Caribbean CEOs are getting AI advice written for American multinationals with large budgets and full engineering teams. I built StarApple AI from Kingston with Caribbean-scale resources, and my team has deployed AI across 19 platforms in this region. Much of the advice I hear in boardrooms in Port of Spain, Bridgetown, Georgetown and Kingston does not fit the businesses it is being applied to.
Jean Auguste Dominique Ingres, Joseph-Antoine Moltedo (born 1775), ca. 1810. The Metropolitan Museum of Art, public domain.
I founded StarApple AI, the Caribbean's first AI company, in 2016. I sit on Jamaica's National AI Task Force and chair the Caribbean AI Association. Most weeks I talk with founders, executives and ministers trying to decide where AI fits in their organisations.
Caribbean CEOs tend to fall into two groups. Some are stuck: AI feels too big, too fast and too expensive to start. Others think they are done because they bought a chatbot or an AI writing tool. The first group is losing ground to competitors. The second has bought a tool without deciding what it is for.
The exposure is real. The World Bank has estimated that 40 to 57 percent of Caribbean jobs face medium-to-high automation risk over the next decade. Few Caribbean firms have a plan for that.
Why US enterprise AI advice fails Caribbean firms
The case studies in business magazines come from companies with data science teams, decades of clean digital records, deep engineering benches and markets big enough to absorb expensive experiments. Few Caribbean businesses have any of those.
Small and medium enterprises provide most private-sector jobs across CARICOM. They run lean management teams, sell into markets of tens of thousands of customers, and keep their data in spreadsheets, WhatsApp chats, accounting software and paper files. A US$200,000-a-year enterprise contract designed for a North American bank is no use to a Barbadian accounting firm with 40 staff.
The approach that works here is narrower. Find four or five workflows where AI cuts time, cuts errors or improves service without heavy integration work. Get good at running those. Expand from there. A CEO who copies the US playbook will overspend, see weak results, and wrongly decide that AI does not work in the Caribbean.
Four decisions to make this year
A CEO who has made these four decisions, even if one of them is a deliberate "not yet", is ahead of one who keeps putting AI on next quarter's agenda.
1. Who owns AI. AI projects handed entirely to IT, or to a junior "AI champion" without budget authority, stall at proof of concept. The champion cannot approve money, headcount or process change. The CEO does not need to understand machine learning but does need to own the priorities and review progress monthly.
2. Buy or build. Most Caribbean businesses should buy. Building bespoke AI needs data scientists the region is short of. Buy tools built on proven platforms, configure them for local conditions, and use local expertise where it exists. Build only where you hold data no global model has: decades of regional credit history at a financial institution, yield and soil records at an agricultural firm, or last-mile delivery data at a logistics operator.
3. Data first. Before any large AI deployment, find out what data you hold, how clean it is and what you should start collecting now for use in two or three years. Most firms that do this find their data is worse than they assumed: duplicate customer records, inconsistent transaction history and manual steps that leave no digital record.
4. Governance before scale. Write down who may use AI, on what data, how outputs are checked before they reach a customer, and who answers for an error. The Caribbean AI Risk Management Council publishes governance frameworks for Caribbean enterprises that are a practical starting template.
Mistakes that stall Caribbean AI projects
The failures I see are mostly organisational, not technical. That matches RAND Corporation's 2024 study of failed AI projects, which put the failure rate above 80 percent and found that the most common root cause was leaders misunderstanding the problem the AI was meant to solve.
The pilot that never scales. Caribbean organisations run pilots well and scale them badly. A customer service bot that handles 200 test conversations often stalls before 2,000 live ones, because no one built the review process, escalation path or staff training. Write down the success criteria and the next steps before the pilot starts.
Trusting a vendor demo built for another market. AI built for global markets fails here in ways that are hard to spot:
- A credit model trained on US data misreads income from the informal economy.
- A voice or chat assistant trained on North American English struggles with Jamaican Patois, Trinidadian Creole or Guyanese speech.
- A routing tool built for US roads ignores road quality, flooding and seasonal access.
- A document tool trained on US forms misreads TRNs, local addresses and Caribbean date and currency formats.
Test every tool on your own data and customers before signing.
What AI-ready Caribbean organisations have in common
Leaders who use the tools themselves. The CEO, CFO and COO have each spent eight to ten hours using the leading AI assistants on their own work. Executives who have never used one are relying on second-hand opinions of what it can do.
Staff who have been told the truth. When AI is introduced as an efficiency tool without an honest conversation about jobs, staff resist without saying so and every rollout slows. Staff who know how their own role will change and have been given time to learn the tools adopt them. The AI Jamaica network and the Caribbean AI Association both run staff AI literacy programmes for Caribbean organisations.
Partners who know the region. Caribbean AI capacity has grown since 2023. Maestro AI Lab is building AI research capacity in the region, and the 14West AI Fund invests in Caribbean AI ventures. Use global tools where they are clearly best, and regional partners for configuration, testing and support.
Frans Hals, Portrait of a Man, early 1650s. The Metropolitan Museum of Art, public domain.
What to expect by mid-2028
- AI agents in routine operations. Tools that carry out multi-step tasks on a user's behalf are likely to be standard business software by late 2027. Firms with clean data and trained staff will adopt them quickly; others will be starting from scratch.
- More AI talent close to home. StartupBlink's 2026 ranking placed Kingston 87th globally and first in the Caribbean as a startup hub. More local talent means less dependence on international consultants who do not know the region.
- Visible job changes. Business process outsourcing, financial back offices and routine administration will need fewer people for the same volume. Firms that plan reskilling and role redesign now will handle it with less disruption.
- Clearer rules. Jamaica's National AI Task Force recommendations and CARICOM-level discussions are likely to produce the first regional AI governance frameworks. Firms with internal governance already in place will adapt with little effort.
Where my advice could be wrong
Telling most firms to buy rather than build has a cost I have not solved. Every firm that buys global tools sends its data and its spending abroad and builds little local capability. If the whole region follows that advice, the Caribbean ends up as a customer of AI and not a producer of it. I still think buying is right for a 40-person firm this year, but it is a weaker answer for banks, telecoms and governments with the scale to build.
What a Caribbean CEO should do this quarter
- You, the CEO: block eight to ten hours over the next month to use Claude, ChatGPT or Gemini on your own real work: a board paper, a contract review, a sales proposal. Note where it helped and where it was wrong.
- You and your COO: list your ten most time-consuming workflows and pick one or two for a six-week test. Record baseline hours and error rates before starting and agree what result would justify scaling.
- Your finance or operations lead: run a two-week data audit covering customer records, transactions and supplier data. Score each source for completeness and duplication.
- Your legal or compliance lead: write a one-page AI use policy covering approved tools, data that must never be entered, human review of customer-facing output, and who is accountable. Check it against your country's data protection law.
- Your HR lead: brief staff on which roles the pilot affects before it starts, and book AI literacy training for those teams within the quarter.
Frequently Asked Questions
What does it cost a Caribbean firm to start using AI assistants?
Business plans for the main assistants, such as ChatGPT Team and Claude Team, cost roughly US$25 to US$30 per user per month; check the vendors' current pricing. A pilot for ten staff therefore costs a few thousand US dollars a year in licences. The larger cost is staff time to learn the tools and to redesign the workflow, so budget hours as well as money.
Is company data safe to put into ChatGPT or Claude?
On business and enterprise plans, the main vendors state that they do not train their models on your data by default, which is not always true of free personal accounts. The data is still processed on the vendor's servers, usually outside the Caribbean. Ban staff from putting customer personal data into personal accounts, and read the vendor's data processing terms before approving a tool.
Which data protection laws apply to AI use in the Caribbean?
It depends on the country. Jamaica's Data Protection Act 2020 came into force in December 2023 and is regulated by the Office of the Information Commissioner. Barbados has its Data Protection Act 2019, and several other CARICOM states have laws at different stages of enactment. If you serve EU customers, GDPR also applies, and the EU AI Act, in force since August 2024, affects AI systems used on people in the EU.
Is there a recognised standard for AI governance a Caribbean firm can follow?
Yes. ISO/IEC 42001, published in December 2023, is the international standard for an AI management system and can be certified by an accredited auditor. The US NIST AI Risk Management Framework, released in January 2023, is free and widely used as a checklist. A mid-sized Caribbean firm does not need certification to start, but using one of these as the template makes later audits and regulator questions easier.
How do I tell whether an AI vendor's claims will hold in my business?
Ask for a paid proof of concept on a sample of your own data, not a demo on theirs. Agree the measure of success in writing before it starts, such as accuracy on 200 real customer queries or hours saved per week. Ask for a reference customer in the Caribbean or a similar small market, and call them.