Where AI actually helps in your business
Every piece of software now carries "AI-powered" on its landing page. Some of it means something real; a lot of it means nothing at all. This lesson won't explain how AI works under the hood. It gives you a practical test you can apply yourself to any pitch that reaches you, before you spend a single riyal.
The basic rule
AI is not a clever employee who understands your business and makes its decisions for you. It is a tool that does one thing well: it takes large amounts of messy text and data and turns them into something organised and usable. Every successful use inside a business traces back to that sentence, and every failed one ignored it.
The three-condition test
Before approving any AI use in your operation, ask three questions. If the answer is yes to all three, the idea is worth trying. If even one fails, postpone it:
- Is the task repetitive? If it happens once a year, doing it by hand is cheaper and faster than setting up a tool and maintaining it.
- Are its inputs text or data you already hold? The tool does not invent information you don't have, and it won't make up for incomplete records or disorganised books.
- Does a human review the output before it has consequences? If the output goes straight to a customer or into an accounting entry unreviewed, the risk outweighs the benefit.
A concrete example: routing customer messages
Imagine a shop that receives dozens of messages a day across scattered channels: a price enquiry, a complaint about a late shipment, a request for a tax invoice, an offer from a supplier. One employee reads each message and routes it to the right department, and an hour of their morning disappears into sorting alone.
This task passes all three conditions: it repeats daily, its input is text you already hold, and an employee reviews the routing before replying. The gain here isn't "intelligence" — it's an hour handed back to that employee every morning, and fewer messages lost in the noise.
Now compare it with another idea you'll hear often: "let AI set your product prices." That one fails the third condition, because the decision touches revenue directly, mistakes are expensive, and their effect only shows up long after the fact.
Genuinely useful vs. noise
| Use case | Verdict | Why |
|---|---|---|
| Reading supplier invoices and turning them into data ready for review | Useful | Repetitive, inputs are documents you hold, and the accountant reviews before approving |
| Summarising a long meeting into decisions and owners | Useful | Saves time, and anyone who attended spots an error immediately |
| Drafting a reply to a recurring enquiry | Useful, conditionally | A draft an employee reviews — not a reply sent automatically |
| Approving a payment with no human in the loop | Noise | That is an accountability decision, not a processing task |
| "Forecasting" next year's sales from two months of data | Noise | The data isn't there, and the tool won't create it |
The accountability question
For every use you consider, answer one question before you proceed: who carries the responsibility if the tool gets it wrong? If the answer isn't a named employee with a clear role, the use isn't ready no matter how appealing it looks. The tool carries no responsibility, and it is not an excuse a customer or a regulator will accept.
The tool suggests, the employee decides, the business is accountable. That order doesn't change no matter how good the tools get.
How to start in practice
Don't start an "AI project." Start with one annoying task your team genuinely complains about, run it for a full month, and measure two things only: the time saved per week, and how often the output needed correcting. If both improve, extend the same approach to a second task. If they don't, stop the experiment without hesitation and without hunting for a justification.
Quick checklist
- The task is repetitive, not exceptional.
- The data you need exists and is organised enough.
- There is human review before any financial or contractual effect.
- There is a named person accountable for the output.
- There is an agreed measure of success and a fixed trial period.