How to Write Good Prompts (and Truly Use AI in Finance)

Every finance professional has been confronted with the potential of Artificial Intelligence. However, most teams are still stuck in the "frustration cycle": they open ChatGPT, ask a broad question, receive a generic wall of text, and consequently conclude that the technology is not useful for the sector's critical and complex needs.

The error is not in the technology, but in the quality of your communication. A well-written prompt transforms AI from a simple text generator into a 24/7 Senior Analyst. It can generate structured reports, automate variance analyses in seconds, and deliver actionable insights. Mastering this technique is what truly separates teams that talk about automation from those that execute financial automation.

The Essentials: How to Write Prompts That Work

The foundation of an effective prompt is understanding that AI is a logical reasoning engine that requires clear instructions, context, and a well-defined output format.

1. Provide Context and Define the Role (The Reasoning of a Controller)

AI needs to know who it is being and to whom the answer is intended. This affects the level of technical depth, tone of voice, and strategic focus of the analysis.

• The Basic (No Context): "Summarize this report"
• The Advanced (With Defined Role): "You are a senior financial analyst specializing in liquidity. I need a summary of this month's cash flow in 3 short paragraphs, focused on the risk of covenant breach. Use warning language for the C-Level."

Why it works: By defining the role (senior analyst) and the focus (covenant risk), you ensure that the priority and depth of the response are aligned with the area's most critical concerns. ### 2. Be Radically Specific (Clarity and Metrics)

Clarity is the pillar of a good prompt. Ambiguous or confusing language leads to misaligned responses. It is essential to detail exactly what you want to achieve, including metrics, period, and calculation assumptions. Another example:

• Poor (Ambiguity): "I need general information about costs."
• Better (Clarity and Metrics): "List the company's monthly operational costs for the last three months (T-3), grouped into categories (rent, salaries, utilities). Calculate the percentage variation of each category versus budget and highlight in red variations above 10%."

Why it works: You avoid broad requests like "analyze costs." By detailing the objective (identify deviations above 10%), AI directs the analysis according to the focus, seeking anomalies instead of just describing data.

3. Show the Ideal Format (Actionable Structure)

AI is excellent at following patterns. If you want a usable output format — a table, a list with headers, or an email structure — you must dictate that structure.

• Poor: "Analyze the month's results"
• Better (With Structure): "Analyze the month's results and organize the response in bullet points: Revenue (% variation vs. Budget), EBITDA (actual margin), Top 3 expenses (impact %), 1 actionable insight for the next month."

Why it works: You eliminate the "wall of text." Defining the desired format saves time and increases the applicability of the response, which can be copied directly into an executive presentation.

4. Break Down into Steps (Chain-of-Thought)

For complex analyses, use the Chain-of-Thought technique or divide the task into logical steps. This forces AI to show its reasoning step-by-step before reaching the final conclusion.

Example for Expense Variance Analysis (3 Steps):

1. "List the 5 expenses that varied most (in absolute value) in the quarter compared to the previous quarter."
2. "For these 5 expenses, analyze the purchase logs and suggest the most probable cause of the variation (e.g., volume increase, accounting error, or inflation)."
3. "Based on the data and the diagnosis from Step 2, suggest 3 concrete and immediate actions to mitigate the impact of these expenses."

Why it works: The sequential approach creates an auditable and structured workflow. If AI makes an error in the cause analysis (Step 2), you can correct only that step, ensuring greater accuracy in the conclusion (Step 3).

Tools to Get Started (And the Critique on Data)

The prompt is the engine, but the tool defines the security and capacity to process your data.

Abstra – The financial automation platform. Ideal for executing complex prompts that require security, logs, traceability, and direct connection with ERPs and databases. It's the choice for those who need governance and auditing in business process automations.

ChatGPT, Copilot, Gemini – Excellent for drafts, brainstorming, summaries of long documents, and quick analyses based on text or copied/pasted data. Critique: Most public versions are not connected to real-time enterprise data and present confidentiality risks (data leakage). Use with caution for confidential data.

BI + SQL + Your Data (The Direct Connection) – The real leap in quality comes when AI consults structured and updated data at the source (such as your ERP). In these cases, the prompt serves to instruct AI to write the correct SQL query and then analyze the result of that query. It's the only way to prevent AI from fabricating numbers.

Why This Matters: The Leap to Strategic Efficiency

Mastering the technique of prompt engineering is a fundamental skill, as it frees the finance team from manual labor and directs them towards strategic thinking.

CFOs and controllers who use structured prompts can:

1. Interpret Data: Transform large volumes of raw data (Income Statements, Balance Sheets) into high-level diagnostics in seconds.
2. Project Scenarios: Run complex "what if" simulations (e.g., impact of exchange rate variation on acquisition cost) with agility and precision.
3. Find Opportunities: Proactively identify anomalies and gaps between budget and actuals.

Teams that are doing this right are already seeing clear and measurable results:

• Drastic reduction in time spent on manual analyses and report consolidation (with reported reductions over 70%).
• More accurate and traceable data, as prompt automation ensures that the same calculation premise is always repeated.
• Faster decision-making, as structured data and insights reach the C-Level early in the month.

In Summary: Structure and Governance Are Key

The difference between a useless response and a strategic financial decision lies in the discipline of your question. Always remember the four pillars:

• Provide complete Context (Role and Audience).
• Be Specific (Metric, Period, Exclusions).
• Define the desired output Format you need.
• Use Sequential Steps for complex analyses.

With financial automation tools like Abstra, you bring prompt discipline into the realm of security and traceability, ensuring your structured question results in an auditable and precise action. Don't settle for generic answers. Ask for more, and ask better.

---

Read also:

• The Role of AI Tools in Financial Automation
• How CFOs Are Using Automation to Gain Operational Efficiency