You Don't Have a Labour Shortage. You Have a Retention Problem.

Every conversation I have with another factory owner about the garment labour shortage runs the same way. Nobody can fill the line. Everyone blames the gig economy, the young, the government, the neighbouring factory that pays five rupees more. It is a conversation about a shortage of people, and it ends where it starts, because a shortage of people is not something a factory owner can fix on a Monday.

Then I read the ILO's 2015 survey of garment workers in Bangalore and the National Capital Region, and it turned the problem the other way round. I run a CMT sewing factory in Nepal and I have lost operators the way everyone else does, so this is not an academic interest. This article is what the evidence says about keeping sewing operators, including the parts that contradict what most of us do.

The Finding That Reframes the Whole Problem

The ILO's FUNDAMENTALS branch published Insights into working conditions in India's garment industry (Geneva, 2015), based on interviews with current and former garment workers in Bangalore and the NCR. It was commissioned in part to investigate labour shortage and staff turnover, so this is the question it was built to answer.

The relevant sentence reads: "despite concerns about labour turnover in the industry, a clear majority of workers (70%) had been employed for 3 years or more in the sector at the time of the interview, including 20% who had worked for more than 10 years. So, the main issue seems to be limited retention of workers at the level of the factory, rather than within the industry as a whole."

Among former garment workers in the same survey, 60% had worked in the sector four years or more, and three-quarters had worked in more than one factory. That is the shape of the thing. People stay in garments. They move between factories. When an operator walks out of my gate, the overwhelming likelihood is that she is walking into someone else's.

I want to be careful about how far that generalises. This is one survey, in two Indian clusters, eleven years old. Regional headcount shortages are separately documented and real, and they coexist with tariff-driven layoffs in the same markets, which is volatility rather than a one-way shortage. But the tenure distribution is the only evidence I have seen that speaks directly to whether the people are missing or merely somewhere else, and it says somewhere else.

You Lose Them in the First Twelve Weeks

If workers churn between factories, the question becomes when they leave yours. The answer is: almost immediately.

Abebe, Caria, Dercon and Hensel studied a large garment firm in Hawassa Industrial Park, Ethiopia, in an IGC working paper published in August 2019. Their figure 1 shows "more than 40% of workers having left after just 12 weeks." In the same paragraph they note that "workers are hardly productive in the first 8 weeks of their employment spells as they have to be trained."

Put the two sentences together and the arithmetic is brutal. The training investment is spent over eight weeks. Over 40% of the intake is gone by week twelve. Most of the people you train never reach the point of repaying what training them cost, which is why the authors treat this turnover as inefficient rather than as a neutral churn.

The pattern shows up in India too, and specifically among women taking their first factory job. Aruna Ranganathan of Stanford GSB studied a menswear factory of about 2,000 workers near Bangalore, roughly 90% women from nearby villages, over two years of data and interviews. As the Stanford King Center reported in May 2019: "36 percent of first-time workers at this factory leave within three months and 11 percent leave after just one month."

Different country, different sample, same window. Your retention problem is concentrated in a period most factories manage least — the weeks when the new operator is slow, unprofitable, and nobody's priority.

Two Bonus Experiments, Two Failures

The standard factory answer to early attrition is a retention bonus. Stay three months and get a payment. It is the first thing everyone reaches for, and the only experiment I can find that tested it inside a garment factory found nothing.

The same Ethiopian team ran a pilot field experiment at the Hawassa firm comparing a retention bonus of 1,250 Birr — about US$45, roughly a month's entry wage, paid only if the worker was still there after three months — against an unconditional bonus of the same value. Their abstract states: "Surprisingly, we find no statistically significant difference in three months retention rates between the two bonus schemes." They add that comparing the experimental cohorts to previous cohorts showed strong differences in retention, which "suggests that low wage levels might be the main reason for high turnover."

Their own reading of their own data, in short, is that the lever was the base wage rather than the design of the bonus on top of it. This was a pilot, and the authors present it as suggestive rather than definitive. It is still the only causal test of a retention bonus in a garment plant that I have been able to find, and it came back null.

The attendance-bonus evidence is worse than null. Alfitian, Sliwka and Vogelsang ran a pre-registered randomised trial (AEARCTR-0002863) assigning 346 apprentices in a German retail chain for one year to a monetary attendance bonus, a time-off bonus, or a control group. Their abstract: "We find that neither form of the bonus reduced absenteeism, but the monetary bonus increased absence by around 45%. This backfiring effect is persistent and driven by the most recently hired apprentices. Survey results reveal that the bonus shifted the perception of absenteeism as acceptable behavior." The paper is published in Management Science.

I have to label that one honestly: German retail apprentices are not South Asian sewing operators, and I would not transplant the 45% figure onto a sewing floor. What travels is the mechanism. Attaching money to attendance told people that attendance was a negotiable, priced thing rather than a norm, and the damage outlasted the bonus. Two experiments on two continents, one in our industry and one not, both found that bonuses did not do what factories assume they do.

The Cheapest Thing That Did Work: Asking

Adhvaryu, Molina and Nyshadham ran a randomised controlled trial across roughly 2,000 workers in Indian garment factories, reported in NBER Working Paper 25866 and published in the Economic Journal. Just after a statutory wage hike that proved disappointing, half the surveyed workers were chosen at random to take an anonymous survey asking for feedback on job conditions, supervisor performance and overall job satisfaction. In the authors' words, "Enabling voice in this manner reduced turnover and absenteeism after the hike, particularly for the most disappointed workers." Quits fell by roughly two percentage points.

The detail that makes this worth your attention: the findings were not shared with management and no action was taken on them. The retention effect came from being asked, not from anything being fixed. I do not entirely like what that implies about how rarely anyone asks. But it is the cheapest verified retention lever in the whole literature, and the barrier to running one is a printed form and a sealed box.

Train Your Weakest Supervisors, Not Your Best Ones

Where retention effects do turn up reliably, they turn up for supervisors. Adhvaryu, Murathanoglu and Nyshadham ran a randomised controlled trial among production line supervisors at a large ready-made garment firm — 1,849 supervisors, 928 treated and 921 control, also circulated as NBER Working Paper 31335.

The headline is a 6–7% productivity gain on lines managed by treated supervisors. The finding worth acting on is underneath it. From the abstract: "Highly recommended supervisors experienced no productivity gains; the average treatment effect of training is driven entirely by low-recommendation supervisors." The weakest-rated supervisors gained more than 11%. The best-rated gained nothing measurable.

There is a second half. "Treated supervisors were 15% less likely to quit than controls over the study period, and this gain was most pronounced for highly recommended supervisors." So training your stars buys retention of those stars, and training your weakest buys output. Both are legitimate purchases. The paper's point is that middle managers had been choosing the first while believing they were choosing the second — they prioritised keeping supervisors they feared losing, and the productivity money went to people who had nothing left to gain.

If you take one operational decision from this article, take that one. Look at who you last sent on a supervisor course and ask which of the two things you were buying.

Who Trains the Newcomer Decides Whether She Stays

Back to Ranganathan's Bangalore study. Having established the 36% three-month cliff, her team found that the single most significant factor in whether a first-time worker stayed was the trainer who prepared her. In the King Center's summary: "Newcomer women whose trainers focused on social skills such as time management and communication, in addition to the necessary technical skills, had a 20 percent higher probability of being retained after three months."

Not a better curriculum. Not more training days. The same training, delivered by someone who also taught the newcomer how to exist in a factory — how to manage her time, carry herself, deal with people she does not know. A 20% improvement in three-month retention for the cost of choosing a different person to stand next to the trainee.

From my own floor rather than from any study: this matches what I see, and it is not a decision most factories make deliberately. The trainer is usually whoever has a free machine and a reputation for patience. It is worth treating as a real appointment. I have written more about the mechanics of that window in what the first 90 days of training a new sewing operator actually look like.

Training Raises Productivity. It Does Not Fix Retention.

This is the part that most industry writing gets backwards, so I want to state it plainly.

Adhvaryu, Kala and Nyshadham evaluated the P.A.C.E. soft-skills programme at Shahi Exports in Bengaluru, a randomised controlled trial across five garment factories with 2,703 workers, later published in the Journal of Political Economy (131(8), 2023). In the December 2019 working-paper version the abstract reads: "Treated workers were 20 percent more productive than controls after the program… Wages rose only modestly with treatment (by 0.5 percent), with no differential turnover… The net return to the firm was 258 percent eight months after program completion." The published journal version reports a somewhat lower productivity effect and states the same absence of a retention effect; I am quoting the working paper because it is the version whose full text is publicly readable, and the figures differ slightly between the two.

Read the middle clause again. No differential turnover. A programme that made workers a fifth more productive and returned 258% to the firm did not keep them any longer than the control group. J-PAL's evaluation summary describes the same study.

Operator training is a productivity investment and a good one. The ROI stands on productivity alone and needs no help. But if a vendor or a CSR deck sells you operator training as a retention fix, they are selling you something this study looked for and did not find.

Two Numbers You Have Probably Been Quoted That Are Wrong

Both of these circulate widely enough that you have likely seen them in a presentation. Correcting them is part of what this article is for.

"Bloom's study found an 11% productivity gain from better management." Bloom, Eifert, Mahajan, McKenzie and Roberts, "Does Management Matter? Evidence from India", was published in the Quarterly Journal of Economics in February 2013, and the published figure is a 17% productivity increase in the first year. The 11% comes from an earlier working-paper draft. The second correction matters more for us: the firms studied were large Indian textile firms, not garment sewing units. It is a genuinely important result and it is routinely cited at the wrong number and about the wrong industry.

"Ethiopian garment factories run 10% monthly absenteeism." Go back to the primary source. NYU Stern's Center for Business and Human Rights, in Made in Ethiopia (Barrett and Baumann-Pauly, May 2019), writes at page 17 that "Attrition at Hawassa factories runs at 5% to 10% a month, according to employers." That is attrition, not absenteeism — people leaving permanently, not people missing a shift. Somewhere in the citation chain the word changed, and a monthly quit rate became a monthly absence rate, which is a completely different operational problem with completely different fixes. Note also that the NYU figure is employer-reported rather than independently audited, and describes 2017–18.

For a real absenteeism number, the best available comes from six consecutive months of worker-level productivity data across four garment factories in Karnataka, summarised on Ideas for India from work by Adhvaryu, Gauthier, Nyshadham and Tamayo: 10–11% of workers absent on a typical day, nearly all of it unauthorised, and beyond a 10% threshold each additional percentage point of absenteeism cost about 0.25 percentage points of productivity.

What I Would Actually Do, In Order

Ordered by how strong the evidence behind each one is, not by how easy they are.

1. Measure first-90-day attrition as a named number. If the cliff is where the evidence says it is, you cannot manage it while it is buried inside an annual turnover average. Count the intake, count who is still there at day 90, and put that ratio in front of whoever runs the floor every month.
2. Choose your trainers deliberately, and tell them the job is bigger than stitching. Ranganathan's 20% retention difference came from trainers who taught social skills alongside technical ones. This costs a staffing decision.
3. Run one anonymous survey. Two percentage points off quits, in a randomised trial, with no action taken on the results. There is no cheaper intervention in the literature.
4. Send your weakest supervisors for training, not your best. Zero measured productivity gain for top-rated supervisors, over 11% for the weakest, in an RCT of 1,849 supervisors.
5. Look hard at your base wage before you design any bonus. The Ethiopian authors concluded base wage levels were the likely driver where their bonus was not. A bonus is the cheaper-looking option, which is exactly why it is the one everyone tries first.
6. Treat operator training as a productivity purchase. It pays for itself on output. Do not put retention in the business case.

What the Evidence Does Not Support

I would rather tell you where the ground is empty than let you walk onto it.

• Canteen, transport and creche as retention fixes. Every list of retention measures includes them. I could not find a single controlled study in garment manufacturing measuring their effect on attrition. The most-quoted figure — a Vietnamese plant halving turnover after opening a kindergarten, from an IFC business-case report — is a footwear factory, self-reported, spans seven years with the kindergarten opening mid-window, has no control group, and the company itself credits the creche only "in part." That does not mean these things do not work. It means nobody has shown that they do, so measure it in your own factory before you spend.
• A replacement cost per operator. The dollar figures in circulation are US and European general-manufacturing benchmarks and are meaningless at South Asian garment wages.
• An industry attrition rate for sewing machine operators. The commonly quoted ranges trace back to aggregator and low-quality journal pages with no primary study behind them. I have no credible number to give you, and neither does anyone quoting one.
• A published operator learning curve. There is no peer-reviewed week-by-week efficiency curve for an individual sewing operator's first months. What exists is practitioner planning figures and one line-level style curve, which are different things.
• That software fixes any of this. I sell production software and it does not. It can tell you your first-90-day attrition number and which lines lose people, which is knowing rather than fixing.

Frequently Asked Questions

Is the garment labour shortage real, or is it a retention problem?

Both are real, but they are different problems and only one of them is yours to solve. The ILO's 2015 survey of current and former garment workers in Bangalore and the National Capital Region found that 70 per cent had been employed in the sector for three years or more, including 20 per cent who had worked more than ten years. Its conclusion was that the main issue seems to be limited retention of workers at the level of the factory, rather than within the industry as a whole. Regional headcount shortages are documented and so are simultaneous tariff-driven layoffs. What the tenure data shows is that the people you lose usually go to another factory, not out of the trade.

Do retention bonuses reduce garment worker turnover?

The one experiment run inside a garment factory found no effect. Abebe, Caria, Dercon and Hensel ran a pilot field experiment at a large garment firm in Hawassa Industrial Park, Ethiopia, comparing a retention bonus of 1,250 Birr, roughly 45 US dollars or about one month's entry wage paid only if the worker stayed three months, against an unconditional bonus of the same value. They report no statistically significant difference in three months retention rates between the two bonus schemes, and conclude that low wage levels might be the main reason for high turnover. A separate randomised trial of attendance bonuses among 346 retail apprentices in Germany found the monetary bonus increased absence by around 45 per cent, and that the effect persisted after the bonus was removed. That study is not a garment factory and not a developing country, but it is the only clean randomised test of an attendance bonus we could find.

Does training sewing operators reduce attrition?

On the available evidence, no. The soft-skills training trial run with 2,703 workers at Shahi Exports in Bengaluru, reported by Adhvaryu, Kala and Nyshadham and later published in the Journal of Political Economy, found treated workers were about 20 per cent more productive than controls with a net return to the firm of 258 per cent eight months after completion, and states in the same abstract that wages rose only modestly with no differential turnover. Training is worth buying for the productivity alone. Buying it as a retention fix is buying something the research did not find. Where retention effects do appear in this literature, they appear for supervisors rather than line operators.

What is the cheapest thing a factory can do to improve retention?

Ask people what they think, anonymously. In a randomised trial across roughly 2,000 workers in Indian garment factories, Adhvaryu, Molina and Nyshadham invited half the surveyed workers at random to an anonymous survey about job conditions, supervisor performance and job satisfaction, just after a disappointing statutory wage hike. Quits fell by roughly 2 percentage points and absenteeism also fell, concentrated among the most disappointed workers. The findings were not shared with management and no action was taken on them, so the effect came from being asked rather than from anything being fixed. The second cheapest lever is choosing who trains your newcomers, which costs a staffing decision and nothing else.

The uncomfortable conclusion of all of this is that the levers with the best evidence behind them are the ones that do not feel like management: ask people anonymously, pick the trainer carefully, train the supervisor nobody rates, and look at the base wage before designing a scheme on top of it. The levers that feel decisive — the bonus, the scheme, the announcement — are the ones that were tested and did not work. I find that annoying. I also think it is what the evidence says.

Santosh Rijal is the founder of Scan ERP, a garment manufacturing ERP system designed for factory floor operations. He works directly with sewing lines, cutting rooms, and production supervisors across Nepal's garment manufacturing sector.