A proposal for an AI project lands on your desk. It promises efficiency, mentions a large language model, and has a number at the bottom. You are not a technologist. How do you decide?
The surveys suggest many leaders are in the same position. Research commissioned by AWS in 2025, covering 1,000 Malaysian businesses, found that 27% had adopted AI, up from 20% a year before, and that 73% of those adopters remained at what it called basic levels of integration. More than half said a shortage of digital and AI skills was holding them back. A separate survey of over a thousand MSMEs reported that 60% named a lack of in-house technical expertise as their biggest obstacle, ahead of cost.
Put plainly: plenty of companies have bought AI. Far fewer have changed how the work gets done.
These are the questions we would ask in your seat. None of them requires technical knowledge.
1. Which task, done by whom, takes how long today?
A good proposal names a specific task: “typing supplier invoices into the system, done by two clerks, about 25 hours a week”. A weak one talks about “leveraging AI across operations”.
If the task cannot be named and timed, the benefit cannot be measured, and you will not know afterwards whether it worked.
2. What does a correct result look like, and who decides?
Before anything is built, there should be an agreed test. Take fifty real documents. What must the system get right on them, and how many is enough?
Agree the pass mark before you see the results. It is very hard to be objective afterwards.
3. What happens when it is wrong?
It will be wrong sometimes. The question is what happens next.
In a sound design, an uncertain result is flagged and goes to a person. In a poor one, the system produces a confident wrong answer and nobody notices until a customer does. Ask to be shown the screen where a staff member reviews and corrects the output. If that screen does not exist, the project is not ready.
4. Who in my company owns it?
Not the vendor. Someone on your payroll who is responsible for the result, has the authority to change the process, and will be there in a year. Projects without an internal owner stop when the vendor’s attention moves on.
5. Where does our data go?
When the system uses a hosted AI model, your documents are sent to another company’s servers. Ask three things. Which provider? In which country? Is our data used to train their models?
These are no longer optional questions. The amended Personal Data Protection Act carries mandatory breach notification, and your customers and auditors will ask you the same thing.
6. What does it cost to run, every month?
The build price is one number. Ask for the other two: the monthly running cost, including AI usage, hosting and support, and how that cost grows if volume doubles. AI usage is billed by consumption, and a system that is cheap in a pilot can be expensive at scale.
Ask who holds the cloud account too. It should be your company.
7. Can I see it working before I commit?
This is the question that removes most of the risk.
A slide deck proves nothing. A working demo, on your documents, that your own staff can click through, proves a great deal: that the vendor understood the problem, that the approach works on your real material, and that your people can use it.
This is how we work with every client. We learn the business, then build a working demo before we send a proposal or a quotation. It costs us effort up front, and it means nobody signs a contract for something they have only read about.
How to read the answers
If the answers are specific, with a named task, an agreed test, a visible review step, a named owner and a demo you can use, the project is probably sound even if you do not follow the technology.
If the answers are general, the project is not ready, however impressive the technology sounds. Send it back with the seven questions. A good team will thank you for them.