The AI‑Augmented Accountant: Mastering the Art of the Prompt - Part 2 – Advanced Application, Risk & Professional Responsibility
This article will count 0.25 units (15 minutes) of unverifiable CPD. Remember to log these units under your membership profile.
In the Part 1 article of this series, we established why prompting is now a core professional skill for business accountants and introduced practical frameworks for producing reliable AI outputs. This second article focuses on implementation: how prompting can be embedded safely, efficiently, and ethically within accounting practice.
Meta Prompting: The “Prompt for a Prompt”
Meta prompting is an advanced technique where the accountant asks AI to help design the optimal prompt before performing the task itself. This is particularly useful when dealing with complex accounting, reporting, or analytical requirements.
Instead of asking the AI to solve a problem immediately, the practitioner instructs it to ask clarifying questions or generate a structured prompt. This improves output quality by defining scope, assumptions, and technical requirements upfront.
🛠️Example 1. Ratio Analysis Prompt
You want to analyse a trial balance for anomalies but don't know how to instruct the AI to look for specific ratios. You use meta prompting:
“Act as an experience financial auditor and data analysis. Create a prompt that I can feed into an AI that will instruct it to calculate liquidity ratios, check for misallocated Capital vs. Revenue expenditure, and flag high-risk accounts. The generated prompt must ensure the AI explains its working.”
Meta prompting is especially useful in independent review and analytical procedures, where scope and criteria must be clearly defined before analysis begins.
🛠️Example 2: Independent Review Analytical Procedures Prompt
“Create a structured analytical review prompt for financial statements that includes ratio analysis, period comparisons, margin trend checks, and unusual fluctuation detection above a defined percentage. Require explanations and calculation steps for each flag. Do not assume missing data.”
Creating a Prompt Library
A prompt library is a structured repository of tested prompts that can be reused across engagements. Treating prompts as operational assets improves consistency, efficiency, and risk control.
Microsoft Excel is suitable for solo practitioners, allowing prompts to be stored as structured data with categories, task names, prompt text, and variables. Filtering and fast retrieval reduce duplication and copy-paste errors.
Microsoft Loop is better suited to teams. Live prompt components can be shared across Teams, Outlook, and Word, ensuring version control, collaboration, and real-time updates.
Typical categories in a firm prompt library may include:
Independent review procedures
VAT checks and classifications
Ratio analysis
Client communication drafts
Risk checklists
SARS submission support.
🛠️Example: VAT Apportionment Information Checklist Prompt
“Create a checklist of information required to calculate a VAT apportionment ratio for a business with both taxable and exempt supplies. Include typical source documents required and common calculation errors to avoid. Do not calculate the ratio, only list the required inputs.”
Risks and Risk Management
The use of AI introduces specific professional risks, particularly in the South African regulatory environment.
Data privacy and confidentiality remain critical under POPIA. Entering unredacted client information into public AI tools may expose personal or confidential data. Accountants should redact identifying information and, where possible, use enterprise AI tools that may contractually exclude data from model training — practitioners must verify vendor terms.
Another key risk is hallucination, where AI confidently fabricates legislation, cases, or figures. AI outputs must be grounded in provided data and independently verified against authoritative sources. Outputs should be treated as preliminary work, not final authority.
🛠️Example 1: Safe Mode Output Control Prompt
“Answer only using the facts provided below. Do not assume missing data. Do not invent legislation or references. If the information is insufficient, state that clearly. Show your reasoning steps and label any uncertainty.”
AI can also be used for structured preliminary risk scans — but always subject to human verification.
🛠️Example 2: Independent Review Going Concern Indicators Prompt
“Using the financial data provided, list indicators that may raise going-concern concerns. Consider liquidity ratios, debt servicing capacity, recurring losses, and negative operating cash flow. Explain why each indicator matters. Do not reach a final conclusion and do not assume facts not provided.”
Practical VAT Applications
Prompting can support VAT work by improving consistency in classification reviews, reasonableness checks, and transaction risk screening. However, AI outputs must always be checked against the VAT Act, SARS guides, and the specific contractual facts.
🛠️Example 1: VAT Transaction Classification Risk Scan Prompt
“Review the transaction descriptions below and flag items that may require special VAT treatment, including zero-rated supplies, exempt supplies, mixed supplies, or potential change-of-use adjustments. Explain the possible VAT risk for each flagged item. Do not assume missing facts and do not give a final tax opinion.”
🛠️Example 2: VAT Reasonableness Check Before VAT201 Submission Prompt
“Using the VAT totals and turnover figures provided, calculate the implied VAT ratio and compare it to expected ranges. Flag unusual deviations and list possible explanations. Show all calculation steps and state any assumptions clearly.”
Other Relevant Issues
Rise of custom GPTs
Custom GPTs or private AI instances allow firms to upload verified reference material such as legislation, professional codes, and internal procedures. When properly curated and maintained, these tools can reduce error risk by limiting responses to approved sources. However, uploaded material must be current and validated — outdated references will produce outdated outputs.
Firms should also consider adopting a formal AI usage policy covering data handling, prompt standards, review requirements, and documentation practices.
Professional Responsibility and Ethics
The use of AI does not diminish professional responsibility. Accountability for advice, calculations, and conclusions remains fully with the accountant, regardless of whether AI tools were used.
AI outputs are decision-support tools only and do not constitute professional advice. Professional liability remains unchanged in cases of error, omission, or negligence.
Practices should implement mandatory human review protocols for all AI-assisted outputs, including:
verification of technical references
reasonableness checks
scope confirmation
confidentiality review
Where AI tools are used in engagements, both the prompt and the output should be retained as part of working paper evidence.
The application of AI requires continued professional scepticism, judgement, and adherence to ethical principles, including integrity, objectivity, competence, due care, and confidentiality, consistent with the CIBA Code of Professional Conduct and global ethical frameworks.
Conclusion and Call to Action
For business accountants, AI prompting bridges operational efficiency and strategic advisory work. Used correctly, structured prompting enhances productivity while preserving professional standards.
Practitioners should select appropriate prompting frameworks, build a reusable prompt library, implement strict data-protection practices, and maintain rigorous review processes.
AI should enhance, not replace, professional judgement. Begin building your prompting discipline now. Apply one structured prompt framework to your next accounting task and evaluate the improvement in clarity and output quality.
Join Us for the 2026 National Budget Speech Viewing & Expert Analysis
25 February 2026 | CSIR international Convention Centre, Pretoria
South Africa’s 2026 National Budget Speech will set the tone for the country’s economic direction, tax landscape, and fiscal priorities. With potential tax increases and key policy shifts on the horizon, finance professionals cannot afford to simply watch the speech, you need to interpret it.
That’s why CIBA is hosting a live Budget Speech Viewing & Expert Analysis Event designed to unpack the announcements as they happen. Join us in Menlyn for an afternoon of insight, analysis, and meaningful professional connection.
🎙️ Our expert panel includes:
Johan Heydenrych, Tax Director at Kreston SA
Ettiene Retief, Head of Tax at Omne
Prof. Phindile Nkosi, Director of the Public & Environmental Economics Research Centre (PEERC)
Together, they will translate policy into practical takeaways for your clients, your advisory role, and your strategic planning.
🍸 After the session, enjoy a cocktail networking event to engage with peers and discuss the road ahead for South Africa’s economy.
📅 Date: Tuesday, 25 February 2026
📍 Venue: CSIR international Convention Centre, Pretoria
⏰ Time: 13:00 – 16:00 (Networking 16:00 – 17:30)
💼 CPD: 3 Units (Taxation)
💰 In-person ticket: R350
💻 Prefer online? Book the virtual option for R250.
Why attend?
✔ Understand real-time tax implications from top experts
✔ Gain clarity on anticipated tax changes
✔ Learn how fiscal decisions will affect SMEs, compliance, and advisory
✔ Strengthen your strategic insights for 2026
✔ Connect with other finance professionals and leaders
➡️ Secure your seat for the in-person event here
