks the important parts.
The harder question is which tasks to give it. This is how we decide.
A simple test
AI works well on tasks that are:
- Repetitive: the same kind of task, done many times
- Rules-based: there is a right answer, and your team could explain how they reach it
- High-volume: the task happens often enough that doing it by hand takes a lot of staff time
If a task meets all three, it is probably worth considering for AI. A task that involves judgement, relationships or mistakes that are hard to spot is probably not suitable. The exception is when a person checks every result.
Where AI works well today
Answering enquiries
Every business answers the same questions every day. Examples are “What are your hours?”, “Do you deliver to my area?”, “Is this in stock?” and “Can I book for Saturday?”
An assistant on your website, WhatsApp or email can answer these questions 24 hours a day. It is trained on your own content. It can ask the qualifying questions your team would ask, and it can book appointments. When it cannot help, it passes the conversation to a person. The full chat goes with it, so the customer does not have to explain again.
Enquiries that arrive after hours get an answer immediately. Your team starts the morning with the routine enquiries already dealt with.
Reading documents
A lot of office work involves reading a document and typing its contents into another system. Staff type supplier invoices into the accounting package, order forms into the stock system and enquiry emails into the CRM.
AI is now good at this. It can read an invoice or form, extract the details and check them against what you expect. It can then enter them into the right system for a person to approve. It can also write the weekly summary that someone currently compiles by hand.
Connecting systems
Many businesses use a website, a CRM, an accounting package and a set of spreadsheets that do not share data. Staff retype information from one system into another.
When those systems are connected, a new order can update stock, create the invoice and delivery note, and send the follow-up. Nobody has to copy details between systems. Background agents can monitor for items that need attention and alert a person.
Not all of this needs AI. Much of it is standard integration work, and that is often the better tool. AI is useful where something has to be read or interpreted, such as a free-text email or a scanned PDF. We choose the simplest version that works, even when a more complex version would look more impressive.
Where it doesn’t fit
AI is quick, does not get tired and does a task the same way every time. It is not good at every task, and some of the most useful advice concerns where not to use it.
- Decisions that need judgement. Examples are pricing an unusual job, handling a complaint and deciding whether to extend credit. These need someone who understands the context and takes responsibility for the outcome.
- Relationships. This includes your key clients, negotiations and difficult conversations. People can tell when a reply comes from AI, and that matters to them.
- Tasks where a wrong answer is hard to spot. If a mistake would only become visible much later, or in front of a client, have a person do the task.
- Rare tasks. If something happens twice a month, setting up and maintaining an automation usually takes more effort than it saves.
- Unclear processes. If your team cannot explain how a task is done, AI will not work it out for them. Fix the process first.
Sometimes the right solution does not involve AI. It may be a better form, a clearer process or a simple integration.
Start with one workflow
Do not start with a large project. Choose one task, test that AI handles it well, and then add another.
- List the repetitive work in a normal week. Include anything your team answers, reads, retypes or follows up many times.
- Pick one. The best first candidate is high-volume, has clear rules and is easy to check.
- Run a pilot alongside your team. Let the AI do the work while people keep doing it too, and compare the results.
- Measure it. Count how many items it handled and how often someone had to correct it. Check whether replies went out sooner.
- Then connect it properly, with approvals, logging and access control in place, before you move to the next one.
This way, you learn what AI does well in your business at very low risk. The second workflow is then easier than the first.
Keep a person involved
AI will occasionally get things wrong. You can still use it if you design your process to allow for mistakes:
- Anything important goes to a person for approval before it happens
- Every action is logged, so you can see what was done and why
- The agent passes the task to a person whenever it is unsure
- Someone reviews how it is performing and adjusts it as the business changes
The agent prepares drafts, sorts items and fetches information. A person makes the decisions. Your team keeps control of the work that needs people, and the agent does the rest.
Handle the data properly
AI that handles client information needs more care than copying text into a chat window.
- Use business-tier AI services that do not train on your inputs. Free and consumer tools may use what you type to improve their models.
- Send only what the task needs. An AI answering questions about opening hours does not need a customer’s ID number.
- Know where personal information goes. Most large AI providers process data outside South Africa, and POPIA has rules about that.
- Where data should not leave your control, run an open-source model on a server you control.
- Give your team a short, clear policy on which tools to use and what not to paste into them.
If you are planning a chatbot, our guide to POPIA and AI chatbots covers this in more detail.
We use this ourselves
We also build and run our own AI products. ChatProjects is a workspace where teams chat with their documents using ChatGPT, Claude and Gemini. TenderPipe tracks tenders automatically and sends new opportunities to an inbox, spreadsheet or CRM. Running these products every day shows us what works in practice and what does not.
Where to start
If you are not sure where AI fits in your business, start with our free AI review. It begins by looking at how your team works in a normal week. We will show you two or three tasks worth automating, and tell you which tasks are not worth automating. If the review shows that a better CRM set-up would help more than AI, we will tell you that too.
Filed under AI Services.