Caribbean AI Wrapper Economy: Warning Signs & Checks
Caribbean businesses, governments, and institutions are being pitched proprietary Caribbean AI at prices that can run 10x to 100x above the underlying model cost. Some of these products are legitimate. Others are US$20-a-month tools with a logo on top, and they carry compliance, continuity, and strategic risks the buyer usually cannot see until it is too late. This is the guide: what a whitelabelled AI actually is, the warning signs your vendor is selling a wrapper, how to check on the first call, and what your business risks by getting this wrong.
The Caribbean's AI procurement systems are being tested for the first time, and they are failing. Regional buyers are paying markups of 10x to 100x above raw model costs for what are, in many cases, thin resale of American, European, and Chinese foundation models. The pattern is now widespread across governments, banks, universities, and SMEs, and it is not going to fix itself.
What a whitelabelled AI actually is
Almost every AI product on the market is built on top of a small number of foundation models. Six model families power the majority of AI products sold into the Caribbean today.
Every Caribbean AI product worth buying is built on top of one of these, or a fine-tune of one of these, or an orchestration across several of these. There is no seventh option that is Caribbean-owned foundation infrastructure, and building one from scratch costs hundreds of millions of dollars and years of research. What separates a legitimate Caribbean AI product from a wrapper is what value the vendor adds on top of the model they chose.
A whitelabelled AI is a product that puts custom branding on one of these foundation models. When the wrapper is real, it fine-tunes on regional data the base model does not have, integrates with the buyer's systems, adds workflow the buyer would otherwise build internally, and prices at 2x to 5x the underlying cost to cover legitimate operational overhead. When the wrapper is a scam, it changes the login page, keeps the base model unchanged, adds no meaningful integration or fine-tuning, and prices at 10x, 30x, or 100x the underlying cost while marketing the product as "proprietary Caribbean AI."
The Caribbean is being pitched both types simultaneously. This guide is how to tell them apart before signing.
Where the money actually goes
Consider a typical US$500 monthly wrapper contract sold to a Caribbean SME for AI customer service. The vendor pitch describes "proprietary Caribbean AI trained for regional business." The actual cost structure is public, and any Caribbean SME can rerun the math in ten minutes with a calculator and the model provider's pricing page.
Two things follow from this cost structure. First, the wrapper vendor's margin is roughly 86% of the contract value, which is higher than a legitimate SaaS company's total revenue would be after all operating costs, and much higher than a fine-tuned or workflow-heavy AI product would earn. Second, US$18 of every US$500 contract flows to Silicon Valley regardless of who the wrapper vendor is. When the wrapper vendor is Caribbean-based, US$482 stays in the region, though it stays with the vendor rather than reaching the buyer's own AI capability. When the wrapper vendor is offshore, the entire US$500 leaves the region.
Across a mid-sized Caribbean enterprise's typical five to ten AI vendor contracts, and at government and public sector scale where individual contracts run into six and seven figures, the numbers move into tens of millions of US dollars per year being priced above what the underlying products justify.
Warning signs your vendor is selling a wrapper
The signs below are things a buyer can observe directly, without depending on what the vendor chooses to disclose. Any single sign is not conclusive on its own. Three or more together should end the sales conversation.
- The demo is a chat interface with no visible workflow. A real value-add product shows integrations with your systems, retrieval on your data, specific tool use, or a domain-specific interface built for a job role. A pure chatbox with a Caribbean flag in the header is almost always a wrapper. If the demo looks like ChatGPT with a different logo, it usually is ChatGPT with a different logo.
- Responses have the fingerprint of a known model. Ask the product the same question you would ask ChatGPT, Claude, Gemini, Mistral, or DeepSeek and compare answers side by side. Identical phrasing, identical structure, identical caveats such as "as an AI language model" leaking through under stress testing, and identical refusal patterns are all signs the wrapper is thin. The underlying model's personality shows through.
- Pricing has no clear relationship to usage. A vendor whose product has real per-query costs will price against them or disclose them. A vendor charging a flat US$500 per month for unlimited use of what is functionally the underlying API is either subsidising heavy users out of margin (rare) or pocketing the difference on light users (common). Flat monthly pricing across widely different usage patterns is a scaling signal from wrapper vendors.
- The product is marketed as "proprietary" but cannot name what is proprietary. Every legitimate whitelabel has something specific it owns: fine-tuned weights, a private dataset, a workflow, an integration pattern, a compliance layer. If the vendor cannot name that thing in one sentence, they are selling a rebrand. "Trained on Caribbean context" is not an answer. "Fine-tuned on 40,000 documents from these named sources" is an answer.
- Basic technical questions produce sales rather than engineering answers. Questions like "what is your prompt template," "how do you handle context length," "what is your evaluation methodology," "what happens when the underlying model updates," and "what is your fallback strategy if the provider deprecates a model" are ordinary product questions. A vendor who redirects them to marketing brochures does not have a product built by engineers.
- Pricing runs orders of magnitude above direct API access. The Caribbean SME being charged US$500 per month for a customer service AI that could be built directly on Claude Sonnet for US$18 per month in API costs is the canonical case. Any buyer can rerun that calculation in ten minutes using the model provider's public pricing page. Markups above 10x require the vendor to name what the extra markup buys. Above 20x, the vendor should be able to point to specific engineering, compliance, or vertical value that justifies the gap.
How to check on the first call
The warning signs above are what a buyer sees on the outside. The checks below are what a buyer asks for explicitly. A vendor selling a legitimate product will answer every one of these on the first call. A vendor selling a wrapper will delay, deflect, or refuse.
1. What is the underlying foundation model, and can you name it specifically?
Why this mattersA specific model name (Claude Sonnet 4.6, GPT-5.5-Pro, Gemini 3.5, Llama 3.3 70B, Mistral Large 2, DeepSeek V4 Flash, or a named fine-tune of one of these) tells you what you are actually paying for. A vendor who cannot or will not name the model is hiding it because disclosure would immediately reveal that the pricing is not justified by proprietary technology. There is no legitimate confidentiality reason to hide a foundation model from a paying customer under a signed NDA.
2. What Caribbean-specific data has this been trained or grounded on, and what is the size of that dataset?
Why this mattersThis is the operational definition of "regional value add." A legitimate Caribbean AI product will name the data: Jamaican legal documents, Trinidadian regulatory filings, CARICOM policy corpora, regional healthcare records where local licensing allows, Creole language sources with specific coverage. A wrapper will say "optimised for Caribbean users" or "trained on Caribbean context" without a specific dataset or size. If the vendor cannot name what data makes the product regional, the product is a base model with a flag.
3. What is the per-query API cost to you, and what is your markup?
Why this mattersA vendor who is confident in their value will disclose both numbers. A vendor who is not will refuse. The healthy markup range for a wrapper with real operational overhead is 2x to 5x total cost. A wrapper with real workflow, retrieval, or integration value can justify up to 10x. Anything above 10x needs to be justified by named engineering, compliance, or vertical value. Above 20x, the vendor should be able to point at specific artifacts that produce the additional value.
4. Where does my data flow, and which country's data protection law applies?
Why this mattersYour data goes to the wrapper vendor, and then to the underlying foundation model provider, and then to whichever cloud region hosts inference. A legitimate vendor answers with named regions, named data residency arrangements, named model provider policies, and named applicable law (Jamaica DPA 2020, Trinidad DPA, Barbados DPA, EU GDPR where relevant, HIPAA where regulated health data is involved). A wrapper vendor answers with vague assurances about "secure processing." The compliance risk sits on your business, not the vendor's, when regulators come asking.
5. Can I migrate to a different underlying model, and what happens if the provider raises prices or deprecates yours?
Why this mattersThe foundation model market is moving fast. Prices change quarterly. Models get deprecated on 12-month notices. New models make old ones obsolete for the same use case. A legitimate vendor documents the migration path, the pricing pass-through terms, and the deprecation notice period in writing. A wrapper vendor answers with "we will manage that for you," which usually means the buyer is locked in and will absorb whatever cost or capability changes the wrapper vendor decides to pass through.
The Caribbean AI Newsletter's earlier buyer's checklist covers a fuller ten-question version of the same exercise. These five checks are the minimum due diligence any Caribbean buyer should apply before signing.
The risks to you and your business
Getting AI procurement wrong is not just a matter of paying too much. Wrapper contracts create five categories of risk, and each becomes real on a different timeline. The financial risk is visible on the first invoice. The strategic risk compounds silently until a competitor with better AI economics starts winning your customers.
The most visible risk is that you are overpaying today. A US$500 per month contract for what should cost US$100 is a US$400 monthly loss. Across a year, that is US$4,800 per contract. Across a mid-sized Caribbean enterprise's typical five to ten AI vendors, the annual loss is US$24,000 to US$50,000 moving from the buyer to wrapper vendors without commensurate value. For a small SME, that is a full-time salary. For a government agency, it is a training programme that never gets funded.
Your wrapper vendor sits between your business and the foundation model. If the vendor changes pricing without notice, deprecates a feature you depend on, pivots away from your industry, or goes out of business, your customer service, back office, or content operation loses its AI layer overnight. Legitimate vendors document their model migration terms, contract escape clauses, and pricing pass-through policies in writing. Wrapper vendors typically do not, because doing so would expose the thinness of the layer they operate.
Every conversation, document, or record processed through the wrapper vendor moves through their infrastructure before reaching the foundation model. Where does that data sit? Which country's data protection law applies to it? Is the wrapper vendor's data handling policy compatible with the Jamaica Data Protection Act 2020, or Trinidad and Tobago's Data Protection Act, or Barbados' Data Protection Act, or the equivalent in your jurisdiction? If your business handles regulated data in healthcare, finance, legal, or telecommunications, a wrapper vendor's opaque data handling policy is a compliance liability that grows every month you remain under contract, and the liability sits on your organisation when regulators arrive, not on the vendor.
Every Caribbean business overpaying for wrapped AI is capability-outsourcing to an intermediary. Meanwhile, competitors building on the underlying model directly, or using a legitimate whitelabel with real integrations, or working with a regional AI company that does real fine-tuning are compounding capability at five to ten times the price efficiency you are. Over three years, wrapper users end up unable to catch up because the capability budget was spent on markup. The compounding effect is why this risk is more dangerous than the direct financial cost. A US$50,000 annual overspend is recoverable; a three-year capability gap against competitors is not.
When the underlying foundation model has a public incident, whether a safety failure, a bias event, a hallucination that reaches a customer, or a policy change that alters model behaviour, the wrapper vendor between you and the model may have no way to detect it, communicate it to you, or roll back before it reaches your customers. Your product using wrapped AI can fail in ways that reach your customers before you know what happened. Ownership of the AI system in the eyes of your customers stays with your brand regardless of who is behind the wrapper, and public incidents of AI failure in the region are already occurring on wrapper products where the buyer had no visibility into the underlying model.
The five risk categories are cumulative rather than alternatives. A Caribbean SME running a US$500 monthly wrapper contract is exposed to all five simultaneously, and the total exposure over a three-year contract term is materially larger than the sticker price would suggest.
Why the Caribbean is uniquely targeted
The wrapper economy is a global phenomenon. American, European, and Asian buyers are also being sold thin resale of foundation models at inflated prices. Three regional conditions overlap in a way that makes the Caribbean pattern harder to catch. Procurement systems were written for perpetual-license enterprise software and do not ask about foundation models, per-query economics, or model migration. Regional literacy about foundation model pricing is low, because the pricing is public but nobody has told buyers to look at it. Vendor sales cycles are calibrated to exploit both conditions, with polished RFP responses, monthly pricing that clears procurement thresholds, and language that borrows the vocabulary of legitimate AI companies.
These three conditions produce the pattern together. A Caribbean SME with a US$500 monthly AI contract has usually gone through a decision process that looked reasonable from the inside. The pitch cleared internal review because internal review was measured against the wrong yardstick.
What real Caribbean AI would build differently
The frustration behind this guide is not with the concept of building on top of foundation models. Almost every AI product worth buying is built on top of foundation models. The frustration is that the region is paying wrapper prices without receiving wrapper value, and the money is flowing to vendors who have not earned it.
A legitimate Caribbean AI company building on top of Claude, GPT, Gemini, Llama, Mistral, or DeepSeek does specific things that a wrapper vendor does not. It fine-tunes on regional data that a base model cannot access: Jamaican legal precedent, Trinidadian regulatory filings, Guyanese oil and gas documentation, Barbadian tourism operational data, healthcare records from Caribbean systems where local licensing allows. It builds retrieval systems over the buyer's own documents rather than relying on the base model's general training. It integrates with the systems the buyer is using rather than delivering PDFs. It publishes its architecture openly enough that a technically literate buyer can reason about what they are paying for.
The regional directory at caribbeanai.org catalogues companies operating at this level, alongside newer entrants still building. The Caribbean AI Risk Management Council has been working on a certification standard designed specifically to attest to what a vendor is actually selling, which is one of the missing pieces of infrastructure. Regional AI companies including Annotera in Jamaica, working on human-in-the-loop annotation and evaluation, and firms like itel investing in AI-driven training and workflow tools, show what building looks like when the goal is regional capability rather than regional resale.
What needs to change at the system level
1. Rewrite standard procurement templates to include AI disclosure requirements
Every Caribbean government, university, bank, and enterprise procurement office should update standard RFP language to require disclosure of the underlying foundation model, any fine-tuning or grounding data, per-query cost economics, data handling and applicable law, and model migration terms. Any vendor unwilling to answer these in a signed procurement response is not a vendor the institution should be signing a contract with. The template change is a two-week task that CARICOM, CARICAD, and national procurement offices could coordinate.
2. Establish a regional AI vendor certification framework with real disclosure requirements
The Caribbean AI Risk Management Council is developing certification standards intended to attest that a vendor has disclosed what they actually sell. Adoption across public sector procurement would make the certification the default filter Caribbean buyers apply before running their own technical review. The certification cannot force vendors to be legitimate, but it can make the legitimate vendors visible and put wrapper vendors on the defensive when they cannot produce equivalent documentation.
3. Build regional AI procurement training programmes for buyer-side officers
Caribbean SME owners and procurement officers should be able to buy AI competently without becoming AI engineers. A two-day training programme, delivered through regional institutions, could cover foundation model pricing across the six major providers, wrapper detection techniques, contract language, and the specific questions to ask before signing. Training of this shape has been delivered for other technology procurement categories and can be built inside a fiscal year.
4. Publish public transparency on regional AI procurement outcomes
Government AI contracts should be published with disclosure requirements including vendor, underlying model, pricing structure, and post-award audit findings. Public transparency creates the market signal that legitimate vendors can compete on. It also creates the record that shows, over three to five years, which vendors delivered what they promised and which delivered a rebrand.
The wrapper economy will keep pitching Caribbean buyers until the region either buys the pitch or builds the systems that make the pitch fail. Eighteen months of contract data across the region says the pitch is currently winning. This publication is arguing for the second option, because the first one is already running and the math has not improved.
Adrian Dunkley, Publisher, Caribbean AI Newsletter
Frequently asked questions
The wrapper economy has been priced correctly for a Caribbean market that does not yet know how to buy AI. The path out is to become the market that knows.Caribbean AI Newsletter · The Island AI Brief, 14 August 2026
Are You Being Scammed by a Whitelabelled AI? A Caribbean Buyer's Checklist
The extended ten-question version of the checks above, six observable red flags, and the pricing math a Caribbean SME can rerun in ten minutes.
Read the piece →Caribbean BPO After AI: Doomed, Different, or Transformed?
Jamaica has lost 12,000 to 20,000 BPO jobs from its 2023 peak of 60,000. Three cases on what happens next, argued at their strongest.
Read the piece →Ten Caribbean AI Companies the Region Needs. Some May Already Exist.
The ten AI companies the Caribbean needs at scale, from Creole voice AI to hurricane forecasting.
Read the piece →The Caribbean AI Newsletter is the leading daily source for artificial intelligence news, analysis, and practical insight for the Caribbean. The newsletter covers policy, workforce, cybersecurity, climate resilience, governance, regional innovation, procurement, and the founders building the region's AI sector.