How to use AI with GreatSoft, safely
A safe first AI task for GreatSoft firms: export a deadline list, draft chase messages, keep sign-off human. No write access, no shortcuts.
It is the second week of provisional season, your IRP6 list runs to a few hundred clients, and someone on your team is copying due dates out of GreatSoft into a spreadsheet to work out who still owes you figures. That manual export is also the safest place to start with AI, and this guide walks the first runnable step end to end. C-Suite Holdings runs managed AI for South African accounting firms, and the part covered here shows how to use AI with GreatSoft for one narrow job: turning an exported list into drafted chase messages, read-only, with your own person signing off before anything is sent.
If you want the wider picture of what the tool can and cannot take on, read what AI can and cannot do with GreatSoft alongside this how-to.
Is there a safe first AI task to try with GreatSoft?
The safest first AI task is taking an exported deadline or document-status list and having AI draft the chase messages, while a human reads and sends every one. You are not connecting anything to GreatSoft, you are not letting AI decide who gets contacted, and nothing touches a client record. You export a list you already maintain, paste it into your own ChatGPT or Claude account, and ask for clear, polite follow-up wording per client, then you check each draft against what you know before it leaves your outbox.
This works as a first step because the input is low-risk and the output is fully reviewable. A drafted email is a suggestion, not an action, so the worst outcome of a bad draft is that you rewrite a sentence, which is exactly the kind of small, bounded experiment that earns trust before you ask AI to do anything heavier.
Practice management systemHow do you start without giving AI write access to GreatSoft?
You start by keeping AI entirely outside GreatSoft and feeding it an export instead, so there is no path for it to change a record. AI never logs in, never holds a credential, and never connects to the system; you run the report inside GreatSoft yourself, save it as a CSV or paste the relevant columns, and that copy is what AI reads. The flow is one direction only: data comes out of GreatSoft, into a working copy, into the AI prompt, and the result comes back to you as text you approve.
That separation is the whole safety model. The moment AI has write access or a login, a wrong inference can edit a live client record, so you remove that risk by design rather than by trusting the model to behave. Treat the export as the boundary: GreatSoft stays the system of record that only your team writes to, and AI only ever sees a snapshot you chose to hand it.
Can you choose which GreatSoft data AI sees first?
The safest data to run AI against first is your operational list: client name or code, return type, due date, and current document-status, with no financial figures or identity numbers attached. This is the layer that drives the chase, and it is far less sensitive than ledgers, tax computations, or supporting documents. You choose what AI sees by controlling the export: start there to use AI with GreatSoft on low-risk fields, prove the workflow, and only widen the data later once you know how the firm handles AI outputs.
Before anything goes into a prompt, strip what the task does not need. AI does not need a tax number to write "your February records are still outstanding", so remove identity numbers, banking details, and any free-text notes that mention a client's affairs, and keep only the fields the reminder actually depends on. The table below shows a sensible starting line.
| Data type | Safe in a first AI draft task | Why |
|---|---|---|
| Client name or internal code | Yes | Needed to address the message; use a code if you prefer |
| Return type (VAT201, EMP201, IRP6, ITR12) | Yes | Tells AI which deadline and document set to reference |
| Due date / period | Yes | Drives the urgency and timing of the reminder |
| Document-status (received / outstanding) | Yes | The core of what the chase is about |
| Tax / ID numbers | No | Never required to draft a reminder; remove before prompting |
| Financial figures, ledger detail | No | Not needed for a chase; high sensitivity under POPIA |
For the full treatment of running the chase without tripping POPIA, see chasing month-end documents without breaking POPIA, which owns that depth.
Do real pitfalls show up when bolting AI onto GreatSoft?
The real pitfalls are giving AI access it does not need, trusting drafts you did not read, and letting AI near filing itself. Each one is avoidable, and each one matters more during a busy season when the temptation to skip a check is highest. Name them up front so they do not surprise you in week three.
How do you keep a human signing off every AI output?
You keep a human in control by making approval a required step, not an optional one, so nothing AI produces reaches a client or a return without a named person checking it. Write the rule down: AI drafts, a specific role reviews, and only that person sends or files. When the rule lives in your process rather than in someone's head, it survives the pressure of a deadline week instead of quietly lapsing.
Practically, that means three fixed points. AI output lands in a draft state (an unsent email, a flagged list), a named reviewer is responsible for that client or that batch, and the act of sending or filing is always a human action taken after the check. This is the same discipline a managed operator would enforce, and it is worth building even for a small in-house experiment, because the habit is what keeps AI a help rather than a liability.
Should you stop doing this yourself and have it run for you?
You should consider a managed operator when the chase has outgrown a once-a-season manual export, when the same follow-up work eats your team's filing weeks every month, and when you want the read-only and sign-off discipline enforced rather than remembered. A solo experiment in your own AI account is the right way to learn what AI can and cannot do here, and for a small list it may be all you need. The strain shows up at scale, when one person exporting and pasting cannot keep pace with the volume, and the careful checks start slipping because there is no time.
That is the point where managed AI for South African accounting firms earns its place. C-Suite runs the chase and a first pass at exceptions read-only on the GreatSoft you already use, with your own named person signing off before anything reaches a client or the ledger, so the volume is handled without the discipline eroding. C-Suite is not a GreatSoft partner and claims no certification or endorsement; it runs alongside your existing setup. If your firm has roughly 3 to 50 staff and the month-end chase is swallowing your best people, book a free Roadmap Session and we will map how it would run on your firm before you commit to anything. South African research from Sage and Plum has put small-business admin at around 120 working days a year, and the chase is a large slice of that for a practice.
Frequently asked questions
Does AI log into GreatSoft to do this? No. AI never logs into GreatSoft, never holds a credential, and never connects to it. You export the data yourself and AI only reads the copy you hand it, so the system of record stays writable by your team alone.
Can AI submit my clients' returns on SARS eFiling? No. SARS eFiling is a government portal, and AI stops at the login. AI can prepare documents, track deadlines, and draft reminders, but a named person on your team logs in and files every return; AI never submits, integrates with SARS, or stores eFiling credentials.
Is C-Suite a GreatSoft partner? No. C-Suite is not a GreatSoft partner and claims no certification or endorsement. It runs read-only alongside the GreatSoft you already use, on the software your firm already pays for.
What is the smallest safe thing I can try this week? Export a short, cleaned deadline list and have AI draft the chase messages while you read and send each one. It needs no integration, no new tool, and no client financials, and it teaches you how AI handles your data before you scale anything.
Where to go next
- For what the tool can and cannot take on with this system: AI and GreatSoft, what it can and cannot do.
- For the POPIA-safe depth behind the chase: chasing month-end documents without breaking POPIA.
- For why the chase decides your whole season: document chasing decides your filing season.
- To see how this would run on your firm: book a free Roadmap Session.