The Department of Statistics Malaysia counts just over a million micro, small and medium enterprises: 96.1% of all business establishments in 2024. They generated RM652.4 billion that year, 39.5% of GDP, and accounted for 48.7% of employment. In 2025 their value added rose to RM689.8 billion, 39.7% of GDP, growing faster than the economy as a whole.
National AI plans are written about data centres and talent pipelines. Whether AI changes Malaysian productivity will be decided in these million businesses, most of which have no IT department.
If you run one of them, here is a way to approach it that does not require a strategy document.
Step 1: Find the job, not the technology
Do not start with “how do we use AI”. Start with a list of tasks that are done by hand, repeatedly, from a document or a message. Typical examples from businesses we have worked with:
- Typing details from a customer’s PDF or photo into your system
- Copying rows from a spreadsheet into a Word template
- Preparing the same report or certificate every week
- Re-entering sales into the accounting system
- Answering the same five customer questions
For each one, write down how many hours a week it takes and what happens when it is done wrong. Multiply the hours by a loaded hourly cost. That is your budget ceiling.
We publish a free automation readiness assessment that walks through exactly this, including the arithmetic.
Step 2: Decide whether it is an AI job at all
Many of these tasks do not need AI. They need ordinary software.
- Rule-based and repetitive (copy this field there, compute this total, generate this document): use automation. It is cheaper, faster and always right.
- Reading unstructured input (PDFs, photos, emails, handwriting): this is where AI is useful, because rules cannot cope with variety.
- Drafting text for a person to finish (replies, descriptions, summaries): AI helps, with a person sending the final version.
A law firm we work with cut filing and checking time by 95%. Almost none of that was AI. It was a spreadsheet upload and template-driven documents. The AI in that system does one small job: reading figures from statements.
Step 3: Keep a person on the output
Any AI step should end with a human check before it reaches a customer, a regulator or your accounts. Design the check so it is quick: show the original document next to what the AI extracted, and let the person correct it in place. In our own systems, review is where we spend most of the design effort.
Step 4: Start with a tool, graduate to a system
For a first experiment, an off-the-shelf AI tool and a careful staff member is enough. You will learn what the model gets wrong on your documents.
Move to something built for you when one of these becomes true:
- The task happens hundreds of times a month
- The output has to land in another system
- You need a record of who approved what
- The data is confidential and should not be pasted into a public tool
Step 5: Mind three risks
Data. Know what you are sending to an AI provider. Customer personal data is covered by the Personal Data Protection Act, which now carries mandatory breach notification.
Credentials. If a developer sets up cloud services for you, ask how access is controlled. A leaked cloud key is the most common way small companies get an unpleasant bill. We wrote up a real case.
Dependence. Make sure you own the account, the data and the code. You should be able to change supplier.
What good looks like
A small manufacturer whose certificates generate themselves. A clinic whose forms are typed by software and checked by a nurse in seconds. A distributor whose invoices appear in accounting without anyone re-keying them.
None of those is an “AI transformation”. Each is one repetitive job, fixed properly. Do that three times and you have transformed the business anyway.