The Eighteen-Month Exit.

I started as a car painter. Twenty-two years later I work in IT portfolio and program management. That path was not luck. It came from people who invested in me before I had a track record and who showed me that staying long enough to grow into a role can matter more than chasing the next title.

Most development hubs do the opposite of what those people did for me. They recruit juniors at scale, train them well, fund the certifications and hand them real client work early. Then they lose a large share of them around the eighteen-month mark. The training is not the problem. The missing piece is perspective. A junior who cannot picture their next five years inside the company will take an external offer on simple arithmetic — a known gain elsewhere against an unknown future here.

The evidence backs this. Median tenure for workers aged 25 to 34 is 2.7 years, about half that of workers in their late fifties. When McKinsey asked more than 12,000 people why they left, lack of career development was the most cited reason, named by 41 percent, ahead of pay. Gallup found that 52 percent of departing employees said their manager or organisation could have done something to keep them. The exit window opens long before the resignation.

The usual response is a bigger cheque after the notice is in. That is the most expensive and least effective moment to act. The durable lever is cheaper and earlier. Real mentoring from people two or three steps ahead, who can describe the next decade and show that patience pays.

The front end works. Dev hubs — meaning large in-house centers and the offshore and nearshore delivery sites and global capability centers — are good at recruiting and onboarding. They run structured training, fund certifications and give juniors real client exposure early. Retention is where it falls apart.

The US Bureau of Labor Statistics put median tenure for workers aged 25 to 34 at 2.7 years in January 2024, against 6.7 years for workers aged 55 to 64. The Stack Overflow Developer Survey, large but self-selected, reports that newer developers now change jobs as often as peers ten years their senior.

• Workers aged 25–34 2.7
• Workers aged 55–64 6.7

Median years with the current employer, by age band (US Bureau of Labor Statistics, 2024).

The same pattern, every region

The pattern is not American and it is not only Indian. Across Europe and Asia the same dev hubs train juniors well and lose them early.

Europe's nearshore hubs show it. Mercer's Comptryx benchmark put European software-role turnover at about 12 percent through 2024. Romania's business-services sector reported an average tenure of roughly three years and a 16 percent turnover rate. In Portugal, a Landing.jobs survey of 1,809 tech professionals found them markedly more willing to change jobs than the year before, with salary, work-life balance and career progression as the top drivers.

Asia is sharper. Indian IT services hit a voluntary attrition peak of 27.7 percent in FY2022, with one Infosys executive capturing the loop exactly: "my attrition is somebody else's lateral and somebody else's attrition is my lateral." Vietnam shows the early exit in its rawest form — ITViec reports that 34 percent of voluntary resignations happen within the first six months and that a new offer carries an average raise of about 31 percent.

• Philippines, BPM (2022)31%
• India IT services (2022 peak)27.7%
• Romania, business services (2023)16%
• Europe, software roles (2024)12%
• Vietnam, local firms (2024 H1)9.6%

Voluntary or total attrition, latest available year. Mixed definitions and scopes, not strictly comparable. Sources: CCAP and WTW (2024), Infosys (2022), ABSL (2024), Mercer (2025), Talentnet-Mercer (2024).

The destination is identical everywhere. An external offer around the early-tenure mark. The trigger skews toward growth in higher-cost hubs and toward pay in lower-cost ones, but the maths is the same. A known gain elsewhere beats an unknown future at home.

The instinct is to assume it is money. The survey evidence says otherwise. McKinsey's Great Attrition work, drawn from more than 12,000 people across six countries, found that lack of career development and advancement was the single most cited reason for leaving, named by 41 percent, ahead of inadequate pay and uncaring leaders. The effect is strongest in exactly the group this paper is about: workers under 35.

Gallup reaches the same place from a different dataset. It reports that 87 percent of millennials rate growth and development as important in a job, that 93 percent left their company the last time they changed roles rather than moving internally, and that managers account for about 70 percent of the variance in team engagement. The Work Institute's 2020 report, built on exit interviews, ranked career development the top exit category at roughly 19 percent. Three independent sources, one conclusion. People leave because they cannot see themselves here in five years.

Developers are independent by nature. The work is continuous self-teaching and problem-solving, so the people who are good at it expect to direct their own time. Self-determination theory has held for decades that autonomy, competence and relatedness are what sustain motivation. Micromanagement attacks that directly.

The first tax is interruption. GitHub ran a two-week diary study of 40 software engineers and found that with few interruptions a developer had an 82 percent chance of a good day. Interrupted for most of the day, that chance fell to 7 percent. Meetings showed the same cliff. Going from two to three meetings cut the chance of making progress on goals from 74 percent to 14 percent.

• Few interruptions82%
• Interrupted most of the day7%

Chance of a developer reporting a good day, by interruption level (GitHub, 2021).

The second tax is the tooling. Microsoft Research surveyed 484 developers on their ideal week against their real one. Coding sat at roughly 11 percent of actual time against an ideal nearer 20 percent. The two activities that ate the difference were meetings and the management of tasks. Heavy project tooling — Jira being the usual name — turns a developer into a clerk of their own work. For a junior deciding whether to stay, that is a daily signal about how far the company trusts them to manage themselves.

There is a counter-model worth naming because it shows the problem is fixable. In Linear's workflow a pull request carries the issue from in-progress to done as it merges, so the developer never leaves the editor to update a board. The repository and its pull requests become the source of truth rather than a separate ticketing system. Shrink the management surface and keep developers close to the work. You remove a reason to leave that no counteroffer repairs.

When the external offer arrives, the junior compares a known gain against an uncertain future. Prospect theory sharpens it. Kahneman and Tversky showed that people underweight merely probable outcomes relative to outcomes received with certainty — the certainty effect. A guaranteed raise elsewhere beats a probable but unpromised future at home, even when the home path would pay more across several years.

There is a second cost most employers never see. The internal path is not only risky, but also ambiguous, because the junior does not know the odds of promotion or the size of a future raise. Ellsberg showed that people will pay to avoid unknown probabilities, not only unfavorable ones. Camerer and Weber confirmed that this ambiguity aversion is robust and economically real. Silence about the internal path is itself a cost the junior subtracts from staying.

The pull on the other side is genuine. Pew found that among workers who changed jobs between April 2021 and March 2022, 60 percent saw a real terms pay rise against 47 percent of those who stayed, and that the median switcher gained 9.7 percent in real terms while the median stayer lost 1.7 percent. A single offer in a hot specialism can run higher. The exact size is not the point. The point is that the external gain is certain and the internal one is not.

By the time a junior resigns, the company has lost the cheapest moment to act and is bidding against a concrete number. The counteroffer also answers the wrong question. It treats the problem as a pay gap, when the junior has usually reframed it as a recognition and trust gap that money no longer fixes.

The cost of getting to this point is measurable. Gallup estimates the cost of replacing an employee at one-half to two times their annual salary and calls that conservative. The Center for American Progress, pooling thirty case studies, put the median replacement cost at about 21 percent of annual salary for non-executive roles, with specialised and longer-ramp roles sitting well above that. A developer is one of those roles. Recruiting alone runs to thousands of euros before the replacement writes a line of code.

A word on a popular statistic. The claim that most people who accept a counteroffer leave within six months is repeated everywhere and supported by no locatable primary study, so it does not appear here as a number. The structural argument stands without it. The resignation is the most expensive and least informative moment to intervene, because the price has been set by an outside bidder.

Mentoring is the obvious answer and the evidence both supports it and explains why the usual version fails. The decisive study is Ragins and Cotton, who examined 609 protégés and found that people with informal mentors rated them as more effective and earned more than people with formal mentors. Formally mentored people did not differ significantly from people with no mentor at all, while informally mentored people clearly did. A name assigned on a form and a quarterly check-in produce roughly the outcome of doing nothing.

The wider literature fits this. Allen and colleagues' meta-analysis confirms that mentoring is associated with better career outcomes, though the effect sizes for objective measures such as pay are small. The often-quoted Sun Microsystems analysis reported much higher retention for mentees and mentors than for non-participants, but it is a single-company correlational study. Herminia Ibarra argued that high-potential people are over-mentored and under-sponsored, and that advice without advocacy changes little. Sylvia Ann Hewlett drew the same line between a mentor who talks with you and a sponsor who spends capital on you.

Visible internal paths do the rest. LinkedIn's analysis of 32 million member profiles found that after three years 70 percent of people promoted internally and 62 percent who moved laterally were still with the employer, against 45 percent who stayed in the same role. Accenture's professional apprenticeship programme, launched in 2016, has put more than 2,500 people through an earn-and-learn route that now accounts for about 20 percent of the firm's entry-level hiring in the US and Canada. That is a ladder with a visible top — exactly the perspective the maths-driven junior is missing.

The economics are not close. Replacing a developer costs from most of a year's salary up toward and beyond a full year once recruiting, ramp time and lost knowledge are counted. A real conversation about the internal path costs a rounding error against that. The lever is cheap precisely because it works before the offer arrives, when perspective still changes the decision and money has not yet become the only language left.

If you want to keep your best juniors, develop them and let them see what loyalty actually buys. Show them the path, in plain terms, from someone who has walked it and who cares whether they stay. The companies that do this lose far fewer people than those that only react to a resignation letter — and they spend less doing it.

A. Where the case is firm and where it is inference

The strongest claims rest on government data, peer-reviewed work and primary survey reports. The tenure figures, the leading reasons people leave, the switcher pay gap, the replacement-cost ranges, the decision-theory results and the Ragins and Cotton finding are all on solid ground. The cross-region attrition figures use mixed definitions and scopes and are not strictly comparable, so the chart is captioned as indicative. The counteroffer attrition figure is folklore with no primary source and is left out. Vendor and self-reported figures (Mercer, Landing.jobs, ITViec, Talentnet-Mercer, Zinnov, LinkedIn, GitHub, Linear, ADP) are flagged where they appear. The author's career arc is a personal account, not evidence.

B. Method

This paper triangulates independent research passes covering the academic and survey literature, the economics of turnover and the decision the junior makes, the practitioner record and a regional pass across Europe and Asia. Every numerical claim is referenced. Government and peer-reviewed primary sources should be checked against their origin documents before any single figure is quoted in isolation.

C. References

1. US Bureau of Labor Statistics (2024). Employee Tenure in 2024. https://www.bls.gov/news.release/tenure.nr0.htm
2. Stack Overflow (2023). Hopping Instead of Hustling, Developer Survey Pulse. https://stackoverflow.blog/2023/10/26/…
3. McKinsey & Company (2021). Great Attrition or Great Attraction, the Choice Is Yours. https://www.mckinsey.com/…
4. Gallup (Adkins, 2018). Millennials Want Jobs to Be Development Opportunities. https://www.gallup.com/…
5. Work Institute (2020). 2020 Retention Report. https://info.workinstitute.com/…
6. Gagné, M., and Deci, E. L. (2005). Self-Determination Theory and Work Motivation. Journal of Organizational Behavior, 26(4), 331–362. https://doi.org/10.1002/job.322
7. Forsgren, N., Storey, M.-A., Maddila, C., Zimmermann, T., Houck, B., and Butler, J. (2021). The SPACE of Developer Productivity. ACM Queue, 19(1). https://queue.acm.org/…
8. GitHub (2021). The Good Day Project, Octoverse Spotlight. https://github.blog/…
9. Mark, G., Gonzalez, V. M., and Harris, J. (2005). No Task Left Behind. CHI 2005, 321–330. https://doi.org/10.1145/1054972.1055017
10. Kumar, S., et al. (2025). Time Warp, Developers' Ideal vs Actual Workweeks. Microsoft Research, arXiv:2502.15287. https://arxiv.org/abs/2502.15287
11. Pew Research Center (2022). Majority of U.S. Workers Changing Jobs Are Seeing Real Wage Gains. https://www.pewresearch.org/…
12. McFeely, S., and Wigert, B. (2019). This Fixable Problem Costs U.S. Businesses $1 Trillion. Gallup. https://www.gallup.com/…
13. Boushey, H., and Glynn, S. J. (2012). There Are Significant Business Costs to Replacing Employees. CAP. https://www.americanprogress.org/…
14. Kahneman, D., and Tversky, A. (1979). Prospect Theory. Econometrica, 47(2), 263–292. https://doi.org/10.2307/1914185
15. Ellsberg, D. (1961). Risk, Ambiguity, and the Savage Axioms. Quarterly Journal of Economics, 75(4), 643–669. https://doi.org/10.2307/1884324
16. Ragins, B. R., and Cotton, J. L. (1999). Mentor Functions and Outcomes. Journal of Applied Psychology, 84(4), 529–550. https://pubmed.ncbi.nlm.nih.gov/10504893/
17. LinkedIn Talent Solutions (2020). 3 Factors Strongly Linked to Better Employee Retention. Vendor data. https://www.linkedin.com/…
18. Accenture (2024). Apprenticeship Programs. https://www.accenture.com/…
19. Mercer (2025). The evolving workforce, European technology (Comptryx). https://www.mercer.com/…
20. Infosys Limited (2022). Form 6-K, FY2022 (voluntary attrition 27.7%). US SEC EDGAR. https://www.sec.gov/…