The 36-month cliff: what new-graduate retention really costs you
In June 2026, there were 31,700 entry-level PMET vacancies were open in Singapore. That is not a market statistic to file away. It is a standing offer, addressed to the people you hired last year.
Ask any HR director in the region where early-career attrition bites and you will hear the same answer: not in the first year, and not in the fifth. It bites somewhere around the third. Long enough that the person has become useful. Late enough that you have stopped paying attention to them.
This is what we see in our own search work, and the data says the same. The question isn't how many people leave. It is what we did to make leaving the better option.
The first year is not the problem.
New joiners rarely leave in year one. Everything is still new: the systems, the people, the unspoken rules of how work actually gets done. There is a learning curve doing the retention work for you.
By month eighteen, that curve flattens. The person can do the job. What they cannot yet see is what comes after it.
By month thirty-six, they have their answer, and it is usually that nothing has changed. Same scope, same title, same manager, same five priorities. Meanwhile, the market has been talking to them for two years. Recruiters know their name. A competitor has already described a role that sounds like a promotion.
At that point, a counteroffer isn't a retention tool. It is a reconciliation: you are matching a number for someone who decided some time ago. By the time you are negotiating, the decision is already made, and you are paying for the delay.
What Japan's numbers teach us about our own
Japan publishes what most markets do not: it follows an entire graduating cohort for three years. Of the graduates who started work in March 2023, roughly one in three had left their first employer by March 2026.
The average is not the interesting part. The spread is. At employers with fewer than five staff, more than half had gone. At employers with a thousand or more, roughly one in four had gone. Same labour market, same cohort, same three years.
It's easy to read that as an argument for being a large employer. It is not. Size is a proxy for something else: whether there is a next role to move into. A small firm hires someone into a job. A large one hires them into a structure where a lateral move, a different team or a new manager is administratively possible. The difference in the separation rate is, more or less, the price of having nowhere to go.
Singapore does not publish an equivalent cohort series. We measure turnover as a flow, which tells you how many people left this month but not whether they were in month eight or year four. Those are different problems, and they need different fixes, and the aggregate number cannot tell them apart.
So we build the answer ourselves, and it is not difficult. Of everyone who joined in a given quarter, how many are still here at twelve months, at twenty-four, at thirty-six? Split by hiring manager, and by where they came from. Three lines on one chart. That is the whole instrument.
The manager decides more than the programme.
Onboarding is the visible part of retention. It has a budget, a checklist, a survey at day thirty. In most organisations, it is also not where the decision gets made.
The first manager is. Whether the work is legible, whether the person understands why it matters, whether a mistake is survivable: one person sets these in the first six months. A good onboarding programme under a manager who cannot hold a conversation about development is a coat of paint.
There is a simple diagnostic. Take your exits under three years and sort them by the manager people reported to. If a small number of managers account for a large share of them, you do not have an attrition problem. You have a management problem wearing an attrition costume, and no amount of engagement survey work will fix it.
What actually holds people
Three things, in order of effect.
A named next step. Not a promise of promotion, which costs credibility when it slips. A documented option: a lateral role, a certification with a date on it, ownership of something that was not theirs last quarter. The Japanese size gradient suggests that the option's existence matters more than its size. People stay where the future is legible.
Time to internal move, measured in days. Most organisations cannot answer how long it takes someone to move internally. It is usually longer than an external move, and everyone in the building knows it. If your internal process takes four months and a competitor's offer takes three weeks, you have made the decision for them.
Honest conversations before the exit interview. The exit interview is where you learn what you could have known at month twenty. Ask earlier, ask specifically, and ask the manager's manager when the answer sounds rehearsed.
Three things worth doing this quarter
Publish cohort survival, not just turnover. Twelve, twenty-four and thirty-six months, by intake, by manager, by source of hire. The first cut will be uncomfortable. That is what makes it useful.
Give every hire under three years a named next step. Not a promise. A documented option with a date and an accountable owner.
Audit the first manager, not the first month. Check whether your highest-risk cohort is concentrated under two or three managers. If it is, that is where the budget goes.
What this means in Asia
In Singapore, the constraint is the external bid. With a deep queue of open entry-level PMET roles, the market does the recruiting for you, and a counter-offer arrives after the fact. The lever is internal mobility: how fast someone can change roles without changing employers, who owns that process, and whether it is reported as a number every month.
In Hong Kong, the same logic applies, with less data. There is no cohort series to lean on, so your own probation and twelve-month figures are the only evidence you will get. In a pool where every competitor sees the same profiles, the retention question is not how to pay more. It is what this person can do here in eighteen months that they cannot do elsewhere.
In Japan, the hiring model itself is the retention policy. Large cohorts arrive at the same time and are placed into a first assignment at the same time, and that assignment sets the odds. Rotation, job allocation and the first manager's span of control are retention instruments, not administrative housekeeping.
The number to watch is not the resignation rate. It is the share of your 2024 intake still with you in 2027, and whether that share differs depending on who hired them.
Sources
Ministry of Manpower, Singapore, Labour Market Report, Second Quarter 2026, press release, 21 September 2026. Figure used: 31,700 entry-level PMET vacancies, 45.3% of total vacancies, June 2026. https://www.mom.gov.sg/newsroom/press-releases/2026/0921-labour-market-report---second-quarter-2026
Ministry of Health, Labour and Welfare, Japan, 新規学卒就職者の離職状況（令和5年3月卒業者）, published 23 October 2026. Figures used: university-graduate three-year separation rate 31.2%; by establishment size, 58.2% under five employees and 24.2% at 1,000 or more. https://www.mhlw.go.jp/stf/houdou/0000177553_00011.html
Both sources accessed 26 October 2026. Figures expressed as words ("one in three", "more than half") correspond to the published values above.
Comments