A business owner hears enough about AI to feel behind. They try something. For a week it is genuinely impressive. By the second week it is used occasionally. By the second month nobody opens it, and the conclusion forms quietly: interesting, not for us.
This happens constantly, and the diagnosis is almost never the technology. It is that the tool was pointed at the wrong task, chosen because it was the most impressive demonstration rather than the most repeated chore.
The three questions
Before considering any AI tool, test the task against three questions. Work that fails any one of them will not stick, no matter how good the tool is.
1. Do we do this repeatedly?
AI pays off on volume. Answering the same customer question forty times a week, writing captions for daily posts, drafting the same category of message repeatedly. Something you do twice a year is not worth the effort of building a habit around, even if the output is excellent.
The instinct runs the other way. Owners reach for the hardest, rarest, most interesting problem, because that is where help feels most valuable. That is precisely the task where a new habit will never form.
2. Is "nearly right" useful?
This one decides most cases and is the one people skip.
For a first draft of a product description, nearly right is genuinely useful. You edit it in a minute and you are ahead. For the VAT figure on an invoice, nearly right is worse than useless, because it looks finished and someone will trust it.
Give AI the work where a good first draft saves time. Keep the work where being nearly right is indistinguishable from being wrong.
The rule of thumb: if checking the output takes as long as doing the work, the tool has not saved anything. It has moved the effort and added a risk.
3. Can we tell when it is wrong?
A caption that misses your tone is obvious immediately. A stock projection that is quietly 15 percent optimistic is not obvious for months, and by then decisions have been made on it.
Only hand over work where you would notice a bad answer quickly. That is not a limitation of current tools, it is basic operational sense: never automate a task whose failures are invisible.
Where this usually lands for a Ghanaian SME
Run those three filters and a fairly consistent shortlist appears.
- Answering repeated customer questions. High volume, obvious when wrong, and a good draft answer beats a slow perfect one. This is why we built CSBot for WhatsApp, where 91.8 percent of Ghanaian internet users already are.
- Writing social captions and product descriptions. Constant, low risk, easy to correct. A human still approves, but starts from something rather than nothing.
- Turning what you know into published answers. Most owners carry the answers to their customers' questions in their head. Getting them written and structured is tedious, repetitive work, and it is exactly what makes a business citable by search engines and AI assistants.
- First-pass summarising. Long documents, message threads, meeting notes. You verify anything important, but you skip the reading.
And what it consistently rules out: anything that touches money without review, anything final-facing with no human approval step, and anything where the business could not detect a wrong answer within a day.
The part that decides whether it works
The tool matters far less than what you feed it.
An AI writing your customer replies with no knowledge of your prices, delivery areas, opening hours or refund policy will produce confident, well-written, wrong answers. Which is worse than no automation at all, because a customer acts on it.
This is why the unglamorous work comes first: write down what your business actually does. Products, prices, terms, hours, coverage, the answers to your ten most common questions. That document is what turns a general tool into something that can speak for your business.
It has a second payoff, and over time the larger one. The same structured information is what search engines and AI assistants read when a customer asks them for a recommendation in your category. You are not writing it twice for two purposes. You are writing it once, and it serves both.
How to start without wasting a month
- Pick one task that passes all three questions. One. The most repeated, lowest risk thing on your list.
- Write the reference document that task needs before touching any tool. Prices, policies, common questions, real answers.
- Run it alongside the current way for two weeks, not instead of it. You are testing whether the output is trustworthy, and you cannot learn that with live customers as the experiment.
- Measure one number. Time spent, response time, posts published. If it has not moved after two weeks, the task was wrong. Change the task, not the tool.
Most businesses that fail at this do so by starting at step three with an ambitious task and no reference document. The order is what makes it survive.
None of this requires understanding how the technology works. That is the standard we hold ourselves to: if a system needs its user to be technical in order to run their business, the system has failed, not the user.