Caribbean households and firms that already pay their taxes keep facing higher GCT and VAT, because chasing everyone else one case at a time costs too much. That is the problem on the table when the Caribbean Organisation of Tax Administrators (COTA) holds its 27th General Assembly and Technical Conference in Georgetown, Guyana, from 27 to 31 July, hosted by the Guyana Revenue Authority. The theme is "Future-Ready CARICOM Tax Administration: Smart, Data-Driven and AI-Enabled for Sustainable Revenue." For the first time, AI and data analytics are the organising idea of the whole assembly.
AI can help CARICOM tax offices collect a larger share of what is already owed, using methods tested in countries with less digital infrastructure than Jamaica, Trinidad and Tobago or Barbados. The obstacle is the data those methods need, which most Caribbean administrations have not yet built to a usable standard. This piece covers the size of the gap, three tools that work, the evidence behind them, the main risk and the order in which to act.
Why the Georgetown agenda is different
COTA has met every year for decades, rotating among member states, and its communiques rarely travel beyond finance ministries. This year the whole programme is built around AI-enabled administration.
Georgetown is also hosting a High-Level Regional Meeting on Tax and Development with the Organisation for Economic Co-operation and Development (OECD). The OECD has spent a decade building rules for automatic exchange of tax information and for taxing multinationals, which mainly serve large economies with the capacity to use them. Pairing that meeting with COTA's AI agenda signals that Caribbean administrators want a say in those standards.
Guyana is a fitting host. Oil revenue has made the Guyana Revenue Authority one of the fastest-growing tax and customs operations in the hemisphere, and a tax base that has changed this much in a decade cannot be managed by picking audit cases on experience alone. Its neighbours can watch a live test.
How much revenue goes uncollected
The Economic Commission for Latin America and the Caribbean (ECLAC) estimated that tax evasion across Latin America and the Caribbean cost US$340 billion in 2016, equal to 6.7 percent of regional GDP, and that income tax evasion exceeded 60 percent in some countries. It linked this to high informality, weak institutions and taxpayers who see little benefit from the taxes they owe.
The Inter-American Development Bank (IDB) estimated the size of the informal economy, using electricity consumption and currency demand methods:
| Country | Informal economy, share of GDP |
|---|---|
| Jamaica | 35 to 44 percent |
| Guyana | 29 to 33 percent |
| Trinidad and Tobago | 26 to 33 percent |
ECLAC measures money that should have reached the treasury. The IDB measures activity that tax offices never see. Together they explain why finance ministries keep raising GCT and VAT on households and firms that already comply: chasing the rest one case at a time costs too much.
Three AI tools tax offices already use
Risk-based audit selection. A machine learning model (software that learns patterns from past cases) scores every return on dozens of variables, such as declared income against visible spending, sector benchmarks, past audit results for similar taxpayers and inconsistencies between related filings. Auditors still do the audits. The model decides which of thousands of returns they look at first.
Invoice matching. In missing trader fraud, a business claims VAT or GCT credit on a supplier invoice that the supplier never reported or paid. Matching every buyer's claim against every supplier's filing is impossible by hand. An automated system flags the mismatch as soon as both returns are filed, instead of months or years later if an audit happens to find it.
Customs risk targeting. Models compare each shipment's declared value, tariff classification and origin with historical trade data and, where available, the partner country's export records. A container declared well below comparable shipments is flagged for inspection, and customs officers stop inspecting by quota or at random.
Evidence from Senegal and Tanzania
A World Bank study of Senegal's corporate tax administration, led by economist Pierre Bachas, found that algorithmic audit selection outperformed the discretionary methods officers had used, finding non-compliant firms the manual process missed. In Tanzania, a risk-based case selection system raised the corrected taxable income found in audits by about 20 percent in its first year, according to research documented by the World Bank.
Neither country has more advanced digital tax systems than Jamaica, Trinidad and Tobago or Barbados, and both worked with small audit teams, patchy records and tight IT budgets. That makes them better guides for CARICOM than examples from North America or Western Europe.
The IMF's 2024 technical note on AI in tax and customs administration, and its guidance on analytics for compliance risk, make the same point: results depend on the quality of the data far more than on the algorithm. A model trained on incomplete or inconsistent records learns their blind spots and automates them.
What makes the Caribbean case different
- Small administrations gain most. No Caribbean tax office can hire enough auditors to review every anomaly. Risk scoring is the only practical way to raise audit coverage without a matching rise in staff.
- Border revenue matters more. Import-dependent island economies collect a larger share of revenue at customs than industrial economies, and the IMF identifies customs risk targeting as one of the more mature, lower-risk starting points. It may be the fastest early result.
- Informality needs different data. A model trained on formal filings is blind to the 26 to 44 percent of activity that never produces one. Reaching it means adding utility consumption, point-of-sale and mobile money records and property registries, the same proxies the IDB used to estimate informality, from the start.
The risk of targeting the wrong taxpayers
The Caribbean's informal economy is mostly vendors, market traders, gig workers and household businesses, many of whom stay informal because registering is slow and costly and brings no visible benefit. Their data is also the easiest to model. If the first visible result of AI in a CARICOM tax office is a wave of automated assessments on small traders while large corporate and cross-border evasion goes untouched, people will rightly call it unfair.
The IMF's guidance treats safeguards as part of the rollout. In practice that means publishing the general criteria behind risk scores, giving flagged taxpayers a clear appeal before an assessment becomes final, and pointing early deployments at large corporate, cross-border and customs cases, where the Senegal and Tanzania evidence is strongest.
The weakest part of this argument is the informal economy itself. The evidence from Senegal and Tanzania concerns registered firms, and no one has yet shown that AI brings informal businesses into the tax net without simpler registration and lower compliance costs alongside it.
What CARICOM tax offices should do after Georgetown
- Commissioners-general: within six months, audit the completeness of filing and invoice data for one tax type, such as GCT or VAT, and publish the share of returns filed electronically with machine-readable invoices. That number shows whether a model can be trained yet.
- Customs departments: start with risk targeting on imports, using declaration data already held in the customs system, and measure the additional duty collected per inspection against the previous year.
- VAT and GCT units: run an invoice-matching pilot on the largest 500 registered businesses before extending it to smaller firms.
- Finance ministries: pass or update rules that set out how risk scores are used, how taxpayers can appeal and which third-party data, such as utility or mobile money records, the tax office may use.
- COTA members: agree at Georgetown to design one shared risk-scoring platform that each administration adapts to its own law, and ask the IDB, the Caribbean Development Bank and IMF technical assistance programmes to fund the pilot.
Frequently Asked Questions
What is COTA and who belongs to it?
COTA is the Caribbean Organisation of Tax Administrators, which brings together the tax departments of CARICOM member states and some other Caribbean territories. It holds an annual General Assembly and Technical Conference and shares training and practice among members. It has no power to set tax law, so any AI platform it designs still has to be adopted by each government under its own legislation.
What customs data do Caribbean administrations already hold?
Many Caribbean customs departments, including those of Jamaica, Trinidad and Tobago and Guyana, run ASYCUDA World, the customs management system developed by UNCTAD. It records every declaration's value, tariff classification, origin and importer, which is the raw material for risk targeting, and it includes a risk management module. That makes customs a practical first project, because the data is already structured and in one place.
Can tax authorities use utility or mobile money records?
Only where the law allows it. Tax statutes usually give the revenue authority power to request information from third parties, and data protection laws such as Jamaica's Data Protection Act 2020 generally permit processing needed for tax collection, but the scope differs by country. Before building a model on third-party data, an administration should get a legal opinion and publish which data sources it uses.
What should a taxpayer do if they think an automated risk score led to an unfair assessment?
File an objection before the deadline stated on the assessment notice, and ask in writing for the grounds of the assessment and the records it relied on. Keep copies of invoices, bank statements and correspondence. If the objection fails, most Caribbean tax laws allow a further appeal to a separate appeals body or court, and a tax adviser or attorney can confirm the route and time limits.
Where can Caribbean tax officials get training on these methods?
The IMF's Caribbean Regional Technical Assistance Centre (CARTAC), based in Barbados, provides technical assistance and training to Caribbean revenue and customs administrations. The Inter-American Center of Tax Administrations (CIAT), whose members include several CARICOM administrations, runs courses and publishes practice guides on compliance risk management. The World Customs Organization also offers training on customs risk management and data analytics.