In July 2026, Indonesia's Minister of Micro, Small and Medium Enterprises, Maman Abdurrahman, said that more than 90 percent of the country's small business owners had not yet put AI to use. That figure is often read as a sign of being behind. We read it differently. In our experience, owners who try AI and then drop it usually drop it because of how they started, not because the AI was not smart enough.
Five habits look perfectly normal, and almost everyone has them. Each feels like progress. Each is also a fast way to waste money and team goodwill.
1. Buying the tool first and finding the job later
The subscription gets paid, accounts get created, and everyone is told to "have a play". Two weeks later one person still uses it and the rest are back to the old way. The tool is not the problem. What is missing is a reason to use it: no specific job that has to become faster or tidier.
Flip the order. Write down one job that makes you or your team complain regularly, then look for a tool that fits it.
2. Starting with the biggest job
It is tempting to point AI at the most expensive work: financial reports, quotes for large customers, replies to sensitive complaints. That is exactly where mistakes cost the most, and you do not yet know how far the tool can be trusted.
Start with work that repeats often and costs little when it goes wrong. Drafting replies to the same few questions ("is it in stock?", "what time do you open?"), tidying meeting notes, or grouping customer reviews. Two questions pick the job: how often does it come up, and how expensive is a wrong result.
3. Not recording the starting point
Without baseline numbers, any result just feels "like it is probably faster". Two numbers, recorded the week before you start, are enough: how long the job takes each time, and how many mistakes or redos happen per week. A sheet of paper or one spreadsheet column will do.
Without those two numbers you cannot decide whether to continue or stop. The decision ends up following the taste of whoever in the team is most enthusiastic.
4. Putting customer data into a personal free account
A staff member in a hurry will often paste a customer list, phone numbers, or a sales summary into an AI app on a personal account. The intention is good. But you do not know where that data is stored or what it is used for, and you cannot pull it back.
Customer data is your responsibility, not the app provider's. A simple rule: use one account in the business's name, and write down which data must never go in, such as ID numbers, card details, and customer bank accounts. One page is enough, as long as everyone reads it.
5. Granting full access from day one
Letting AI send messages to customers, change prices, or edit stock directly feels efficient. Until the day the tool answers a complaint in the wrong tone to many customers at once, and the mistake is already sent before anyone had a chance to read it.
Access should grow slowly, following the evidence. Start with an AI that only reads, then one that drafts and is always reviewed by a person. Pressing send stays a human job until you have months of records showing the results are consistent.
One person who owns it
One more thing often gets missed: who owns the trial. If everyone is allowed to experiment and nobody is responsible, the results stay scattered in people's heads and never become a decision. Pick one person. Not necessarily the most tech-savvy one, but the one who understands the job best. They record the two numbers, review the AI's output, and report to you at the end of week four. Someone who does that job every day will spot a strange draft faster than any IT person.
Four weeks, one job
With those five traps avoided, the plan is simple. In week one you pick one job and record the two numbers. In week two the AI produces its version while the team keeps working as usual, so you can compare the two without risk. In week three the AI's output is used once one person has checked it. In week four you recount and decide.
The decision rule is simple too. If the time per job drops and the person reviewing is not correcting more often than they would simply do the work themselves, move on to a second job. If not, stop. You have lost one month of subscription, not one year, and you know exactly why.
Our view: AI in a small business does not need a grand strategy. It needs one job, two numbers, and one access limit. A business with all three knows within a month whether AI is useful to it. A business without them will keep paying for a subscription and guessing.
If you would like help choosing that first job, tell us about your business on WhatsApp at +62 823 1499 5005.