Should you hire an AI person?
Most firms this size ask the question the wrong way round. It is not "can we afford an AI hire". It is "do we have a whole job here".
That is the question that decides it, and it is answerable before you write a job description.
What the hire actually costs
A capable AI lead in Singapore or Australia is S$150,000 to S$250,000 a year once employer costs are in. Call it S$15,000 a month at the midpoint.
Then add what nobody budgets. Roughly four months to find one, during which nothing happens. Recruitment fees if you use an agency. Equipment and tooling. And senior time managing someone whose work you cannot evaluate, which is the hidden cost and the largest one.
Against that, a retainer that ships something new every month runs S$6,000 to S$15,000, and a fractional AI engineer bills US$6,000 to US$18,000 a month for eight to twenty hours a week. So the honest comparison is not cheap versus expensive. Every option in this category costs real money. They differ in what happens afterwards.
The failure mode nobody warns you about
You hire someone good. They ship two useful things in the first quarter. Then the pipeline of obvious AI work runs out, because at a fifty-person firm there is a finite amount of it, and they are now underemployed in the role you hired them for.
So they start absorbing adjacent work. The reporting nobody owns. The integration that keeps breaking. Onboarding the new starters onto the tools. Within a year you have an expensive generalist doing IT, and they know it, and they leave.
This is the normal outcome under roughly a hundred people. Not because the hire was wrong, but because the job was half a job and nobody said so at the start.
The other failure mode: it all sits with one person
Whether you hire or not, this one applies.
Firms usually appoint the keen one. They get good, and every AI capability the firm has now lives in a single employee. Then they resign, or move teams, or get busy, and it leaves with them. That is not a risk, it is the base case, because keen people with new skills get offers.
Four people is the smallest number where the knowledge stops being a person and starts being a practice. It is also the smallest group that argues about the right approach, which is where most of the real learning happens.
If you do hire, hire on the condition that their job includes teaching three colleagues. If that is not in the role, you have bought a dependency with a notice period.
When hiring is genuinely right
Three conditions, and you want all three:
You have more than a hundred people, or enough process complexity that the work does not run out.
Somebody senior can evaluate the work. If nobody in the building can tell good from bad here, you cannot manage the role, you can only hope.
The work is continuous, not a transition. Some firms need capability built once and then run. That is a project shape, and projects are badly served by permanent hires.
If all three are true, hire, and stop reading. A good person in the building every day beats any outside arrangement.
When it is not
If you are between twenty and a hundred people, the work is a transition rather than a permanent function, and nobody internally can yet judge the output, a hire is the expensive way to find that out.
The cheaper path is to get one real process from manual to running, with your own people involved, and see what the shape of the work actually is. Six months of that tells you whether there is a whole job here, and by then you have people who can interview for it properly.
What we would say on a call
We sell an alternative to this hire, so read that into the above as you like. But the argument we make is not that you should never hire. It is that most firms this size are not yet in a position to write the job description, and hiring to find out is a S$200,000 experiment with a four-month lead time.
Find out first. Then hire, if the answer says so.