Practical ai for accounting & finance

Speaker

Nestene Botha (CA)SA

About The Speaker

Nestene completed her BCom (Hons) Cum Laude at North West University in 2012, ranking 2nd in her Honours class. She went on to complete her Master’s in Chartered Accountancy and her professional articles, registering as a Chartered Accountant (SA) in 2015. She then lectured on the audit programme at the University of Cape Town until 2017.

In 2017, she founded The Audit Pro, a virtual auditing firm that grew to 70 clients and R1 million annual turnover. Alongside running her firm, she continued delivering technical training to universities and professional bodies.

During Covid, she co-founded Explore ProTech Entrepreneurial Haven, providing entrepreneurs and professionals with training, strategy, visibility, and networking support. Nestene has worked with institutions including SAICA, UCT, North West University, Milpark, Mancosa, and Durban University of Technology. She is passionate about building practical, high-level training programmes that drive real results.

She has been recognised as one of SAICA’s Top 35 Under 35 Chartered Accountants in South Africa and one of the Top 50 Women in Accounting globally by Practice Ignition.

Upcoming Event

practical AI for accounting & finance

  • 12th & 13th November 2026
  • Voila Hotel, Bagatelle
  • 9AM - 5PM
  • Rs 28,000 (MQA Approved & HRDC Refundable up to 90%)
Register Now
Course Outline
course overview

This two-day, face-to-face course is built around one idea: participants learn AI by using it, not by watching someone else use it. From the first hour of Day 1 the laptops are open and everyone is working in a live AI tool. The theory is kept deliberately short and is introduced only where it changes what participants do next.

The course focuses on the transferable skills that work in any of the four tools accountants in Mauritius are most likely to have on their desk — Claude, ChatGPT, Microsoft Copilot and Gemini — with a small number of tool-specific techniques added at the end. Prompting, working with documents and data, controlling hallucination and verifying output are the same skills whichever platform a firm has standardised on, so nobody is disadvantaged by the tool their employer pays for. 

The centre of the programme is the workflow build. Participants bring a real process from their own work — a monthly management pack, a reconciliation, a client query, an audit file preparation routine — and spend a substantial block of Day 2 redesigning and building it in groups, with the facilitator on the floor coaching each table. Each group then presents its rebuilt workflow to the room. Participants leave with something that runs on Monday morning, not with a list of ideas.

course overview

By the end of the workshop, participants will be able to:

  • Operate at least one major generative AI platform (Claude, ChatGPT, Microsoft Copilot or Gemini) confidently for day-to-day accounting work, including uploading and interrogating documents and spreadsheets.
  • Apply structured prompt engineering techniques across all six levels, and adapt a prompt written for one platform so that it works on another.
  • Identify which of their own daily, monthly and year-end tasks are suitable for AI assistance, and which are not.
  • Use AI to automate routine accounting tasks while retaining professional judgement and review.
  • Use AI to support financial analysis, variance investigation and management reporting without compromising accuracy.
  • Use AI to draft and improve client communication, translating technical accounting content into clear language.
  • Detect and correct hallucination, arithmetic error and fabricated referencing in AI output, and describe the review controls that make AI-assisted work safe to sign.
  • Apply the confidentiality and data protection requirements of the Data Protection Act 2017 (Act No. 20 of 2017) to decisions about what may and may not be entered into an AI tool.
  • Redesign an existing work process end to end so that AI fits into it naturally, and build a working version of it.
  • Present a redesigned workflow to colleagues, and critique someone else’s.
  • Produce a personal 30-day implementation plan and contribute to a firm-level AI use policy.
Course Details
Part 1
Hands On, and the Transferable Skillset

Session 1.1 — Straight In: Working With the Tool You Have

The course opens with participants working, not listening. Everyone gets their AI tool open, loaded with a real accounting document, and produces something useful inside the first hour. This establishes the room as a working room and lets the facilitator see exactly where each participant is starting from.

  1. Setting up: signing in, choosing a model, and what the paid plans unlock.
  2. The interface in five minutes — chats, projects, files, and where the settings live.
  3. First exercise: upload a trial balance and a set of prior-year financial statements, and ask three questions of them.
  4. Second exercise: turn a page of messy meeting notes into a client-ready action list.
  5. What just happened, and what did not — a first look at where the tool helped and where it guessed.
  6. Where everyone is starting from: a quick round of who has which tool, and what they have already tried

Session 1.2 — AI Foundations for Accountants

A short, grounded explanation of what generative AI actually is, pitched at the level an accountant needs in order to use it well and explain it to a client or a partner. Kept brief on purpose.

  1. What generative AI is, and how it differs from the software accountants already use.
  2. Why it predicts rather than calculates, and what that means for every number it produces.
  3. The Core Four in Mauritius: Claude, ChatGPT, Microsoft Copilot and Gemini — strengths, weaknesses and best-use scenarios for each.
  4. Free plans, consumer paid plans and enterprise plans: what changes at each tier, particularly around data handling.
  5. Choosing the right tool for the task at hand.

Session 1.3 — Risks, Ethics, Confidentiality and Professional Responsibility

A short, grounded explanation of what generative AI actually is, pitched at the level an accountant needs in order to use it well and explain it to a client or a partner. Kept brief on purpose.

Session 1.4 — Prompt Engineering: The Six Levels (Hands-On Workshop)

The core transferable skill of the whole course, taught as a workshop rather than a lecture. Participants write, test and improve prompts at every level, and see the quality of the output change as they climb.

  1. Level 1 — Basic requests, and why most people stop here.
  2. Level 2 — Structured prompts: context, constraints and output format.
  3. Level 3 — Role-based prompting: instructing the AI to work as a specialist.
  4. Level 4 — Chain-of-thought prompting: making the reasoning visible so it can be checked.
  5. Level 5 — Few-shot prompting: teaching house style with examples.
  6. Level 6 — System prompts and custom instructions: setting up once, benefiting every time.
  7. Making a prompt portable: adapting the same prompt for Claude, ChatGPT, Copilot and Gemini.
  8. Workshop: each participant builds and tests a prompt at each level on a task from their own work, and keeps the results as the start of a personal prompt library

Session 1.5 — Applied AI, Part 1: Routine Tasks and Financial Analysis

The first two of the applied categories, worked through as practical exercises on realistic accounting material. The session closes with the workflow discovery exercise that sets up the whole of Day 2.

  1. Category 1 — Automating routine tasks: extracting and structuring data, drafting recurring deliverables, and preparing supporting schedules.
  2. Practical: build a month-end schedule from source documents, then check it.
  3. Category 2 — Enhancing financial analysis: ratio analysis, variance investigation, trend commentary and anomaly spotting.
  4. Practical: produce and interrogate a management commentary from a set of management accounts.
  5. Workflow discovery exercise: participants map their real daily, monthly and year-end workflows, score each one for volume, pain and AI suitability, and select the single workflow they will rebuild on Day 2.
  6. Groups are formed around similar workflows, ready for the Day 2 build.
  7. Day 1 wrap-up and what to think about overnight.
Part 2
Build It, Present It, Implement It

Session 2.1 — Applied AI, Part 2: Efficiency and Client Communication

The remaining applied categories, again as practical exercises, with a bias towards the tasks that take practitioners the most time.

  1. Category 3 — Increasing efficiency: research and summarisation, working paper documentation, meeting notes and follow-up tracking, and email triage.
  2. Practical: reduce a long technical document to a one-page brief a partner can act on.
  3. Category 4 — Elevating client communication: drafting letters and reports, and translating technical accounting content into plain language a client will actually read.
  4. Practical: turn a set of financial statements into a client-ready summary, then critique it as the client would.
  5. A brief look at the two categories the course does not have time to work through marketing the practice, and personal and professional development — with take home prompts for each

Session 2.2 — Where AI Gets It Wrong

The control session. Participants are shown, and then made to find for themselves, the specific ways AI output fails in accounting work — and the review routine that catches each one.

  1. Hallucination in practice: fabricated standards references, invented case law, plausible wrong numbers.
  2. Why AI arithmetic fails, and when to make it show its working or use a calculation tool instead.
  3. Context window limits, and what happens to a long document that does not fit.
  4. Practical: a deliberately flawed AI-drafted note is circulated — participants find every error.
  5. The review routine: what to check, in what order, and what to keep on file as evidence of review.
  6. Recording AI assistance on an engagement: what a reviewer should be able to see afterwards

Session 2.3 — Workflow Build Workshop (Group Work)

The largest single block of the course. Groups take the workflow they selected on Day 1, strip it back, redesign it around AI, and build a working version — prompts, templates, instructions and review points. The facilitator moves between tables coaching each group. This session is protected time and will not be cut short.

  1. Mapping the current process honestly, including the parts nobody admits to.
  2. Deciding what AI should do, what a person must do, and where the review point sits.
  3. Cleaning up the inputs — because a workflow that runs on messy input will fail whatever tool is used.
  4. Building it: writing the standing instructions, the prompts and the templates the workflow needs.
  5. Testing it on real (de-identified) material, and fixing what breaks.
  6. Preparing a short presentation of the rebuilt workflow

Session 2.4 — Implementation, Governance and Your 30-Day Plan

The closing session turns two days of practice into something that continues after everyone goes back to the office.

  1. Rolling it out beyond one person: getting colleagues and leadership on board without mandating it.
  2. What belongs in a firm AI use policy — the practical version, one page, not twenty.
  3. Cherry on top: a short rotating demonstration of the tool-specific power features worth knowing — Copilot inside Microsoft 365, Claude Projects and Claude for Excel, ChatGPT custom GPTs, and Gemini inside Google Workspace.
  4. Measuring whether it worked: what to track over the first three months.
  5. Each participant writes a personal 30-day action plan, with the first action dated.
  6. Wrap-up, questions, and how to get help afterwards
Training Methodology
  • Practical-first. The majority of both days is spent doing, not listening.    Every theory block exists to make the next exercise work.
  • On-the-floor coaching. The facilitator moves through the room and sits with participants and groups while they work.
  • Real work wherever possible. Participants bring their own processes and documents (de-identified), so what they build is immediately usable.
  • Peer learning. Participants work in groups, present to one another and review one another’s work.
  • Tool-agnostic. Everything taught works on Claude, ChatGPT, Microsoft Copilot and Gemini. Nobody is disadvantaged by the tool their firm has chosen.
  • Deliberately unhurried. The content has been reduced so the practical sessions finish properly.
  • Every participant must bring a laptop. This is a working course, not a demonstration. A tablet or a phone is not sufficient — participants will be uploading documents, working in spreadsheets and building prompt libraries throughout both days.
  • A paid AI subscription is strongly recommended.
  1. An enterprise or business plan (Claude for Work / Enterprise, ChatGPT Business or Enterprise, Microsoft 365 Copilot, Google Workspace with Gemini). This is the best option — it is what a firm should be using for client work, and the data handling terms are materially different from the consumer tiers.
  2. A consumer paid plan (Claude Pro, ChatGPT Plus, Copilot Pro, Google AI Pro). Participants can get by comfortably on this, and it is inexpensive for a single month if the firm does not yet have an enterprise arrangement.
  3. A free plan. Workable but limiting — participants on a free plan will hit usage caps during the practical sessions and will not be able to complete some exercises.

 

Any of the Core Four is fine: Claude, ChatGPT, Microsoft Copilot or Gemini. Setup guidance will be circulated in advance so nobody spends course time creating accounts.

Participants should also bring:

  • A real work process they would like to improve, and the documents that go with it (de-identified, or we will de-identify them together in Session 1.3).
  • No prior AI experience is required. All levels are welcome.
Register Now