Built for the islands
Issue 26Briefing

The Monthly Intelligence Report

Built for the islands

Jamaica’s AI and jobs study, a test for public information services and the records a pilot should leave behind.

Caribbean AI Association·October 2026Published

An AI trial can improve the work inside a firm while demand for that firm’s services slips. October’s Jamaica study puts those two changes side by side. The issue also returns to the public-service test raised in CAIA’s October review edition. Before a team relies on an answer or recommendation, someone should be able to trace it to the source and decide when it needs to be held.

Jamaica’s study separates adoption scenarios from demand stresses

What the output figures compare

The Association’s study of jobs, AI and export demand used Jamaica’s July 2026 employment data as a starting point. It compared five AI adoption scenarios under matched non-AI conditions. The study separately tested four cases of weaker demand for services from overseas. It covers 32 occupations, from accountants and bank tellers to hotel staff, tour guides and musicians. The scenarios illustrate what could happen under stated assumptions. The study has not been peer reviewed and assigns no probability to any path. CAIA’s 20 October Jamaica report.

In January 2030, the broad-adoption case puts Jamaica’s output 3.61% above the matched case without AI adoption. Average wages in the group the model classifies as cognitive occupations are 1.97% lower. Those are jobs the model groups around tasks such as preparing accounts or analysing documents. The unemployment rate is 3.76% under broad adoption and 3.40% in the matched case. The output figure compares the value of what Jamaica produces at the same point in time. See the core scenarios and their comparisons.

How a firm can use the demand cases

The combined export-demand case assumes broad AI adoption and weaker overseas demand. The comparison path assumes no AI adoption and keeps the original export-demand assumptions. The model puts output 0.79% below that comparison path in January 2030. The authors report output for this case and leave the unemployment rate unestimated. Take the output difference as a prompt to test your demand assumptions before you estimate how many jobs a firm might need. See the demand-stress scenarios.

Put that distinction to work in a service firm. Estimate what it costs to handle one contract using the current process. Test where an assistant could cut checking or preparation time. Then change the demand assumption on its own. Use one case in which the client renews the contract and another in which it buys less work. If the firm plans to move people into different roles, name those roles and the time staff would need to learn them. That work connects the scenarios to the choices a manager faces before renewal.

Give public employees a source they can open

What Jamaica’s environmental agency reported

In June 2025, Jamaica’s National Environment and Planning Agency described its use of drones, geographic information systems and mobile tools in monitoring and enforcement. It reported testing in Portland, St Catherine and St Mary and a pilot in St Elizabeth. It also described adding AI to help detect problems. For any proposed model, the agency still needs a way to check alerts against what a staff member can establish on site. Jamaica Observer coverage of the agency announcement.

If a monitoring tool flags a change on a map, keep the date and location of the picture beside the reference image so the officer can check it. Ask the officer to name the detail they would need to confirm before arranging a visit. Keep a record of cases where the source was too old, a boundary was unclear or the alert covered the wrong area. The officer needs that evidence to decide what the agency can do next.

Build a public answer around the official record

A test for Project Daylight

CAIA’s Project Daylight proposes making public information easier to find and understand. The October review edition described a proposed assistant that would connect explanations to official material and refer cases for human review. A public test has to show how it handles versions of a law or notice, especially when a rule changes by date or territory. Project Daylight.

In the October review edition’s editor’s note, Adrian Dunkley wrote, “A project announcement tells us what its organisers intend. A prototype shows what has been built.” A test of Daylight can make that distinction concrete. Give it a question about an application deadline, with two official notices that show different dates. The reviewer should be able to find the current notice and explain how they determined it was current. Then remove the notice the answer needs. See whether the assistant tells the resident what is missing and identifies the responsible office.

The same test helps the team design for people who take a screenshot of an answer and share it outside the service. Include the issuing authority and the date close to the answer. If a later notice changes the deadline, keep a dated correction with the original answer inside the service.

Make research questions local enough to check

Start with the decision a forecast would inform

The October review edition examined Caribbean climate modelling and the Association’s Think Tanks. Reef looks at where AI systems can fail and what institutions can check before people rely on them. Mangrove studies how forecasts and environmental monitoring could inform decisions about farms, coastlines and disaster response. A local forecast test brings those questions together. CAIA Think Tank directory.

For an island rainfall forecast, name the catchment the user needs to assess, the land that drains to the same river or stream. Write down when they need the answer. A grid cell in the model may cover slopes, valleys and coast at once, even though people in those places could experience different rain. Before a forecast guides a decision at the scale of a community, compare it with local observations and record where the test had no measurements. The absence of a station in one valley should be visible to the person using the result.

Put a record of the trial in front of the next decision maker

A small handover with a clear owner

Before extending a trial in November, ask the team to bring one case the tool handled well, one that staff corrected and one that staff held. Include the original record, the answer, the person who approved or stopped it and the time staff spent checking it. A colleague should be able to follow what happened without hearing the whole project history.

If the team can recover those details, use them to decide where the tool should be tested again. If a case fell apart because the source changed or the reviewer was unavailable, put that case back into the next trial. Give the person who approves changes the authority to stop it when the same problem recurs.

Read every issue of The Monthly Intelligence Report