AI Master Class: Provisional Tax - Automating Workflows
Duration: 2 hours
Price: R195.00
AI Master Class: Provisional Tax - Automating Workflows
Presenters : Nestene Botha
Overview
Every provisional tax period, tax practices lose billable hours to the same exercise: reading each client’s prior-year return, working out which supporting documents the current year’s estimate requires, and drafting the request to the client. It is repetitive, unglamorous work, and done manually it can take an afternoon per client.
Generative AI is already on practitioners’ desks: almost nine in ten of those surveyed in The Tax Faculty’s recent working paper on AI adoption use it. Yet only nine in a hundred have automated a single workflow from beginning to end, and cost ranks only seventh among the obstacles to wider use. The gap is neither adoption nor budget: very few practices have yet built an AI-assisted workflow that covers an entire process.
Rather than discuss AI adoption in theory, this video closes that gap on one workflow, built step by step: provisional tax information requests. Starting from a client’s prior-year return, an AI skill identifies the supporting documents that this year’s estimate is likely to require, tests the possible scenarios with the practitioner until the client’s position is clear, and finishes with an email to the client in the practitioner’s drafts folder, ready to be reviewed and sent. Each run leaves a file note, so the review takes minutes and the audit trail is complete. Practitioners receive the starter skill used in the demonstration, a single file to install, run, customise and keep.
Video Content
- Part 1: Introduction to AI-assisted workflows: what the term means, why it is currently the recommended route to AI adoption in an accounting practice, and what professional bodies around the world have said about it.
- Part 2: What a skill is: one file, with no separate prompting method to learn, and why it now runs the same way in Claude and in ChatGPT.
- Part 3: The workflow, built step by step: provisional tax information requests, from last year’s return to this year’s document list and client email, with the practitioner’s knowledge of the client as the one input the machine cannot supply, and the skill ensuring that every human-in-the-loop review takes place.
- Part 4: The audit trail: what makes the workflow defensible rather than merely fast, with a record of what the skill read, what the practitioner corrected, what was sent and when.
- Part 5: From a skill to a workflow: the same skill in VS Code, run across the client list and writing into each client’s own folder.
- Part 6: Picking it back up: weeks later, with the documents in hand, straight through to the calculation, and a review that takes minutes because the trail was never lost.
- Q&A and open discussion
Competencies Developed
- Understand what an AI-assisted workflow is, why it is currently the recommended route to AI adoption in an accounting practice, and what professional bodies internationally have said about its use.
- Explain what an AI skill is and how a single skill file can be installed, run and customised in both Claude and ChatGPT without learning a separate prompting method.
- Identify the supporting documents a client’s provisional tax estimate is likely to require, working from the prior-year return and testing possible scenarios until the client’s position is clear.
- Prepare clear, complete information requests to clients through an AI-assisted process that leaves a reviewed draft email ready to send.
- Keep a defensible audit trail of AI-assisted work, recording what the skill read, what the practitioner corrected, and what was sent and when.
- Exercise professional judgment over AI outputs by applying knowledge of the client and making sure every human-in-the-loop review is completed before anything reaches the client.
- Apply a single skill across a client portfolio so that each engagement’s records stay in the client’s own folder and can be picked up again at the calculation stage.