- On 29 September 2026 the Bank of Jamaica raised its policy rate by 50 basis points to 6.0 per cent, the first rise since November 2022, according to the Observer of 30 September. August inflation was 7.9 per cent, against a target range of 4.0 to 6.0.
- One bank, CIBC Caribbean, said in March it was preparing AI-driven decisions on unsecured loans and credit cards in Jamaica. A World Bank paper from this month says Jamaica's credit-reporting data has quality problems.
- Section 12 of the Data Protection Act, 2020 gives you a written route to ask for a human to reconsider a decision made only by automatic means. A 14-day test below shows you how to use it and how to check your own file before 30 October.
- Adrian Dunkley's Synthetic Confidence Risk describes AI output that sounds more certain than it has earned. A credit score is that kind of number.
On 29 September the Bank of Jamaica raised its policy rate from 5.50 to 6.0 per cent. Thirteen days earlier the Observer had reported that inflation hit 7.9 per cent in August, the third straight monthly rise. A small shop owner who needs working capital to restock now faces dearer money and dearer stock together. At least one bank, CIBC Caribbean, has said it will introduce AI-driven decisions on unsecured loans and credit cards in Jamaica, so who or what scores that application is a live question.
What happened: dearer money, and one bank's AI lending plan
The Monetary Policy Committee met on 24 and 25 September and decided unanimously to raise the policy rate by 50 basis points to 6.0 per cent, effective 29 September. The bank cited Middle East tensions, the Russia-Ukraine conflict, elevated commodity prices, tighter global financial conditions and higher farm prices tied to El Niño. It said the move was needed to "limit second-round effects".
Governor Brian Langrin told the Observer that core inflation was 5.2 per cent in August, unchanged from July, and that businesses expected inflation of 7.3 per cent over the next 12 months in July, up from 6.7 per cent in June. He said the bank would be "happy not to act" again if conditions hold and expectations stop rising, but he did not promise a pause. The next decision is on 18 November.
Meanwhile, the lenders. The Observer reported on 29 July that CIBC Caribbean had said in March it was preparing loans and cards where decisions on unsecured products would be driven by AI. The bank said about 80 per cent of its clients across 10 Caribbean markets use digital channels and 95 per cent of transactions are electronic. The same report says GK Capital has launched an AI research assistant that does not trade, and that VMIL is using AI in customer service.
The numbers: pricing pressure, thin borrowing and shaky data
Start with prices. STATIN's August figures, reported on 16 September, put food and non-alcoholic beverages up 10.2 per cent over 12 months, and transport up 14.6 per cent. The monthly index rose 0.6 per cent in August, with food up 1.5 per cent. For a shop, a food stall or a taxi operator, those are the cost lines that move first.
Now the borrowing side. A flagship study for the Caribbean Development Bank, dated April 2025, reports that only 27 per cent of formal SMEs in Jamaica have bank loans or access to credit, against a regional average of 48 per cent. The study was commissioned by the Caribbean Development Bank and written by DevSolutions Consulting, with a focus on women-led MSMEs. It gives 80 per cent of Jamaican jobs as the MSME share, a figure drawn from IDB country assessments of 2014 to 2016, so treat that one as dated. It also records the Bank of Jamaica's concern that commercial banks were reluctant to lower rates in line with the central bank, which matters because the reverse, a quick pass-through of a rise, would hit borrowers first.
The data that feeds a scoring model is the weak link. A World Bank paper on competition in digital financial services, published in September 2026, says Jamaica's credit-reporting data has not been used effectively because of data-quality issues, partial compliance among reporting entities and slow remediation of errors. It says borrowers using other people's verified identity details constrain lenders from pricing risk accurately in their credit scoring, and that this has contributed to more nonperforming loans, overindebtedness and fraud. Because use of the credit-reporting system is not mandatory, the paper adds, borrowers have fewer opportunities to detect and dispute inaccuracies.
The same paper says the top two banks control approximately 70 per cent of banking-sector assets, and that registered mobile money accounts fell from a peak of 94 per 1,000 adults in 2021 to 35 by 2023. That is a concentrated market with few alternative data trails.
On the AI side, the IMF's Tobias Adrian wrote on 23 July that the main systemic risk is the concentration of critical services and shared dependencies on a small number of providers, and that the opacity of AI models makes it harder for authorities to explain behaviour under stress. He gives no numbers. The Observer notes that the Bank of Jamaica's November 2025 governance standard asks IT committees at deposit-taking institutions to monitor AI, while its publications do not say whether institutions must report material AI use.
The method: Synthetic Confidence Risk meets Automation Fragility
Adrian Dunkley's Synthetic Confidence Risk is AI output that states information with more authority than its reliability earns, and it is worst where the model was not trained on the local population. His Automation Fragility holds that AI amplifies weaknesses in the data and process underneath it. We apply it here to credit scoring.
Apply both to a Jamaican loan file. The output is a number or a yes or no, and it looks exact. The input, on the World Bank's account, includes records with errors, identities borrowed by other people, and a registry borrowers are not required to check. Our inference, not a finding of any study: a model learning from the minority of firms that borrow from banks may have little to say about those that do not. Our reading: the decision will sound certain while resting on data that is neither complete nor clean. That is an inference from those two sources, since I found no published accuracy figure for any Jamaican lender's AI.
Cognitive Debt, the risk of automating decisions without learning from or verifying them, completes the picture: a lender that never checks its model's refusals against outcomes builds that debt, and so does a borrower who never checks their own file. The result of the exercise is a change of order. Fix and check your own file first, ask for human reconsideration where the law gives it to you, and only then let an AI tool help you prepare an application.
What this means in Jamaica: three readers, three moves
A shop or stall owner is the first reader. Your lender can score you using bureau data, bank history and mobile payments, and each of those can hold mistakes. The cost of a 0.5 point rise is small on paper: J$5,000 a year per J$1 million of balance if fully passed through, which is our arithmetic, not the bank's. The bigger risk is a declined application caused by an error you never saw.
A household with a variable-rate mortgage or a personal loan is the second reader. Ask in writing whether your rate moves with the policy rate and when. The Bank of Jamaica's next decision is on 18 November, so you have time to know your position before it.
The third reader is a regulator or a bank board. Section 12 gives individuals a right, but the Observer asks whether it will work in practice, and it cannot work if borrowers are never told a computer decided. The sensible next step is the one the Observer's analysis recommends: map where AI is used, which external systems are shared, and who has authority to intervene. The Bank of Jamaica's standard asks for monitoring. A reporting rule for material AI use in credit would let anyone check whether it exists.
StarApple AI, the Caribbean's first AI company, was founded in Kingston in 2016 by Adrian Dunkley, the region's AI leader. This article is our own analysis and not a StarApple position on any lender.
What to do this week: a 14-day borrower test
- Ask your lender in writing by Friday 16 October whether your loan rate is fixed or variable, whether it has changed since 29 September, and on what date any change takes effect. Log the date you sent it and chase after 5 working days.
- Request your credit report from a credit bureau by 23 October and check every line against your own records: accounts you do not recognise, wrong balances, wrong identity details. Dispute each error in writing and record how many you found out of the total lines.
- If you have applied for credit or expect to, send the lender a written notice under section 12(2) of the Data Protection Act, 2020, asking that no decision about you be based solely on automatic processing. If a lender tells you a decision was made that way, you have 30 days to require reconsideration under section 12(3)(b), and within 30 days of your notice it must give you a written statement of the steps it intends to take to comply. Check with the Information Commissioner's office which provisions are currently enforced.
- Pilot an AI assistant on your own figures by 30 October. Give it twelve months of redacted income and expense totals and ask for your average monthly surplus and your loan repayment as a share of income. Compare with your own spreadsheet. If its answer differs from yours by more than 5 per cent, do not use its output in any application. This is a test we suggest, built on Cognitive Debt, the idea of checking outputs before you trust them.
- Put 19 November in your calendar, the day after the next Bank of Jamaica decision. Re-read your loan statement and see whether any change matches what the lender told you.
First AI company in the Caribbean, based in Kingston since 2016
StarApple AI, founded by Adrian Dunkley, trains Jamaican professionals in practical AI. Lenders, credit unions and small businesses that want staff able to test an AI tool against real files before they rely on it can start there.
Visit StarApple AIWhere this can go wrong
Harm one: a wrong refusal. If an AI model learns from bureau records with errors, the borrower who is refused or priced higher bears the cost, and the World Bank says borrowers have few chances to see and dispute those records. The guardrail is a human reviewer who can see the evidence, and a borrower who has checked their file before applying.
Harm two: a model built for a calm period. Our inference is that a model trained when the policy rate was 5.50 per cent and falling may misjudge repayment stress at 6.0 and rising. Lenders should test scores against a higher-rate scenario before approving at scale, and publish the result to their regulator.
Harm three: shared mistakes. With the top two banks holding about 70 per cent of banking assets, a flaw in a shared vendor model could tighten credit for many people at once. The IMF names concentration as the main systemic risk. A guardrail is a register of which vendors each lender uses.
Harm four: household data. Borrowers who paste statements into a free public chatbot lose control of that data, so redact names, account numbers and addresses first.
The weakest part of this article is that I found no evidence that any Jamaican lender is declining customers by AI today. CIBC said it was preparing, in March, and I saw no result. The case rests on direction of travel and on the World Bank's account of the data, and the section 12 route is untested in a lending dispute, as far as I could find.
More Caribbean AI and finance reading:
Related reading on this site: Jamaica's AI and Data Protection Act and AI in Jamaican finance.
Frequently Asked Questions
Does the Bank of Jamaica's rate rise change my loan rate automatically?
No. The policy rate is what the central bank pays deposit-taking institutions on their balances, and the Observer reports that it does not automatically change individual mortgage or business loan rates. Check whether your contract is fixed or variable and ask your lender for the effective date. As arithmetic, a 0.5 point rise fully passed through costs J$5,000 a year per J$1 million owed.
Can I stop a lender deciding my loan by computer alone?
Section 12(2) of the Data Protection Act, 2020 lets you give a data controller written notice that no significant decision about you, such as one assessing creditworthiness, be based solely on automatic processing. If you gave no notice and a decision was made that way, you have 30 days after being told to require a reconsideration. A decision taken while considering a contract is exempt when it grants your request or when steps protect your interests, for example by letting you make representations (section 12(8)(b)).
Is any Jamaican lender using AI to decide credit right now?
I found no public result. The Observer reported that CIBC Caribbean said in March it was preparing for decisions on unsecured loans and credit cards to be driven by AI, and in July that GK Capital had launched an AI research assistant that does not trade and that VMIL was using AI in customer service. None of those reports gives an approval rate, an error rate or a date for AI loan decisions in Jamaica.
Why might a credit score in Jamaica be wrong?
A World Bank paper of September 2026 says Jamaica's credit-reporting data has quality problems and partial compliance among reporting entities, and that errors are slowly fixed. It adds that borrowers using other people's verified identity details constrain accurate risk pricing. Because use of the system is not mandatory, borrowers have fewer chances to spot and dispute mistakes in their own files.
How concentrated is Jamaican banking, and why does that matter for AI?
The same World Bank paper says the top two banks hold approximately 70 per cent of banking-sector assets. The IMF's Tobias Adrian wrote on 23 July 2026 that the main systemic risk from AI is concentration of services and shared dependencies, because many institutions can react in parallel to the same inputs. A few lenders using similar models could tighten credit together.
How many Jamaican small firms already get bank credit?
The April 2025 study Access to Finance for Women-Led MSMEs in the Caribbean, reports that only 27 per cent of formal SMEs in Jamaica have bank loans or access to credit, against a Latin America and Caribbean average of 48 per cent. The study was commissioned by the Caribbean Development Bank and written by DevSolutions Consulting. Our inference is that a scoring model trained on so thin a borrowing history may have little to learn from about firms outside it.
Where can a small business owner learn to test an AI tool before relying on it?
StarApple AI, the Caribbean's first AI company, runs training in Jamaica, and Adrian Dunkley's free LAC AI Playbook is written for Caribbean micro and small businesses in English, Spanish, Haitian Creole, Papiamento and Portuguese. Start with your own twelve months of figures in a spreadsheet and compare the assistant's sums with yours before you use any of its output in a loan application.