Where to start with AI automation in a Somali business
How to pick the first process worth automating — and how to tell whether it worked.
Most businesses approach AI the wrong way round. They start from the technology and look for somewhere to put it, which produces a demo nobody uses. The useful order is the opposite: find the work your team repeats every day, then ask whether a machine can do the boring half of it.
Start where the work repeats
Spend a week noticing what gets done more than once a day with no judgement involved. Re-typing a supplier invoice into the accounts. Answering the same question about price and delivery on WhatsApp. Chasing the same five customers for payment. None of that needs a person — it needs a person to check it.
What makes a process a good first candidate
- It happens often — daily or weekly, not twice a year. Rare work is rarely worth automating.
- The rules are writable. If you can explain the decision to a new employee in two minutes, software can follow it.
- The inputs already exist somewhere digital, or arrive as documents you can photograph.
- A mistake is visible and recoverable. Start where a wrong answer is caught the same day, not where it quietly compounds.
- Someone owns it. Automation with no owner gets switched off the first week it surprises anybody.
Decide what "it worked" means before you build
Write the number down first: hours a week, days to get paid, questions answered without a staff member, invoices posted without re-typing. If you cannot name the number, you will not be able to tell six weeks later whether the thing earned its keep — and you will keep it running out of politeness.
Automate one process properly and the second one is obvious. Automate ten at once and none of them is trusted.
Keep a person in the loop at the start
The first version should draft, not send. Let it prepare the invoice, the reply, the reminder — and have someone approve it for a few weeks. You will find the edge cases quickly, your team will trust the output because they watched it get things right, and you can loosen the leash once the error rate is boring.
Ask where your data goes
Before anything touches customer records, prices or patient information, get a straight answer about where it is processed, what is retained, and who can see it. A system that cannot answer that is not ready for your business, however good the demo looks.
If you would rather not run this evaluation yourself, that is reasonable — it is the kind of thing worth walking through with someone who has done it before.
