Talk with us
[email protected]+1 876 418 5388Kingston, Jamaica
AI · Data governance

AI will make your data look better. It won’t make it true.

AI is only as good as what goes in, and only as safe as where it goes.

AI has made it easy to produce work that looks finished. Whether the content is right is a separate question, and AI does not answer it.

The polish problem

A rough spreadsheet becomes a clean dashboard. Scattered notes become a confident board paper with headings, charts and a summary. The presentation is excellent.

Most people know the old computing phrase: garbage in, garbage out. AI has changed the second half. Garbage now comes out looking like gold. It reads well, it is well formatted, and it carries the authority of a finished product. Underneath, it is still garbage.

For boards and executives in Jamaica’s financial services and other regulated businesses, this creates two risks. The first is what goes into the tool. The second is where it goes.

Garbage in, gold out

AI works with what it is given. If the customer list has duplicates, the monthly report leaves out reversed transactions, or the sales figures come from three spreadsheets that do not agree, AI will not fix that. It will summarize it, chart it and present it with confidence.

The evidence is building:

  • Data quality decides whether AI projects survive. Gartner predicts that through 2026, organizations will abandon 60% of AI projects that are not supported by AI-ready data. In its survey of 1,203 data management leaders, 63% said their organizations either lack the right data practices for AI or are not sure.
  • Polished output still has to be checked. Researchers from Stanford and BetterUp Labs, writing in the Harvard Business Review, coined the term “workslop” for AI-generated work that looks good but lacks substance. In their survey of 1,150 full-time US desk workers, 40% had received it in the past month. Each incident took about two hours to sort out, an estimated US$186 per employee per month (Entrepreneur).
  • A respected name does not make the content right. In October 2025, Deloitte agreed to refund part of an AU$440,000 fee to the Australian government. Its report contained references that did not exist, a citation to a court judgment that was never made, and a quote wrongly attributed to a judge. The corrected version disclosed that an AI tool had been used (The Register).
Polish used to be a sign of effort. AI has removed that signal.

A well-prepared report used to suggest someone had done the work. Reviewers now have to check the substance, because the presentation tells them nothing.

Where the data goes

The second risk is less visible. When staff paste customer records, policy details or internal figures into an AI tool, that information leaves the company. Where it goes depends on the tool.

Many staff are not waiting for approval. Microsoft and LinkedIn’s 2024 Work Trend Index, a survey of 31,000 people in 31 countries, found that 78% of AI users bring their own AI tools to work. In practice, that often means a free or personal account, used on company information.

The difference between a personal account and an enterprise version is mostly in the contract, not the brand name. An enterprise agreement can rule out training on the company’s data, set where data is stored and for how long, require sign-in through company accounts, and give the company audit logs and administrative control. A personal account usually offers none of that to the employer.

In Jamaica, the Data Protection Act, 2020 places responsibility on the organization that controls the personal data, including when another party processes it. Personal data sent outside Jamaica must be protected to the standard the Act requires. Our understanding is that customer data entered into an unapproved personal AI account, with no contract and often stored abroad, would be difficult to defend to the Office of the Information Commissioner. Institutions regulated by the Financial Services Commission or the Bank of Jamaica would face the same question from their supervisor.

When assessing a tool, ask for evidence rather than assurances:

  • A signed data processing agreement that states whether your data is used for training, where it is stored, and how long it is kept.
  • Independent assurance such as ISO/IEC 27001 certification or a SOC 2 report for security, and, increasingly, ISO/IEC 42001 for how the provider manages AI.
  • Company control: sign-in through company accounts, audit logs, and the ability to remove access when someone leaves.

We covered approved tools and vendor contracts in more detail in Unapproved AI is a data breach waiting to happen. Banning AI outright rarely works. Staff who find it useful will use it on their phones. The practical answer is to approve a suitable tool, set it up properly, write a short acceptable-use policy, and train people on it.

What boards and executives should ask

That gives a board three simple checks.

The data going in

  • Which reports do we rely on most, and who checks that the figures behind them are complete and correct?
  • Do our systems agree with each other, or do we reconcile spreadsheets by hand?
  • If AI summarized our data tomorrow, would we trust the result?

The tools in use

  • Which AI tools are our staff using today, and which did we approve?
  • Do we have a signed agreement with each provider that says how our data is used and where it is stored?
  • Do we have a written policy on what staff may and may not put into an AI tool?

The output coming out

  • Who reviews AI-assisted work before it reaches a customer, a regulator or this board?
  • Do we require staff to say when AI was used to prepare a document?
  • Who is accountable when an AI-assisted report is wrong?

Real work first, then the tool

AI is useful. It can save hours on drafting, summarizing and analysis. But the value comes from the work underneath it: clean data, approved tools, and people who still read what comes out. Technology and substance have to move together. When they do not, AI makes weak work look finished, and that is harder to catch than weak work that looks weak.

How ready is your organization for AI?

Twelve questions for boards and executives, covering data, tools, oversight and execution. About four minutes.

Take the AI Readiness Check →

Sources

This article is general commentary and does not refer to any particular organization other than those named in the sources. It is not legal advice.

← Back to all insights