Reading Time: 6 minutes

Your taxes are paying for it

The region ranks near the bottom of global AI readiness. Spending is accelerating anyway. A free assessment programme wants to stop the waste before it starts.

The Oxford Insights 2024 Government AI Readiness Index ranked Latin America and the Caribbean as one of the least prepared regions in the world for AI adoption. 

Only Sub-Saharan Africa scored lower. None of the highest-ranked CARICOM nations, including The Bahamas, Barbados, and Trinidad and Tobago, placed in the regional top ten. That ranking did not slow down spending.

Trinidad and Tobago has stood up an entire Ministry of Artificial Intelligence. The Caribbean Telecommunications Union launched a Caribbean AI Task Force in July 2025. UNESCO published a Caribbean AI Policy Roadmap built from consultations with over 1,000 institutions across 20 countries. CARICOM and the UNDP have formally committed to coordinating AI adoption across the region. Budgets are moving. Procurement cycles are underway across Jamaica, Guyana, Barbados, The Bahamas, and beyond.

The ambition is rational. The sequencing, according to one of the region’s longest-serving AI practitioners, is where things go wrong.

That failed AI project? You paid for it.

Government AI projects do not fail the way private sector projects fail. When a startup wastes money on a tool it cannot use, the founders absorb the loss. When a government ministry procures an AI system that sits unused for six months, citizens absorb that cost through taxes. Your taxes.

Adrian Dunkley, founder of StarApple AI, the Caribbean’s first AI company, has spent over a decade implementing AI systems across insurance, finance, and public sector organisations in the region. He says the failure pattern is remarkably consistent.

“Think about it like this. The government buys a brand new car. Looks great in the car park. But nobody in the office has a driver’s licence, the roads to the office are not paved, and there is no gas station within 20 miles,” Mr Dunkley said. “That is what happens when you buy AI before you check whether your organisation can actually run it.”

The procurement moves forward without assessing whether the agency’s data is accessible, whether staff can verify AI outputs, whether existing processes are documented well enough to automate, or whether anyone has clear ownership of the decisions the AI is meant to support.

The tool arrives in an environment that cannot support it. Usage drops. The few staff members who try it either trust its outputs without verification or abandon it after the first confusing result. The agency reports the project as completed. The outcomes tell a different story.

Four ways government AI projects die before they start

AI does not fail randomly. It fails in patterns that are identifiable before procurement begins.

Data fragmentation is the most common trigger. Government agencies store information across multiple systems that were never designed to communicate. A ministry of health might hold patient records in one system, staffing data in another, and procurement records in a third. An AI tool purchased to “find efficiencies” across these systems cannot access half the information it needs. It ends up working within one data silo, delivering none of the cross-functional value it was bought to provide.

Staff readiness is the next problem. AI outputs sound authoritative regardless of whether they are correct. Mr Dunkley has a name for this: Synthetic Confidence Risk.

“Generative AI like ChatGPT does not automatically say ‘I am not sure about this one.’ It gives you the wrong answer with the same confidence it gives you the right one,” he said. “If the person reading that output does not know enough to spot the difference, the mistake goes straight through. And every time a wrong answer slips by and nothing bad happens immediately, you trust the tool a little more. That is how the damage builds.”

He calls that accumulation Cognitive Debt, and warns it compounds quietly until something breaks loudly enough for someone to notice.

Process documentation is another weak point. You cannot automate a process that has not been written down. Many government agencies run on institutional knowledge held by long-serving staff rather than documented procedures. When AI is introduced to “speed up” a workflow that exists in people’s heads rather than on paper, the tool has nothing structured to work with.

Then there is decision ownership. AI can recommend but it cannot be accountable. When a system generates a recommendation and nobody in the organisation has clear authority to act on or reject it, the result is a diffusion of accountability that makes it impossible to trace why a decision was made or who should have caught an error.

The hidden cost nobody talks about

Failed AI projects erode institutional trust in technology. Staff who have a negative first experience with AI tools become resistant to future attempts, even when the tools and conditions improve. Agencies that quietly shelve AI projects give ammunition to political opponents of digital investment, making it harder for the next administration to fund even well-designed programmes.

There is a regional dimension to this that gets too little attention. CARICOM nations are watching each other. A high-profile failure in one country discourages adoption across the bloc. A well-executed deployment in Jamaica or Barbados or Guyana, on the other hand, creates a template that neighbouring nations can study and adapt.

“When one Caribbean country messes up an AI project, it does not just waste that country’s money. It scares off the next ten countries from even trying,” Mr Dunkley said. “And that delay has a price too. We end up further behind, consuming AI systems built elsewhere by people who have never set foot in our institutions, designed for problems that look nothing like ours.”

He calls this structural gap Preparation Asymmetry: the widening distance between nations that build AI systems and nations that inherit the consequences without having participated in the design.

“We cannot close that gap by spending faster. We close it by spending smarter. And you cannot spend smart if you do not even know where you are starting from.”

We survey land before we build. Why not AI?

“Nobody builds a bridge without checking the ground first. Nobody rolls out a vaccination programme without an epidemiological assessment,” Mr Dunkley said. “But somehow, government after government is signing AI contracts without anyone first asking: can we actually use this?”

An AI assessment, he argues, will either confirm that an agency is ready to proceed or reveal exactly what needs to happen first. Either answer saves public money. One saves it by confirming a sound investment. The other saves it by preventing a premature one.

One free call could save your country millions

Mr Dunkley’s team is now offering free AI Readiness and ROI Assessments for government entities across all Caribbean countries. Each assessment produces a readiness rating, recommendations prioritised by impact, and a risk mitigation plan. There is no cost, no sales pitch, and no vendor lock-in.

When asked why free, Mr Dunkley did not hesitate: “Because it is your money. You pay taxes. Your government is about to spend some of that on AI. You deserve to know whether anyone checked if the thing will actually work before they signed the cheque.”

Government agencies can register at: https://starapple.typeform.com/starappleai4u

For enquiries: [email protected]

Read This, Then Call Your Representative

If you work in a government ministry, agency, or state body anywhere in the Caribbean, register for an assessment before the next procurement decision is made.

If you are a citizen, a taxpayer, or part of the Caribbean diaspora watching public funds being allocated from abroad, share this with someone who can act on it. Forward it to your permanent secretary, your director of IT, your minister’s office. Call your representative and ask one question: before we spend public money on AI, has anyone assessed whether we are ready for it?

The assessment is free. The cost of skipping it is not.

---

Adrian Dunkley is the founder of StarApple AI, the Caribbean’s first AI company, with over 15 years of applied AI experience across insurance, finance, and government. He has been recognised as Caribbean AI Innovator of the Year, IBM Mentor, Forbes Technology Council Member, AWS Grant AI Awardee, and EY Entrepreneur of the Year.