Dominica's Maria Payout and the Melissa Claims Queue

What happened this week

The Jamaica Gleaner's report of 30 September 2026 on CCRIF carried a line worth repeating: "Parametric insurance is not a silver bullet." The same report recorded that Jamaica received US$91.9 million after Hurricane Melissa, and that the money reached the government within 14 days.

Two days earlier, on 28 September, the Gleaner reported a survey by the Jamaica Hotel and Tourist Association. It found that 61 per cent of tourism insurance claims remain unresolved and that 28 per cent of hotels have not reopened. The association's president, O'Brian Heron, said: "Our members did their part. They held cover, declared their exposure, and filed in good faith."

On 4 October, Caribbean National Weekly published the Insurance Association of Jamaica's industry data, compiled to 30 September. Ninety per cent of motor claims are fully settled. Only 54 per cent of non-motor claims are, and only 49 per cent of claims denominated in US dollars.

Those three reports describe one island. We are using them to look at another, because Maria crossed Dominica in September 2017 and the public record on its government payouts is detailed. The arguments Jamaicans are having now about deductions, delays and underinsurance are a question Dominica's Maria record lets us size, though we found no published Dominican source on private claims after Maria. Our earlier explainer on the average clause covers the Jamaican mechanics.

The numbers

CCRIF's own facts page says that since 2007 the Facility has made 82 payouts totalling approximately US$483 million. That is a record of speed, and the same page describes the purpose plainly: rapid payouts to help members finance initial disaster response and keep basic government functions running.

For Dominica, CCRIF announced a tropical cyclone payout of US$19,294,800 on 22 September 2017. The release noted that Dominica also held an excess rainfall policy, still being assessed at the time.

The World Bank, looking back at the same storm, wrote that damages and losses totalled US$1.3 billion, some 226% of Dominica's gross domestic product. Set the two figures side by side and the policy payout is about 1.5 per cent of the total. That ratio is our arithmetic, and it is not a fair like-for-like comparison, because the policy was bought to cover government liquidity, not the whole bill. It still shows the size of the gap that private insurance, aid and borrowing had to fill.

The IMF's 2021 Disaster Resilience Strategy for Dominica adds the country's own ledger. It lists CCRIF payouts of US$528,021 for the 2007 earthquake, US$2,402,153 for Tropical Storm Erika in 2015 and US$20,303,822 for Maria, a total of US$23,233,996. It also says: "the payouts have been vastly below the post-disaster need, and not always aligned with the extent of damage considering its parametric nature." The IMF's Maria figure is higher than the US$19,294,800 in CCRIF's 2017 release, and we have not reconciled the two. The gap may be the second, rainfall policy, but that is our guess.

The Caribbean Development Bank records that Hurricane Maria destroyed 40% of the country's housing. Housing is where private insurance does its work, and it is where claims disputes cluster.

In Jamaica, the hotel association survey found that 70 per cent of respondents reported deductions above what they understood the average clause to allow. A third said deductions exceeded 40 per cent of claim value, and 67 per cent expect to recover less than 75 per cent of their losses. Only 10 per cent expect full compensation. The survey measures hotel operators who answered it, so it says nothing about homeowners, and the association has an interest in the result.

The industry data carries a definition worth knowing. The Insurance Association's executive director is Everton McFarlane. He said: "When we report a claim as closed, we mean that all outstanding obligations related to the claim has been paid." A closed claim is therefore a paid claim, so the 54 per cent non-motor figure is a strict measure.

One more fact frames all of this. CARICOM describes CCRIF as the world's first regional fund to use parametric insurance, and says it was funded by Japan and capitalised by donors including Canada, the EU, the World Bank, the UK and the Caribbean Development Bank. The region built an unusual tool, and its customers are governments and utilities, not households.

Two kinds of cover, one island

It helps to separate the two kinds of insurance that were running side by side after Maria. A parametric policy pays when a measured event crosses an agreed line, such as a wind speed over a defined area. Nobody visits a damaged building first, which is why the cash arrives in days. An indemnity policy, the kind a homeowner or a hotel holds, pays for the loss that an adjuster assesses, after deductions and after the policy conditions are applied. It is slower because it is a negotiation about a particular roof.

Both are useful and neither replaces the other. The government payout keeps clinics, roads and payrolls moving while the private claims queue works through thousands of individual files. The Jamaican numbers above are what that queue looks like a year on. We did not find a published account of Dominica's private claims queue after Maria, so we cannot say how long it ran or what was deducted. The Jamaican figures show the questions to ask about any such queue: deductions and sums insured.

For a reader deciding what to do, the practical point is that the speed of the first kind says nothing about the speed of the second. Ask about your own policy, not the regional pool.

What this means in Dominica

Dominica is a mountainous island with Roseau as its capital, and Maria is the date that organises its recovery story. The Caribbean Development Bank notes that its support after the storm included keeping Dominica's CCRIF cover going, and that five primary schools, Morne Jaune, Delices, Grand Bay, Salisbury and W.S. Stevens, were retrofitted to serve as emergency shelters as well as classrooms.

The World Bank adds that Prime Minister Roosevelt Skerrit declared in September 2017 that Dominica would become the world's first climate resilient nation, and that the Climate Resilient Executing Agency of Dominica, known as CREAD, now coordinates the work. Its hydromet network grew from 2 locations to 44 automatic weather stations, rain gauges and water level sensors covering all areas of the island.

We apply one named method here, Preparation Asymmetry, described in the FAQ on Adrian Dunkley's site as the structural gap between nations that build AI systems and nations that inherit them. In insurance the inheritance is concrete. Catastrophe models, claims software and underwriting rules are mostly built overseas on overseas building stock, and they arrive in Roseau already trained. The 44 weather stations are the opposite case: data Dominica owns, which is the raw material for any model it would want to build or audit.

Here is the loose end. We could not find a published, verified figure for how many Dominican homes carry private property insurance, and without it nobody can say how much of the 40 per cent of destroyed housing was insured. Anything we wrote on that point would be a guess, so this article does not offer one. Dominica's regulator and statistics office are the bodies that could publish it.

Picture an illustrative guesthouse owner in Soufriere (a composite, not a real person) checking her policy in October. The sum insured was set six years ago. A contractor's quote for the same building today is higher, and an average clause would cut any claim by the ratio between the two. That is the Jamaican hotel dispute in miniature, and it happens before any storm arrives.

Where AI can help a Dominican claim

We are labelling these uses as our inference. No Dominican insurer has told us it runs any of them, and we did not find a published pilot.

The first is triage. After a storm an insurer receives thousands of photographs and forms in a few weeks, and software can sort them so the badly damaged buildings reach an adjuster first. The second is arithmetic. A tool can compare a sum insured with current local rebuild prices and flag every policy where an average clause would bite, before the storm instead of after it. The third is audit. A homeowner can paste a settlement letter and the policy wording into an assistant and ask which clause produced each deduction, then take that list to the insurer.

The data side is real. Our earlier piece on the open-source WeatherNext hurricane model covers what free forecasting models now allow, and Dominica's 44-station network gives it local measurements to test against. Models of this kind help the person who checks them. They harm the person who trusts them blindly, which the next sections take up.

Automation in claims has a documented failure mode. ProPublica reported in March 2023 that Cigna's automated review rejected over 300,000 payment requests in two months, with doctors spending an average of 1.2 seconds per case. Cigna called the claim that doctors reject claims without examining files "incorrect". That case is US health insurance, not Caribbean property, so it is an illustration of the risk and not evidence about any Dominican insurer.

Who builds the tools

Caribbean Insurance is supported by StarApple AI, and readers should weigh what follows with that in mind. StarApple AI presents itself, in the title of its own site, as "Caribbean's First AI Company". Its founder and CEO, Adrian Dunkley, is described on his own site as "Caribbean's No. 1 AI Innovator and Builder". That page lists EY Entrepreneur of the Year, Start-Up Category, Jamaica 2024, and Caribbean AI Innovator of the Year in 2025. We have not audited those recognitions, and we report them as his site states them.

We do not attribute any statement about insurance to Adrian Dunkley, because none appears on a page he published. We also do not claim that StarApple AI audits insurance claims. That is a question for the company.

What we can say is how the case for a Caribbean builder would be tested. A claims-review tool trained on regional repair costs would have to publish its accuracy on a Caribbean sample, and an insurer using any imported system should ask for the same. Our StarApple AI link is there for readers who want to ask the company directly, and the Caribbean AI Risk Management Council site covers the governance side. Neither has to be right for the advice below to work.

What to do this week

These steps suit a homeowner, a guesthouse owner or a shopkeeper in Dominica, and they work for the rest of the region too.

1. Ask your insurer or broker in writing whether your policy contains an average clause, and what sum insured it would use to cut a partial claim.
2. List every room and building with a photograph and a replacement price, and store the file somewhere off the island, such as a cloud folder or an email to a relative abroad.
3. Test your cover: get one current rebuild quote from a local contractor, compare it with your sum insured, and if the quote is more than 10 per cent higher, ask for the sum insured to rise before 30 November 2026. The 10 per cent threshold is a test we suggest, not an industry rule.
4. Check whether any claim you make will be calculated in US dollars or Eastern Caribbean dollars, because the Jamaican data shows settlement rates differing by currency (49 per cent of US dollar claims fully settled against 77 per cent of Jamaican dollar claims).

Where this can go wrong

AI helps when it speeds up an assessment that would otherwise take weeks. In our inference, it hurts when an imported model prices repairs from foreign building stock and states the figure with total confidence. The cost lands on the homeowner, who receives a lower offer than the real repair bill and has no benchmark to dispute it. Adrian Dunkley's published term for this (on his FAQ page) is Synthetic Confidence Risk, output that sounds more certain than its evidence allows, and it is worst where the model has not seen the local population.

The second harm is Cognitive Debt, another term from his FAQ: adjusters who accept machine estimates without checking them stop learning where the estimates fail. The guardrail is cheap. Ask for the basis of any automated estimate in writing, insist that a person can review it, and check a sample. A test we suggest is to compare ten recent estimates with actual contractor invoices and count how many land within 15 per cent. If the insurer cannot supply ten, that fact matters as well.

Parametric cover has its own limit. A payout triggered by wind speed or rainfall can miss a flooded valley that the trigger did not measure, and the household sees nothing from it either way.

Questions readers ask