Almost every business conversation now finds its way to artificial intelligence. A director wants a chatbot. A finance team wants automatic reports. Sales wants an agent that follows up every lead. Management wants to ask the system a question and receive an answer immediately.
None of these ideas is foolish. Many are already possible. The problem is usually the order in which businesses try to implement them.
A company may still be approving purchases through WhatsApp, tracking stock in several spreadsheets and preparing its monthly reports by copying figures from one file into another. Then someone asks whether AI can predict demand, write management reports and answer questions about profitability.
It probably can produce an answer. Whether that answer can be trusted is a very different question.
The demo is not the business case
AI demonstrations are impressive because they compress a complicated task into a few seconds. A document is uploaded, a prompt is entered and a polished answer appears. That experience is useful, but it can create the wrong expectation. It makes AI look like a product that can simply be switched on across a business.
Real businesses are messier than demonstrations. Customer names are duplicated. Product codes are inconsistent. One department follows a formal approval process while another relies on a phone call. Important context sits in email threads, meeting notes and the memory of one experienced employee.
When AI is placed on top of that environment, it does not remove the disorder. It reads the disorder. Sometimes it even presents the result with enough confidence to make the disorder look organised.
This is why the first question should not be, 'Which AI tool should we buy?' A better question is, 'Which business problem are we trying to solve, and what must be true for us to solve it safely?'
AI needs something solid to work with
Useful business AI depends on four things that are not particularly glamorous: a clear process, reliable data, appropriate access and a person who remains accountable for the outcome.
A clear process. If nobody can explain how a purchase is requested, approved, received and paid for, an AI agent cannot be expected to manage that process reliably. The process does not have to be perfect, but it must be understood.
Reliable data. An AI system can summarize or interpret the information it receives. It cannot quietly repair years of inconsistent product records, missing costs and duplicate customers without creating new risks.
Appropriate access. An assistant that only reads a report carries a different level of risk from an agent that can create payments, edit employee records or confirm sales orders. Those permissions should never be treated as a technical afterthought.
Human accountability. Someone must still own the decision. AI may prepare, recommend, flag or draft. The business must decide what it may approve, what requires review and who is responsible when something goes wrong.
The operational system matters more than the chatbot
This is where ERP systems become important. Not because every company needs more software, and not because an ERP system automatically makes a business efficient. A poorly implemented ERP can be just another expensive source of frustration.
Its real value is that it can give the business one operational record. Sales, purchases, inventory, projects, employees and accounts can follow connected workflows instead of living in separate files. Approvals are recorded. Transactions have owners. Documents can be traced. Management reports come from the same activity that runs the business.
Once that foundation exists, AI becomes much more useful. It can review a live sales pipeline and identify opportunities that have gone quiet. It can prepare a summary of overdue customer invoices. It can flag unusual stock movements, extract information from supplier documents, draft project updates or help an employee find the correct internal procedure.
The difference is context. The AI is no longer guessing from an isolated prompt. It is working with the records, rules and events that the business already uses.
The most valuable AI in a business may not be the one employees chat with. It may be the one quietly notices that something important has not happened.
Do not use AI for everything
There is also a tendency to use AI for tasks that ordinary automation handles better. That is often unnecessary and sometimes dangerous.
If a purchase above a certain amount must be approved by a director, that is a rule. Configure the rule. If an invoice becomes overdue after thirty days, the system does not need to think about it. Trigger the reminder. If a customer discount cannot exceed an agreed limit, enforce the limit in the workflow.
AI becomes useful when the input is less predictable. It can interpret an email, classify a document, summarize a long discussion, compare written requirements, suggest a response or identify a pattern that would be difficult to express as a fixed rule.
A sensible design uses both. Traditional automation handles the predictable steps. AI handles language, ambiguity and judgement, with a human involved where the consequence is material.
AI agents raise the stakes
The conversation is now moving beyond assistants that answer questions. AI agents can monitor events, use business applications and take actions. That is powerful. It also changes the risk.
An inaccurate answer from a chatbot is inconvenient. An agent with broad access can change records, contact customers or trigger a transaction before anyone notices the mistake. Giving an agent an administrator account simply because it is easy is the digital equivalent of handing every office key to a new intern on the first morning.
Businesses should start with the least access required. Read-only access is often enough for early use cases. Actions should run through dedicated accounts, clear permission boundaries and logs. High-impact steps should require approval. Credentials must be revocable, and the business should know exactly what happens when access is removed.
These controls do not make AI less useful. They make it usable in the real world.
A practical African business opportunity
For African businesses, the opportunity is significant, but it is not found in copying every AI trend from Silicon Valley. It is found in solving the operational problems that consume time every day.
A distributor may need better demand visibility. A construction company may need tighter control of project costs and materials. A manufacturer may need to connect production, inventory and quality records. A service business may need to stop losing commitments across email, meetings and messaging apps.
In each case, AI can help. But the value comes from connecting it to the actual work of the business. A clever assistant that sits outside the operating system will always have a partial view. A well-designed automation that understands the transaction, the customer, the approval and the next expected action can change how work gets done.
This also gives growing companies a chance to avoid some of the complexity built up by older businesses. They can standardize earlier, centralize their data and add AI to a cleaner foundation. The aim should not be to appear advanced. It should be to run a better business.
Business Automation With AI
At ABN Consulting Group, we use the phrase Business Automation With AI, or BAWA, to describe this approach. The important word is not AI. It is business.
We begin with the outcome: reduce the time needed to close the month, improve sales follow-up, control stock losses, shorten customer response time or give managers a clearer view of project performance. We then look at the process, the data and the existing systems. Predictable steps are automated first. AI is added where it can interpret, summarize, recommend or coordinate better than a fixed rule.
The final part is accountability. We decide what the AI may see, what it may do, what needs human approval and how its actions will be reviewed. That may sound less exciting than launching an all-purpose agent. It is also far more likely to produce a system that people trust and continue using.
Where a business can start
Start with one workflow that is frequent, frustrating and measurable. It might be following up quotations, processing supplier invoices, preparing weekly project updates or answering repeated employee questions.
Write down how the work happens today. Identify where information enters, who makes each decision, where delays occur and which system holds the final record. Fix obvious gaps before adding AI. If five people use five different product names, standardize the names. If no one owns the approval, assign an owner.
Automate the parts that follow clear rules. Then use AI for the parts that require reading, interpretation or a useful first draft. Keep a person in the loop until the results are dependable. Measure time saved, errors reduced, response time improved or revenue recovered. If the result cannot be measured, the project is probably still too vague.
After that first workflow works, move to the next one. A collection of small, reliable improvements will create more value than one impressive AI project that never becomes part of daily work.
AI is a multiplier
AI will become part of ordinary business software. In many cases, employees will use it without thinking of it as a separate tool. That makes it even more important to build on the right foundation.
AI multiplies what is already present. When a business has clear processes, useful data and responsible controls, it can multiply speed, insight and consistency. When the foundation is weak, it can multiply confusion and risk.
There is no award for having the most chatbots, the largest model or the most ambitious agent. The real advantage belongs to the business that uses technology to solve a real problem, in a way its people can understand, trust and improve.
AI is not the business strategy. The strategy is deciding what kind of business you want to run, then using AI where it genuinely helps you run it better.