Before You Add AI to Your Business, Fix This First.

“Maybe we need AI.”

Orders are late again. Customer details live in three spreadsheets. Someone captures the same information twice, while delivery instructions sit in a WhatsApp conversation nobody else can find.

There are two “final” versions of the quotation. A follow-up was forgotten. Your team is busy, customers are waiting, and you have heard that AI can automate almost anything.

“Maybe we need AI.”

It is an understandable thought. But do you actually need AI, or do you first need to fix the process?

Before choosing a tool, find out where the work gets stuck, where information comes from and why someone has to step in. That gives AI a useful job to do.

Before you make your business smarter, make sure it makes sense.

1. AI is a tool, not a business plan

A demonstration turns a pile of documents into a neat summary in seconds. Suddenly, the question becomes: “What can AI do for us?”

Start one step earlier: “What problem are we actually trying to solve?”

“Our quotes take too long” is a starting point. Look closer: is the delay caused by writing, missing prices, waiting for approval or finding the customer’s requirements? Each suggests a different solution.

AI may help. So might ordinary automation, a feature in software you already pay for or clearer responsibilities.

This matters locally. GSMA’s June 2026 report on South African SMEs and AI identifies skills, organisational capacity and data quality among adoption barriers. A promising tool still needs a business able to use it well.

2. A bad process doesn’t become a good process just because it’s automated

Consider an illustrative order process. A customer emails an order. One employee captures it in Excel. Another copies it into accounting software. A third receives the job details by email.

The delivery address is missing. The quantity is ambiguous. Nobody knows who should ask the customer.

Now introduce AI to read the order and move the details between systems. It could flag omissions, if designed to do so. But speeding up the transfer alone could spread incomplete information faster.

Accounts prepares the wrong invoice, production starts the wrong quantity and someone has to unwind the work.

Automating a broken process doesn’t necessarily fix it. Sometimes it just makes the mistakes happen faster.

Agree on required information, who checks it and what happens when something is missing. Then choose how technology can support those steps.

3. Your information is everywhere

The customer’s details are in one spreadsheet. The latest price is in another. The order confirmation is an email, delivery instructions are on WhatsApp and the invoice lives in accounting software.

Which version of the truth is your AI supposed to use?

If a customer changed their address yesterday, an old order may look convincing while being wrong. A polished answer does not prove the information behind it is current.

Decide which record is authoritative, who updates it and how changes reach the people who need them. One reliable source can span connected systems; it need not mean replacing everything with one enormous application.

If that sounds familiar, our guide to outgrowing spreadsheets, email and WhatsApp explores the warning signs.

4. Sometimes you don’t need AI at all

A missing order number rarely needs artificial intelligence. It needs a required field.

Many frustrating jobs have clear rules. Ordinary software can:

• Use a form to capture required information upfront.
• Notify the next person when an order changes stage.
• Show outstanding work on a shared dashboard.
• Create a quotation or invoice from details already entered.
• Remind staff when a follow-up becomes overdue.
• Keep one agreed record instead of several competing spreadsheets.

When the instruction is predictable—“if this happens, do that”—rule-based automation is often a sensible fit. It still needs testing, maintenance and someone responsible for exceptions.

AI becomes more relevant when work involves interpreting varied language, extracting information from inconsistent documents or recognising patterns. Often the useful solution combines both: AI interprets an enquiry, fixed rules check required fields and a person approves the quotation.

Choose the tool after you understand the task. Our business process automation guide explains where ordinary automation can help.

5. Where AI actually becomes useful

Once the workflow is understood, AI can support specific jobs:

• Summarising long customer enquiries for the person preparing a quote.
• Extracting order details from differently formatted documents.
• Suggesting categories and routing requests to the right team.
• Flagging unusual stock movements for someone to investigate.
• Drafting responses for staff to check and send.
• Helping staff find answers in approved business knowledge sources, with references they can verify.

Define what happens when information is missing, the suggested category is wrong or the system cannot find a reliable answer. Important prices, commitments, payments and sensitive decisions may still need human review.

A 2026 South African Journal of Information Management review links useful SME AI adoption with trusted data, connected systems, skills and process discipline. It synthesises international studies through a Western Cape lens; it is not a survey of South African businesses.

You do not need a perfect business before trying AI. You need a defined task, suitable information and a way to check whether it helps.

6. Start with one annoying process

Listen for the complaint your team repeats: “We capture this twice.” “We’re always waiting for approval.” “I spend Friday building the same report.”

Pick one. Follow a real job from start to finish with the people doing it. Note where they chase missing documents, search emails, check job progress or wait for someone else.

Ask what each intervention is achieving. Is someone making a judgement, or filling a gap the process should have prevented?

Record a baseline: time spent, turnaround time, corrections or missed follow-ups. Trial one change on a manageable set of jobs and compare the results, including the time spent checking outputs.

A small, clear improvement is easier to assess than “introduce AI across the business”. If choosing software becomes the next step, compare custom and off-the-shelf options before committing.

7. The five-minute AI readiness test

Choose that one process. Answer “yes”, “partly” or “not yet”:

1. Can we explain the process from beginning to end?
2. Do we know where information comes from?
3. Do we know who is responsible for each step?
4. Is the information reasonably accurate and accessible?
5. Can we identify the actual problem we want technology to solve?
6. Would we know how to measure an improvement?
7. Do we know which decisions still require human involvement?

Mostly confident answers suggest you have a useful basis for exploring a small trial. They do not prove that AI is needed or that a particular product is suitable.

Several “partly” answers point to things to clarify with your team. Mostly “not yet”? Start by mapping the work and agreeing who owns the information and decisions.

This is a reflection exercise, not a certification or scientifically validated assessment. Begin with the question you cannot answer confidently. What would you need to find out before spending money?

Keep confidential information out of casual experiments

Do not casually upload customer records, employee details or confidential business documents into unfamiliar AI tools. First check who can access the information, how the provider uses and retains it, and what controls are available.

Use fictional or appropriately de-identified examples for early experiments. Agree staff permissions, confidentiality requirements and human oversight before connecting real records.

South African businesses must consider applicable POPIA obligations, including lawful processing and appropriate security safeguards. Overseas processing may require additional consideration. Choosing an AI product does not by itself establish compliance.

Understand the problem. Improve the process. Then choose the technology.

That is where UNKNOT starts: understanding how work happens and what keeps getting in the way.

The answer could be workflow automation, connecting existing systems, custom business software or AI applied to a specific task. Sometimes simplifying a handover is enough.

Before you make your business smarter, make sure it makes sense.