1.1 — Defining AI

Artificial Intelligence refers to computer systems designed to perform tasks that normally require human intelligence — recognising patterns, making decisions, understanding language, and learning from experience.

1.2 — Key Terminology

Essential terms for every accounting professional:

• Machine Learning (ML) — AI that learns from data rather than following explicit rules
• Deep Learning — A subset of ML using neural networks with many layers
• Natural Language Processing (NLP) — AI that understands and generates human language
• Large Language Model (LLM) — AI trained on massive text datasets (e.g. ChatGPT, Claude)
• Algorithm — A set of rules or steps a computer follows to solve a problem
• Training Data — The information used to teach an AI model

1.3 — Types of AI

• Narrow AI (ANI) — Designed for a specific task (e.g. fraud detection, invoice matching). All AI that exists today.
• General AI (AGI) — Hypothetical AI that can perform any intellectual task a human can. Does not yet exist.
• Generative AI — AI that creates new content (text, images, code) based on patterns in training data.

1.4 — A Brief History

• 1956 — Term “Artificial Intelligence” coined at Dartmouth Conference
• 1997 — IBM’s Deep Blue defeats chess world champion
• 2011 — IBM Watson wins Jeopardy!, demonstrating NLP
• 2022 — ChatGPT launches, bringing generative AI mainstream
• 2023–2026 — Rapid adoption across accounting, audit, and finance

2.1 — Machine Learning Basics

Traditional software follows hard-coded rules. Machine Learning flips this — you give the system thousands of examples and it figures out the rules itself.

2.2 — Training, Models & Inference

• Training — Feeding data to an algorithm so it learns patterns
• Model — The result of training — a mathematical representation of learned patterns
• Inference — Applying a trained model to new, unseen data

2.3 — How Generative AI Works

Tools like ChatGPT use transformer architecture — they predict the most likely next word in a sequence. They don’t “understand” content like humans; they recognise statistical patterns.

• Prompt — The instruction you give the AI
• Token — A chunk of text (~¾ of a word) the AI processes
• Hallucination — When AI generates plausible but factually incorrect content
• Context Window — The amount of text the AI can “remember” in one conversation

2.4 — Prompt Engineering Basics

Good prompts for accounting work should:

• Be specific (“Summarise this trial balance variance” vs “Look at this data”)
• Provide context (“I’m a tax practitioner reviewing a SARS assessment…”)
• Define the format (“Give me a bullet-point list of…”)
• Set constraints (“Only consider South African tax legislation”)

3.1 — AI in Bookkeeping & Data Entry

• OCR + AI extraction — Scanning invoices and auto-populating accounting systems
• Smart categorisation — Auto-classifying transactions based on learned patterns
• Bank reconciliation — AI matching bank feeds to GL entries

3.2 — AI in Audit & Assurance

• Full-population testing — Analyse 100% of transactions instead of sampling
• Anomaly detection — Flagging unusual journal entries or after-hours postings
• Risk assessment — Predictive models to identify high-risk areas
• Document review — AI reading contracts to extract key terms

3.3 — AI in Tax

• Tax research — AI searching legislation, case law, and SARS rulings
• Compliance automation — Auto-populating tax returns
• Transfer pricing — Analysing comparable transactions across jurisdictions
• Tax planning — Modelling scenarios and their tax implications

3.4 — AI in Advisory & FP&A

• Cash flow forecasting — Predicting future cash positions
• Scenario modelling — AI-powered “what if” analysis
• Report generation — Drafting management reports from financial data

3.5 — Generative AI for Daily Work

• Drafting client emails, engagement letters, and proposals
• Summarising lengthy regulations or technical guidance
• Creating Excel formulas and explaining complex spreadsheets
• Preparing meeting notes and action items

4.1 — The Confidentiality Risk

• Data leakage — Information entered into AI tools may be stored or used for training
• Cloud processing — Most AI tools process data on external servers
• Prompt injection — Malicious inputs designed to extract sensitive data

4.2 — POPIA & AI

• Personal information entered into AI tools constitutes processing under POPIA
• The AI provider becomes an operator (data processor)
• Cross-border data transfers require adequate protection levels
• You must have a lawful basis for processing client data through AI

4.3 — Practical Guidelines

• Never paste real client data (ID numbers, bank details, tax numbers) into public AI tools
• Use enterprise/business versions with data protection agreements
• Anonymise or use dummy data when testing
• Check your firm’s AI usage policy before using any new tool
• Keep an internal register of AI tools used
• Get informed consent from clients where AI is used on their data

5.1 — AI Bias in Accounting Contexts

• Credit scoring — Models trained on biased data may perpetuate discrimination
• Hiring algorithms — May disadvantage certain demographic groups
• Fraud detection — May disproportionately flag certain regions
• Risk assessment — Historical patterns may not reflect current realities

5.2 — Professional Scepticism & AI

• Never accept AI output at face value — always verify
• Understand the limitations — AI can be confidently wrong
• Document your review — record how AI was used and validated
• You remain accountable — AI is a tool, not a replacement for judgement

5.3 — Ethical Frameworks for AI Use

• Transparency — Disclose when AI is used in professional work
• Fairness — Evaluate whether AI recommendations are equitable
• Accountability — Maintain clear human oversight
• Competence — Only use AI tools you understand sufficiently
• Due care — Same standard of care as manual work

5.4 — Regulatory Landscape

• EU AI Act — First comprehensive AI regulation, categorising by risk level
• South Africa — AI Policy Framework and ongoing POPIA enforcement
• IFAC & IRBA — Guidance on AI use in audit and accounting
• IESBA — Developing AI-specific ethical guidance for the profession

6.1 — Start With Low-Risk Experiments

• Draft routine client emails with AI (review before sending)
• Ask AI to explain a complex IFRS standard in plain language
• Generate Excel formulas for common calculations
• Summarise lengthy SARS binding rulings
• Create meeting agendas and minutes from notes

6.2 — Build Skills Progressively

• Week 1–2: Try one AI tool for one simple task per day
• Month 1: Identify 3 repetitive tasks AI could assist with
• Month 2–3: Pilot AI on a specific engagement or project
• Ongoing: Stay informed via IRBA, IFAC, and CIBA updates

6.3 — Pre-Use Checklist

• ☐ Does my firm have an AI usage policy?
• ☐ Am I using an enterprise version with data protections?
• ☐ Have I anonymised client-identifiable data?
• ☐ Will I validate all AI output before use?
• ☐ Am I documenting AI use in working papers?
• ☐ Do I understand the tool enough to evaluate output?
• ☐ Have I considered whether client consent is needed?

6.4 — The Future-Ready Accountant

AI will not replace accountants — but accountants who use AI will have a significant advantage. The profession is shifting from data processing to interpretation, advisory, and strategic thinking.