We lost a release because three engineers were away in the same two-week window, and the part that still bothers me is that every one of those absences had been known about for months. Nobody had hidden anything. The dates existed, in three separate inboxes, and never once in the same place as our delivery plan. I treated National Service as something that happened to me rather than something I could read off a calendar, and it cost us a quarter.
This is the plan I run now. It is six steps, most of the work is annual rather than per-absence, and the first step costs an afternoon. It is written for an engineering manager or founder in Singapore with a team where several people are operationally ready national servicemen, which in practice is most local engineering teams.
The System Gives You More Warning Than Almost Any Other Absence
This is the part I had wrong for two years. I had filed call-ups mentally alongside sick leave, as something inherently unpredictable. The notice structure says otherwise.
Published guidance for employers and servicemen sets out notice periods by type of activity. High-key in-camp training, meaning seven days or more, is generally announced around six months ahead. Low-key in-camp training of under seven days gets around three months. Make-up training and other call-ups carry roughly one month, and reporting notes that servicemen who have deferred may get only about a month for their make-up training.
Read that ordering again, because it is the opposite of what intuition suggests: the longest and most disruptive absences come with the most notice. A two-week absence that lands as a surprise in your sprint planning was knowable roughly six months earlier. That is not a scheduling problem, it is an information routing problem inside your own team, and it is cheap to fix.
| Activity | Typical notice | What it does to a sprint | Your lever |
|---|---|---|---|
| High-key in-camp training (7 days or more) | About 6 months | Removes a person for one to three sprints | Full planning: ownership transfer, scope cut, no deferment needed |
| Low-key in-camp training (under 7 days) | About 3 months | Removes a person for part of one sprint | Handover protocol and on-call swap |
| Make-up training and other call-ups | About 1 month | Short notice, harder to absorb | Standing buffer rather than per-event planning |
| Annual ceiling per person | Up to 40 days in a financial year | Exposure limit, not a forecast | Use for worst-case risk on a single owner |
One caveat on all of the above. Notice conventions, schemes and payroll mechanics get updated, and summaries age badly. Treat this table as the reason to go and check the current position on the MINDEF releases and in the employer material, not as the authority itself.
π‘ Our Expert Take
The six-month figure is the most underused number in Singapore engineering management. Six months is enough time to do something no other absence lets you do: deliberately transfer ownership of a system while its owner is still sitting next to the person taking it over. Parental leave gives you a comparable window and most teams use it well. Call-ups give you the same window several times a year across several people, and almost nobody uses it at all, because the dates arrive in a personal inbox rather than in a planning tool.
Step 1. Build a Forward Call-Up Register
One shared place, forward-looking, holding every known call-up date for the team. That is the whole artefact.
How to populate it without making anyone uncomfortable. Say plainly why you want it: so that cover is planned in advance and nobody is asked to choose between a call-up and a release. Ask for dates and duration only. You do not need unit, vocation, activity type or location, and asking for them changes the character of the request. Make sharing the norm by putting the dates straight into the team calendar as ordinary unavailability, in the same way as annual leave.
Two practical notes. Ask people to forward dates when they receive the notice, not when the date approaches, because the entire value of the register is the lead time. And ask the team to keep their own employment details current in the MHA NS Portal under Manage my profile, since that is what makes the payroll side in step two work smoothly rather than turning into a correction exercise later.
Then do the thing that makes the register real: put it next to your roadmap. A register that lives in a human resources folder nobody opens during planning has not solved the problem I had. It needs to be visible in the same conversation where quarterly commitments are made.
Step 2. Sort Out Payroll, DIRECT and CPF Once
Get this done once, properly, and it never costs you attention again. Leave it unresolved and it generates a small administrative fire around every absence.
The mechanics, as published. Make-up pay covers the difference between civilian income and service pay. Employers registered under the DIRECT scheme keep paying wages on their normal payroll schedule, as if the employee had not been away, and claim reimbursement from MINDEF afterwards. Where claims are auto-generated from CPF data, MINDEF indicates that make-up pay and service pay are paid within ten working days from the start of the activity. CPF contributions remain payable on total wages including make-up pay, that obligation sits with the employer whether or not they are on DIRECT, and it is not reimbursable.
For an engineering team, DIRECT is almost always the right answer, for a reason that has nothing to do with finance. It makes the absence invisible in the engineer pay packet. An engineer who sees no change in their salary during a call-up does not spend the week before it worrying about money, and does not come back to a payroll query. Anything that makes a predictable absence boring is worth the registration effort.
Settle three questions in writing while you are in there: who files the claims and by when, whether the engineer needs to do anything at all, and what happens for someone whose pay includes a variable component. Then put it in your onboarding material so the next hire never has to ask.
Step 3. Convert Call-Up Days into Quarterly Capacity
Annual figures hide the problem. Call-ups do not arrive evenly, and the quarter is where the damage shows up.
Take your register and total the known call-up days per quarter, not per year. Convert to engineer days, then into whatever unit your planning actually uses. Then subtract it from the commitment before the quarter is planned, not as a footnote afterwards.
An example from our own numbers, to show the shape rather than to be copied. A team of seven, five of them operationally ready, with the register showing one high-key training and two low-key trainings in the same quarter, came to roughly 24 engineer days. That is close to five engineer weeks out of a quarter, concentrated in two months rather than spread. Spread across a year it looks like a rounding error. Landing inside one quarter, against a fixed commitment, it is the difference between shipping and not.
Add a modest standing buffer for the short-notice category rather than trying to plan each one. Make-up training and other call-ups arrive with about a month of warning, which is too short for ownership transfer but long enough to swap an on-call shift. A small buffer absorbs them without a replan. Our engineering backfill plan uses the same principle of holding a declared buffer instead of pretending capacity is fully committed.
Step 4. Remove Single Points of Failure During the Notice Period
Here is where the six-month notice earns its keep, and where almost nobody collects.
List the systems where exactly one person can deploy it, debug it at 2am, or explain why it was built that way. Every team has three or four. Cross-reference that list against your call-up register. Where a single owner has a high-key training coming, you have been handed months of warning to fix a risk you were carrying anyway.
What to actually do with the time, in rough order of value:
- Pair on the next real incident or change in that system, with the second person driving and the owner watching. Simulated handovers teach much less than a real one.
- Write the decisions down, not the steps. Runbooks go stale; the reasoning does not. Recording why a system works the way it does is what lets someone else make a judgement call in the owner absence, and it is the use case our architecture decision records method is built for.
- Transfer the credentials and access before the week of departure, and have the second person use them once while the owner is still there.
- Rename the owner in your documentation for the period, so escalation does not route to someone unreachable.
Notice that none of this is cover work. It is structural risk reduction that you would want regardless of National Service, with a deadline attached that makes it actually happen. We treat it as the same exercise as reducing the bus factor, and the call-up register is simply the most reliable forcing function available for it in Singapore.
Short a person rather than short a plan?
Sometimes step three produces a number that no amount of handover discipline fixes, and the answer is an extra pair of hands for a quarter rather than a permanent hire. HireDeveloper.sg sources pre-vetted engineers for Singapore employers and we are straightforward about which option fits: a contract engineer for a defined window, a permanent hire, or cutting scope. If cutting scope is the right answer we will say so.
Get startedStep 5. Write the Handover Protocol, and Stop Defaulting to Deferment
Standardise the absence so it stops being a negotiation. One page, three phases.
The week before: work in progress either merged or explicitly parked with a note on its state, ownership of anything live reassigned by name, on-call shifts swapped and the rota republished, and the register and team calendar updated. During: a named first point of contact per system, an explicit statement that the person is not expected to be reachable, and permission for the team to make reversible decisions without them. On return: a short written catch-up rather than a two-hour meeting, and no on-call shift on the first day back.
That middle point matters more than it looks. If the team believes they should check with the absent person before deciding anything, you get an absence that blocks work and an engineer answering messages during training. Writing down that reversible decisions proceed without them is what makes the cover plan real.
Now the part that is genuinely a judgement call. Deferment exists for critical work commitments, and there are quarters where using it is right. The failure is making it the default, and it fails for a specific mechanical reason: deferment moves the absence rather than removing it, and the replacement make-up training typically arrives with around a month of notice instead of six. You have taken your most critical engineer and converted their predictable two-week absence into an unpredictable one, which is a strictly worse position. A team that defers every call-up has not solved a capacity problem, it has traded away its own lead time.
π‘ Our Expert Take
There is a retention dimension here that rarely gets discussed openly. Engineers notice how an employer handles a call-up. A manager who treats it as an unwelcome interruption, asks whether it can be pushed back, and makes the person feel responsible for the gap, is remembered for it. A manager who already has the dates, has arranged cover and says have a good one, is also remembered. Neither costs money. Over several years across a local engineering team these small moments compound into something that shows up in your attrition numbers, and no compensation review repairs it afterwards.
Step 6. Run the Return, Then Review Once a Year
Two weeks away from a codebase that several people have been changing is a real re-entry cost, and most teams spend it badly by handing the returner a backlog and an incident on the same morning.
What works: a written summary of what changed and what broke, waiting for them rather than delivered live. No on-call on the first day. One owned task on the first day rather than four, chosen so that getting back into the code is the point. And a deliberate check on whether the ownership you transferred in step four should transfer back or stay where it went, because quite often the second person has grown into it and the right answer is to leave it with them.
Then the annual piece, which takes an hour. Once a year, compare what the register predicted against what actually happened: total days, which quarters were hit, which handovers held and which did not, and whether anything was deferred and why. Write the numbers down. The next year plan should be built on measurement rather than on how the last bad quarter felt.
One last note on fairness, since it comes up in mixed teams. The cover work in step four should not land repeatedly on the same two colleagues. Rotate who takes on transferred ownership, and record it, or you will have solved a capacity problem by quietly creating a resentment one. The mechanics of planning around a known absence are much the same as for parental leave cover, with the difference that call-ups recur, which means an unfair pattern has many chances to repeat.
π‘ Our Expert Take
If you only do one of these six steps, do the register. Everything else is downstream of knowing the dates, and the honest reason most teams do not have one is not resistance from engineers, it is that nobody ever asked. In my experience people share the dates immediately once they understand it is for planning cover rather than for judging availability. The whole thing is one afternoon of asking, and it converts the most predictable absence in Singapore engineering from a recurring surprise into a line in the capacity plan.
FAQ: National Service Cover for Singapore Engineering Teams
How much notice does an employer get for in-camp training?
High-key in-camp training of seven days or more is generally announced about six months ahead. Low-key training of under seven days gets about three months. Make-up training and other call-ups carry roughly one month, and servicemen who have deferred may receive only about a month for make-up training. The useful consequence is that the longest absences come with the most warning, so being surprised by a two-week absence is an information routing problem inside your own team rather than a scheduling constraint. Check the current position in MINDEF employer material for your own planning.
How many call-up days per year should I plan for?
MINDEF material states that operationally ready national servicemen may be required to attend call-up activities up to a maximum of 40 days in a financial year. Most people are called up for considerably less in a typical year, so plan against the dates your team actually holds rather than against the ceiling. Keep 40 in mind for one purpose only: as the worst-case exposure on a system that has a single owner, which is the scenario worth engineering around.
Does the employer keep paying salary, and what about CPF?
Make-up pay covers the difference between civilian income and service pay. Under the DIRECT scheme the employer continues paying wages on the normal schedule as if the person had not been away, then claims reimbursement from MINDEF; where claims are auto-generated, MINDEF indicates payment within ten working days from the start of the activity. CPF remains payable on total wages including make-up pay, stays the employer responsibility whether or not they are on DIRECT, and is not reimbursable. Confirm the mechanics with MINDEF and CPF against your own payroll setup before relying on them.
Should we ask an engineer to defer training for a release?
Sparingly. Deferment exists for critical work commitments and there are quarters where it is the right call. Making it routine backfires mechanically: the absence moves rather than disappears, and the replacement make-up training usually arrives with about a month of notice instead of six. You convert a predictable two-week gap into an unpredictable one on your most critical engineer. There is also a retention cost, because engineers notice which managers treat a call-up as an interruption and which ones have already arranged cover.
Build the register, then hire against what is left
Steps one and three take an afternoon between them, and they tell you whether you have a planning problem or a staffing one. If it turns out to be staffing, we source pre-vetted engineers for Singapore employers, including engineers for a defined window rather than a permanent seat. We will also tell you when the cheaper answer is to transfer ownership of one system and leave the headcount alone.
Lets get started