Two people open the same artificial intelligence (AI) assistant to tackle the same task. The first types, “help me with my budget”. Back comes a generic lecture about tracking expenses that could have been written for anyone, anywhere. The second types a paragraph — who the AI should be, what numbers it’s looking at, and exactly what they want back — and 90 seconds later has a board-ready variance commentary with the three biggest drivers named and quantified.

Same tool. Same subscription. The gap between them is not the technology. It is the prompt.

Over the last three weeks this column has shown what AI can build for you — formulas, a cash-flow forecast, a clickable dashboard. This week is about the skill underneath all of it: how to ask. Get this right and every tool you already pay for — ChatGPT, Claude, Copilot — gets several times more useful overnight. Get it wrong and you’ll keep concluding, unfairly, that “AI isn’t that impressive”.

The anatomy of a good prompt

A weak prompt is a keyword. A strong prompt is a brief — the kind you’d give a sharp new hire. Three things turn one into the other.

Role. Tell the AI who to be. I like to start with “You are a world-leading…” — a world-leading management accountant, a world-leading copywriter, a world-leading tax advisor. It sounds almost too simple, but naming the expert pulls the answer up to that level instead of leaving it at the average of everything ever written on the internet.

Context. Give it the situation. What’s the business, what are the numbers, who is the answer for, what does “good” look like here. The AI cannot read your mind or your files unless you tell it — and the single most common reason an answer disappoints is that you knew something important and didn’t say it.

Specific output. Say exactly what you want back, and in what shape. A 150-word summary. A table with three columns. Five subject lines. A polite first-draft email. “Help me with my budget” has no output; “give me a five-line variance commentary naming the three largest movers” does.

The anatomy of a good prompt: role, context and a specific output turn a keyword into a brief. (Branded graphic by PGH Consulting, LLC)

Bad prompt: “Write something about our new service.”

Good prompt: “You are a world-leading marketing copywriter. We’re a Kingston accounting firm launching a fixed-fee bookkeeping package for small businesses at J$35,000 a month. Write three Instagram captions under 40 words each, friendly and professional, each ending with a clear call to book a free consultation.”

The first makes the AI guess. The second makes it deliver.

“Help me with my budget” versus a real brief — the same tool, minutes apart, on a Kingston distributor’s numbers. (Branded graphic by PGH Consulting, LLC (illustrative demonstration figures, not client data).)

Ten prompts to steal this week

Copy these, swap in your own details, and watch what comes back. Each one follows the same Role–Context–Output pattern.

1. The accountant — “You are a world-leading management accountant. Here is my month-end profit-and-loss versus budget [paste the figures]. Write a five-line variance commentary for our directors: name the three largest movers, quantify each in dollars and per cent, and suggest one question I should ask the department head responsible for each.”

2. The FP&A analyst — “You are a world-leading financial planning analyst. Using these last 12 months of actuals [paste], build a simple driver-based forecast for the next quarter. State every assumption you make as a bullet list, flag the two line items most sensitive to error, and give me a best-case, base-case, and worst-case revenue number.”

3. The business owner — “You are a world-leading small-business advisor. My shop’s monthly revenue is J$1.8 million, costs are J$1.5 million, and I’m deciding whether to hire a second salesperson at J$120,000 a month. Walk me through the break-even maths in plain English, list three risks I may be overlooking, and tell me what to track for 60 days before I commit.”

4. The bookkeeper — “You are a world-leading bookkeeper. Here is a list of bank transactions with descriptions [paste]. Suggest a sensible expense category for each, flag any that look like duplicates or personal spending, and list the five transactions you’re least sure about so I can review them myself.”

5. The sales lead — “You are a world-leading B2B sales writer. We sell point-of-sale systems to Jamaican retailers. A prospect with three stores asked for a proposal. Draft a one-page proposal with a short problem statement, three benefits in their language, simple pricing tiers, and a clear next step. Keep it under 350 words.”

6. The marketing lead — “You are a world-leading marketing strategist. We’re a café in Mandeville launching a weekday breakfast deal. Give me a two-week content plan: seven post ideas, the best day and time to post each, one caption per post, and three hashtag sets aimed at a local Jamaican audience.”

7. The HR manager — “You are a world-leading HR professional familiar with Jamaican workplace norms. Draft a clear, fair remote-work policy for a 20-person firm: eligibility, core hours, equipment, and data-security expectations. Keep the tone supportive, not legalistic, and mark anywhere I should have a local attorney review the wording.”

8. The operations manager — “You are a world-leading operations consultant. Here are rough notes from our weekly team meeting [paste]. Turn them into clean minutes: decisions made, action items with an owner and a due date for each, and a short list of open questions still needing a decision.”

9. The tax professional — “You are a world-leading Jamaican tax advisor. Explain, in plain language a small-business owner can follow, how General Consumption Tax (GCT) registration and filing works once a business crosses the registration threshold — what they must charge, what they can claim, and the filing rhythm. Add a ‘verify this with Tax Administration Jamaica (TAJ) or your accountant’, note where rules may have changed.”

10. The audit professional — “You are a world-leading internal auditor. I’m testing the purchase-approval control for a small company. Draft a short testing plan: what evidence to request, a sensible sample size for roughly 400 monthly transactions, three exceptions to watch for, and how I should document a pass or a fail in my workpaper.”

What not to do

Five habits quietly sink most prompts. Don’t be vague — “make it better” gives the AI nothing to aim at, while “make it 30 per cent shorter and more formal” gives it a target. Don’t pile ten unrelated requests into one message; the AI will do all of them at half quality, so take them one at a time. Don’t assume it knows your numbers, your clients or your last email — if you didn’t paste it, it can’t see it. Don’t accept the first draft as the finished article; it’s a starting point, not a ruling. And don’t ask it to invent facts: figures, tax rules and legal wording must be checked against your source data and a qualified professional, because a confident-sounding AI can still be wrong.

That last point matters most in our field. AI is a brilliant drafter and a poor auditor of itself. The judgement stays with you.

Iteration is the real skill

Here is the secret the prompt collections rarely mention: the people getting extraordinary results are not writing one perfect prompt. They are having a conversation. The first answer comes back, and instead of sighing and giving up, they steer — “good, but cut it in half,” “now make it sound less corporate,” “add a number to each point,” “rewrite it for someone who’s never seen our accounts.”

Each instruction sharpens the result. Three quick rounds will beat one heroic prompt almost every time, and it takes less effort than rewriting the thing yourself. Treat the AI like a capable junior colleague who responds instantly to feedback and never takes it personally. Tell it what’s not quite right, and let it try again.

Refine once and the answer sharpens — a single follow-up turns a good draft into board-ready bullets. Branded graphic by PGH Consulting, LLC (illustrative demonstration figures, not client data).

Master a role, the context, a specific ask — and the willingness to say “close, now do this” — and you won’t need a list of prompts at all. You’ll write your own.

What to try this week

1. Pick the one prompt above that matches your job. Paste in your real (non-sensitive) details and run it.

2. Don’t stop at the first answer. Give it two follow-ups — “shorter,” “add numbers,” “change the tone” — and watch it improve.

3. Save the version that finally lands as your own template. Next time, you start from there.

4. Notice one task you do every week that’s really just “draft, then refine”. That’s your next candidate for AI.

Better prompts, better outcomes — a worked variance summary plus the four steps to try this week. Branded graphic by PGH Consulting, LLC (illustrative demonstration figures, not client data).

Always verify figures, tax rules, and legal wording against your source data and a qualified professional before acting on any AI-generated answer.

Peta-Gaye Hardy is the founder of PGH Consulting, LLC, where she helps finance and operations teams adopt AI in practical, low-risk ways. She writes the weekly AI in Finance & Business column and is based between Jamaica and the United States. Learn more at www.pghconsultinggroup.com. Follow on Instagram @pghconsultinggroup.

Disclosures: This article is informational and does not constitute investment, tax, legal, or accounting advice. AI tools can produce errors and figures should always be verified against source data. Some features described require paid subscriptions. The author has no commercial relationship with ChatGPT (OpenAI), Claude (Anthropic), Microsoft Copilot, or any product mentioned and was not compensated by them. Readers should consult a qualified professional before acting.