Barbados Is Assessing AI's Effect on Jobs. Start With Clerks

On 4 October 2026 the Barbados government said it is moving to assess how artificial intelligence affects Barbadian workers. The Barbados Statistical Service already counts one group that deserves the first look: 14,000 clerks, about 84 per cent of them women. The official occupation figures offer a starting point for employers who want to examine their own clerical work.

What happened on 4 October?

The announcement was short. CBC Barbados reported that Mia Mottley spoke at the Barbados Workers' Union's 85th anniversary gala at the Hilton Barbados Resort. The report says: "Government is moving to assess the impact of artificial intelligence on the Barbadian workforce." It adds that work on the Competitiveness and Productivity Commission is expected to start soon.

The report names no agency to run the assessment, no deadline and no budget, and it carries no figures. An assessment announced at a union gala will be judged by whether it measures the jobs where workers are concentrated, so it helps to know what the existing numbers already say.

How many clerks work in Barbados?

The Barbados Statistical Service's Annual 2025 labour force bulletin estimates 126.3 thousand employed people and 8.8 thousand unemployed, a labour force of 135.1 thousand and an unemployment rate of 6.5 per cent. Its table of employment by occupation and sex (table 8.4) is the most useful single page for this question.

Occupation group, 2025	Men (thousand)	Women (thousand)	Both (thousand)
Service workers and shop workers	9.1	16.6	25.7
Elementary occupations	10.1	7.0	17.1
Professionals	5.0	9.9	14.9
Clerks	2.2	11.8	14.0
Technicians and associate professionals	6.8	6.2	13.0
Craft and related workers	11.3	1.3	12.6
All employed (including groups not shown)	64.1	62.2	126.3

From those figures we calculate the following, and these percentages are our arithmetic, not the Statistical Service's. Clerks are about 11 per cent of everyone employed. Women hold 11.8 of the 14.0 thousand clerical jobs, about 84 per cent. Treat these as national-level survey estimates only: the survey samples 2,160 households and the published cells are rounded to 0.1 thousand, so the 84 per cent has a margin we cannot calculate from the bulletin. Clerical work employs roughly one in five employed women (11.8 of 62.2 thousand, about 19 per cent) and about one in thirty employed men (2.2 of 64.1 thousand, about 3 per cent).

A labour-market study needs to distinguish between occupations. The same technology would land differently on a payroll office than on, say, a fish vendor at Oistins, to take an illustrative example of an occupation with little desk work.

What does the ILO say about clerical work and AI?

The ILO's Working Paper 140, published in May 2025, updates its global index of exposure to generative AI. It reports that "clerical occupations continue to have the highest exposure levels". Globally, one worker in four is in an occupation with some exposure. The share is 34 per cent of employment in high-income countries and 11 per cent in low-income countries. Of all employment worldwide, 3.3 per cent falls in the highest exposure category, and that includes 4.7 per cent of women's jobs against 2.4 per cent of men's.

A second ILO publication brings the gender point closer to home. In a research brief dated 5 March 2026, the ILO found that women are more exposed to generative AI than men in 88 per cent of the countries analysed. It says more than 40 per cent of women's employment is exposed in several economies, among them small island developing states in the Caribbean and the Pacific. The brief ties this to women's concentration in clerical, administrative and business support roles, such as secretaries, receptionists, payroll clerks and accounting assistants. It does not name Barbados.

For the wider region, a joint ILO and World Bank study published on 31 July 2024 estimated that 26 to 38 per cent of jobs in the region could be influenced by generative AI. It put full automation at 2 to 5 per cent of jobs and found that 8 to 14 per cent of jobs could see productivity gains. It also found that gaps in digital access and infrastructure could hold back about half the jobs that stand to gain, some 17 million, of which 7 million are held by women.

Read those figures with the ILO's own caution. The index measures possible exposure, not realised job losses, and the ILO expects transformation rather than replacement to be the most likely outcome. Actual effects depend on the country and the sector.

Which Adrian Dunkley method fits this story?

The method is People First, Problem Second, AI Last, which StarApple AI publishes on Adrian Dunkley's FAQ page. It starts with the human need, defines the problem precisely enough to know what a solution would look like, and chooses the technology last. The method was written for adoption decisions. Applying it to a national workforce assessment is our extension, and we did not seek comment from Adrian Dunkley for this article.

For this assessment, workers need to keep earning and employers need accurate office work. The problem, stated without the letters A and I: which tasks inside clerical jobs take the most hours, and which of them depend on judgement or on dealing with a customer? Only after that does the question of a tool arise. Exposure estimates help identify where to look; workers and employers then need to describe the tasks themselves.

An occupation is a bundle of tasks, and the ILO's index scores tasks and then rolls them up. A "clerk" in the Statistical Service table could be a bank teller, a records officer in a ministry or a dispatcher at a distribution company. The table cannot tell them apart, and neither can a headline percentage.

What does this mean for Barbados specifically?

Three local features sharpen the picture. The first is the size of the services economy. The World Bank's data site, using a modelled ILO estimate, puts employment in services at 82.6 per cent of total employment in Barbados for 2025. Services include many kinds of work. That sector share cannot tell us how much of it is desk-based or exposed to AI.

The second is the concentration of women in the clerical group. Our inference, not an official finding: if exposure is highest in clerical work and 84 per cent of clerks are women, then any Barbadian effect of AI on clerical hiring would be felt first by women. No Barbados study that we could verify has tested this.

The third is the state of the evidence. The Statistical Service's occupation table for unemployed people groups clerks with service workers, shop workers and skilled agricultural workers. That combined group accounts for 3.0 thousand of the 8.8 thousand unemployed, 2.1 thousand of them women, according to table 8.8 of the same bulletin. Because the table lumps three occupation groups together (clerks, service and shop workers, and skilled agricultural workers), it cannot show whether clerks are over-represented among people out of work, which is what an assessment would want to know.

For wider context on the skills side, our earlier reports cover the national digital skills campaign and which Bajan jobs AI may change. Our sister site AI Guyana and AI Jamaica track the same questions in their own labour markets.

We tried to check whether the IMF's 2024 working paper on AI and productivity in Latin America and the Caribbean carries a Barbados breakdown of exposed and complementary jobs, but the IMF site refused our requests, so we cite no such figure. If you hold the paper, check its country tables before quoting anyone's Barbados percentage.

What should the assessment measure?

Our suggestions, offered as an outsider's checklist and not as anything the government has said it will do. First, count tasks, not job titles. The ILO scores tasks, so a Barbadian assessment that stays at the level of "clerks" will repeat the limit of the Statistical Service table. Second, split every finding by sex, because the clerical group is where the numbers diverge most. Third, ask employers what they have already tried. An employer that has trialled a drafting tool for six months knows things no index records, and those results, failures included, belong in the evidence.

Fourth, publish the method alongside the result. If the assessment reuses the ILO's global index, it should say so and flag that the index was not built from Barbadian job descriptions. Fifth, repeat the exercise. The Statistical Service publishes its labour force survey every quarter, which means a baseline taken now can be compared with one taken in a year. A single snapshot of exposure would age quickly.

A last point for employers who are waiting for the national result before they act. Nothing stops a firm from measuring its own clerical work this month, and the three steps below take about three weeks.

What are the risks?

The first risk is a wrong reading of the exposure numbers. A high exposure score invites the headline that clerks will lose their jobs, and the ILO says that outcome is not what its figures show. Acting on that headline could mean cutting staff before anyone has measured whether a tool does the work accurately. The second risk falls on women in clerical roles, who are the most exposed to any change in clerical hiring, since 84 per cent of Barbadian clerks are women (our inference, not an official finding). The third is error: a tool that drafts letters or reconciles figures can be confidently wrong, and in our judgement a customer or a taxpayer would bear the cost. Against that, the ILO and World Bank study above found that 8 to 14 per cent of jobs in the region could see productivity gains. By our inference, a clerical team that checks outputs carefully might save time on routine drafting and reconciliation, and no Barbados study has measured it, which is what the test below is designed to measure. The guardrail for all three is the same. Measure before you decide, keep a named person responsible for checking outputs, and involve the workers who do the job, which is a practical reason to bring the Barbados Workers' Union to the table early.

What can you do this week?

These steps are a test we suggest, built from the People First method. They are not an official procedure, and the numbers are thresholds we picked for a first look.

1. Write, by Friday 16 October 2026, the human need and the problem for one clerical role in two sentences each, without using the word AI. If either sentence needs the word to make sense, the problem is not defined yet.
2. Log, by Friday 23 October 2026, every task the person in that role does over five working days, with minutes spent. Mark each task as rule-based, judgement or customer contact. Calculate the share of logged minutes that are rule-based.
3. Test, by Friday 6 November 2026, a tool on one rule-based task using at least 20 real items. Record minutes per item before and after, and have a person check every output. Count the outputs with an error. Keep a person checking every item throughout the trial. If more than 2 of the 20 contain an error that would have reached a customer, pause the trial and investigate. Fewer errors in this small sample would not establish that unsupervised use is safe.

What the test cannot show: it covers one role in one workplace, 20 items give a wide margin, and times logged by the worker themselves are rough. It says nothing about national exposure. Its value is a record of your own work that a union representative, a manager or a ministry can read.

Frequently asked questions