Gold and teal ribbons cross and separate against a dark background, illustrating how a workflow can be taken apart and redesigned.

Work and skills Caribbean

Before you give a task to AI, ask why anyone still does it

10 November 2026 Article

Original illustration prepared by AI Boss Limited.

The Caribbean AI Association’s REWEAVE framework gives teams a way to decide which tasks to drop, where AI can help and what staff need to learn. Its test is what happens to service quality, workloads and the hours a team expects to save.

Picture a claims officer copying a customer’s name, policy number and accident date from one screen into another. AI could do the copying. The first question for the insurer is who needs the second record. If the original system already serves everyone who needs it, the team may be able to drop the step.

Exposure tells you where to look

The International Labour Organization estimated in 2025 that about one in four workers globally were in an occupation with some exposure to generative AI. That is an estimate of potential task exposure. It cannot tell you whether a particular employer should change a role or whether a worker will lose their job. ILO senior economist Janine Berg said the index helps countries “assess potential exposure and prepare their labour markets for a fairer digital future.” ILO study and statement

What happens at work depends on the task. Researchers followed 5,172 customer support agents as one company introduced an AI assistant in stages. They estimated that access increased issues resolved per hour by 15% on average, with larger gains for less experienced agents. A field experiment with consultants found gains on some assignments and more wrong answers on a task outside the tool’s tested capabilities. Customer support study · Consulting experiment

Give the work a new home

Start by mapping one workflow. Review the work as people do it. Eliminate steps that serve no remaining purpose. Weigh who should handle each task. Equip staff to check the tool and handle the cases it cannot. Allocate any usable time to work that needs it. Verify the result against how the work performs today. Evolve the role when the evidence supports the change.

In the claims example, the team could retire the second record, use an approved template to acknowledge receipt, and let AI suggest which documents appear to be missing against the approved checklist. An officer checks those suggestions before contacting the customer. Coverage decisions and disputed cases stay with the authorised officer. The same team needs a way to handle the queue when the system is down.

Give people time to learn the new job

If an experienced colleague has to correct every draft, the workload may have moved desks. Count that time. Ask the person who does the checking what slows them down and whether they can decline an unsafe draft without delaying the customer.

A new worker has a different problem. Routine files used to teach them what an ordinary claim looks like. If those files go through faster or skip their desk, protect time for a supervisor to walk them through a mix of straightforward and difficult cases. Ask them to show that they can spot a missing record, explain the next step to a customer and finish a case if the tool goes offline.

Try it with one workflow

Bring a colleague and two cases to a short meeting. Pick one routine case and one that went off script. Write down what the customer needed, who touched the file and how long the work took, including the checking. Choose one step to remove or change. Before trying it, agree who can stop the trial and which mistakes would trigger a stop. If you work alone, test your own process on dummy material first and check the rules before using customer records. If customers use voice notes or a local language variety, include those in the test alongside standard forms.

Track service alongside time. Did the customer get a complete answer? Did more files need to be reopened? If the team recovers enough time to schedule it, put it on the roster for something specific, such as overdue callbacks or supervised practice. The next review should show whether the hours reached that work.

What the framework is designed to test

The white paper proposes comparing REWEAVE with a rollout that starts with a tool. Both groups would use the same safeguards and measure staff time, service quality and workload. That comparison has yet to be carried out.

Read the white paper and use the worksheets

More research

Independent Caribbean analysis from the Caribbean AI Association. More analysis · Join the association