Getting Started With Generative AI: A Field Guide for Caribbean Businesses
Issue 01TipsInaugural Issue

The Monthly Intelligence Report

Getting Started With Generative AI: A Field Guide for Caribbean Businesses

A pragmatic first guide for the Caribbean business owner asking whether to start with generative AI, which tool to pick, how to prompt it, and where to draw the line on data.

Dr. Marcus Charles·August 2024

Note from the President

Welcome to the first edition of The Monthly Intelligence Report. The Caribbean AI Association was incorporated this summer because the most consequential technology shift of our lifetimes was happening without us, and a region of forty million people deserves a seat at the table where its future is decided.

This newsletter is one part of how we intend to claim that seat. Each month you will find a feature from one of our contributors and a short note from this office. We will write about tools and tactics, about policy and ethics, about the worries that keep our members awake and the openings that should keep them excited. The voices will vary on purpose. The Caribbean does not speak with one accent, and neither should its association.

I am grateful to the founding members who put their names on our charter, to the early sponsors who funded the lights, and to every reader on this list. We are 612 strong as I write this paragraph. By December I hope we are five thousand.

Walk good.

Adrian Dunkley Founder and President, Caribbean AI Association


Feature

Getting Started With Generative AI: A Field Guide for Caribbean Businesses

By Dr. Marcus Charles

I have spent the last six months in twenty-three meetings with Caribbean business owners who all asked the same question in slightly different words. Should we be doing something with this AI thing yet, and if so, what? Here is the short answer. Yes. Start this week. The longer answer follows.

Pick one tool and learn it well. The three you should consider are ChatGPT (from OpenAI), Claude (from Anthropic), and Gemini (from Google). All three offer a free tier that is good enough to learn on, and a paid tier at roughly twenty US dollars per user per month that is good enough to run a department on. Do not subscribe to all three in the first month. Pick one, run every reasonable task through it for thirty days, and form an opinion. My recommendation for a non-technical owner today is ChatGPT, because the interface is the simplest and the ecosystem of how-to material online is the deepest. My recommendation for a technical owner is Claude, because its writing is more careful and its long-context performance suits the kinds of document work most Caribbean offices do every day.

Decide what data you will and will not give it. Your tool of choice should be configured to not train on your data. ChatGPT Team and Claude Team both default to this. Free tiers do not. If you are sending a client invoice or a customer record into any of these tools, you should be on a paid plan with the training opt-out confirmed in writing. Print the page that says so and file it. When your data protection regulator comes asking, and in some of our jurisdictions they will, you want the answer ready.

Find the three tasks that eat your week. For most Caribbean small businesses I have looked at, those tasks are: replying to customer enquiries, writing the same kinds of documents over and over, and summarizing long emails or PDFs that should not have been sent in the first place. Each of these is now twenty minute work that used to be two hour work. The compounding gain is real. A small hotel in Port Antonio that I worked with reclaimed nine hours of the owner's week within a month, by using a single tool to draft replies to booking enquiries, write event proposals, and summarize the daily housekeeping notes into a one page brief.

Write your prompts as if you were briefing a competent intern. Tell the model who it is talking to, what the situation is, what it is being asked to produce, what good looks like, and what to avoid. Three sentences of context will buy you twenty minutes of saved editing every time. The single most common mistake I see is two-word prompts that produce two-word answers and then a complaint that AI is overhyped.

Keep a human in the loop on anything that goes out under your name. Models still make confident mistakes. They will invent regulations, invent customer names, and occasionally invent entire products you do not sell. Read every output before it leaves your hands. The good news is that reading a draft is much faster than writing one.

Train your staff before they teach themselves badly. If you do not give your team an approved tool and basic guidance, they will use the free tier of whatever is fastest and they will paste your customer data into it. That is happening in your office right now whether you have authorized it or not. Better to spend a Saturday morning getting everyone on the same page than to find out from your auditor.

Budget for the next eighteen months. AI tools cost more than you think when you scale them, and less than you think compared to the work they replace. For a five person Caribbean business, plan on roughly one hundred US dollars per month for tools, plus the cost of one good training session. Compare that to a part time salary. The math is not subtle.

You do not need a strategy document to begin. You need a Tuesday morning, a willingness to try things, and the discipline to write down what worked. Send me what you learn. I read every reply.


Dr. Marcus Charles is a CAIRA member based in Bridgetown and consults with Caribbean enterprises on AI adoption.

Originally published in The Monthly Intelligence Report, August 2024.

Read every issue of The Monthly Intelligence Report

One feature, one President's note, every month. Written by the CAIRA contributor bench from across the Caribbean and the diaspora.