How to use AI with Pastel: a safe first step
A runnable, read-only first AI task for a Pastel practice: a first-pass read of one exported report, with a person checking every line.
You want to know how to use AI with Pastel on a real job without going near the books, and the question is which task is small enough to be safe and useful enough to be worth the hour. The answer is a first-pass read of one exported report, run through a paid-plan AI, with a person checking every line before anything counts. C-Suite Holdings runs managed AI for SA accounting firms, and the part we run is narrow: the document chase and a first pass at exceptions, read-only, on the software you already use, with your own person signing off. This guide teaches the single runnable first step for how to use AI with Pastel, the report to start with, and the pitfalls, so you can decide whether to keep doing it yourself or hand it off.
What is the safest first AI task on a Pastel practice?
The safest first AI task on a Pastel practice is a first-pass read of one exported report, where the AI summarises and flags what it sees and a person checks every line before acting in Pastel by hand. You export a single report that needs human judgement (a customer age analysis or an unallocated-items list), hand the AI the file, and ask it to group the items and surface anything that looks off. Nothing is posted, nothing is written, and every flag is reviewable before it touches the books.
This is the right place to start because the risk is low and reversible. The AI is reading and organising, not deciding, so a wrong grouping costs you a few seconds of reading rather than a misstatement in a client's ledger. That property, useful on the upside and cheap on the downside, is exactly what you want from a first step on real client work.
This article stays on the report-reading rung. The bank-feed reconciliation work and the VAT201 and EMP201 preparation inside the close are a separate, deeper workflow covered in AI for month-end close; the first step here is the safer one below it, a triage read you check by hand.
How do I get Pastel data to an AI tool safely?
You get Pastel data to an AI tool by exporting one report to a file, stripping the identifiers, then uploading the de-identified copy to a paid-plan AI account with training switched off. Pastel holds the books locally or on a firm server, so the natural and most controllable route is an export, a CSV or PDF of the single report you want read, rather than any live connection. Because the AI only ever sees an exported copy and never reaches Pastel, there is no path for it to write back; the ledger changes only when a person changes it in Pastel.
The mechanics in order:
- In Pastel, run and export the report you want read (for example a customer age analysis or an unallocated-receipts listing) as CSV or PDF.
- Open the file and replace client and counterparty names with neutral labels (Customer A, Supplier B), keeping the amounts and ages, so no personal information leaves the firm.
- Upload the de-identified file to a paid business-plan AI account (ChatGPT, Claude, or similar) with training on your data switched off, and for client work, a data processing agreement in place.
- Read the AI's summary against the source rows, accept what is correct, ignore what is not, and make any actual change in Pastel yourself.
Which Pastel report should I start with?
Start with a customer age analysis or an unallocated-items list, because both are bounded, low-stakes, and reversible: the AI helps you read and prioritise, not change a balance. A customer age analysis is a clean first target because the task is triage, ranking overdue balances and drafting a short internal note on the worst few, and a wrong call there costs a second look, not a posting. An unallocated-items list (receipts or payments not yet matched to an invoice) is the other strong starting point, because the AI can group the items and flag the obvious candidates while a person keeps the matching decision.
Both reports are self-contained, so the AI does not need the whole ledger to be useful, and both carry no authority to change anything, because the work happens on an exported copy. Leave the trial balance and the statutory returns for later: a first-pass read of a trial balance for unusual movements is a reasonable second experiment, but the age analysis and the unallocated-items list give you the cleanest, safest win on day one.
| Report | Why it is a safe first pick | What you ask the AI to do |
|---|---|---|
| Customer age analysis | Triage only; no balance changes | Rank overdue balances, draft a one-line note on the top few |
| Unallocated-items list | Bounded; matching stays with a person | Group items, flag obvious candidates, never confirm a match |
| Trial balance (later) | Read-only review, slightly broader | Flag movements that look unusual against the prior period |
What should I never paste into a consumer AI tool?
Never paste identifiable personal information into a consumer AI tool: client and individual names, ID numbers, contact details, bank account numbers, and anything that ties a financial line to a named person. Under POPIA your firm stays the responsible party for that data wherever it goes, so a free consumer account that may train on your inputs is the wrong tool for client data. The rule is to de-identify before the file leaves your control, and to use only a paid plan with training switched off and, for client work, a data processing agreement in place.
The practical version is short. Replace names with neutral labels before you upload, because the AI does not need a real name to rank an overdue balance or group an unallocated receipt, and keep ID numbers and bank account numbers out of the file entirely, since a category, an amount, and an age carry the analysis while the identifiers do not. The deeper treatment of chasing and handling client documents without breaching POPIA lives in document chasing decides your filing season, and it is worth reading before you make any of this a habit.
What are the common Pastel-plus-AI mistakes?
The common mistakes are trusting a confident-but-wrong answer, trying to wire the AI into Pastel, and pasting identifiable client data into a free account. Each one is avoidable with the same discipline: keep the AI on an exported copy, keep a person on every line, and keep the data de-identified on a paid plan.
How do I check the AI got it right?
You check the AI by reconciling its read back to Pastel: take each item it summarised or flagged, find it in the source report and in Pastel, and confirm the amount, the age, and the customer before you act. The AI's job is to organise and surface, so the check is fast, you are not re-doing the work, you are verifying that the grouping and the flags line up with what is actually in the report. Where the AI ranked an overdue balance, confirm the balance and the age against the age analysis, and where it flagged an unallocated receipt as an obvious match, open Pastel and let a person make the matching decision.
Write the rule down so it survives a busy month-end. A one-line internal note ("AI reads are drafts; [name] reconciles every flagged line to Pastel before any action or client contact") keeps the discipline from eroding when the week compresses, and the sign-off is the control that lets you use AI on real client work without putting the ledger at risk, the same boundary a managed setup preserves rather than removes.
Should a firm hand this to a managed operator?
A firm should hand this to a managed operator when the export-and-review loop turns from a useful experiment into a recurring monthly job across many clients, and the manual handling starts costing the senior time it was meant to save. One person reading one client's age analysis on a quiet afternoon is a fine do-it-yourself task. The same person exporting, de-identifying, and reconciling reports for thirty clients, every month, against filing deadlines, is the point where ad-hoc exports and copy-paste stop scaling and the sign-off step starts slipping.
The signals are concrete: you are exporting and de-identifying the same Pastel reports by hand every cycle, the reconcile-to-Pastel check gets rushed when the month tightens, and the single view of what is outstanding across clients lives in someone's memory rather than a system. That is where a managed operator earns its place, running the chase and the first-pass exceptions read-only on the Pastel you already use, on a schedule, with your own person still signing off. C-Suite is not a Sage or Pastel partner and claims no certification or endorsement; it runs read-only alongside the Pastel you already have. To see how that would run on your firm, book a free Roadmap Session.
Frequently asked questions
Can AI change anything in my Pastel data with this method? No. The whole method runs on an exported copy of a report, and the AI never connects to Pastel. Nothing changes in the books unless a person makes the change in Pastel by hand after reconciling the AI's read against the source.
Which AI account should I use for this? A paid business-tier account (ChatGPT, Claude, or similar) with training on your data switched off and, for client work, a data processing agreement in place. A free consumer account is the wrong tool for client data because it may use your inputs to train the model.
Which Pastel report is the safest one to start with? A customer age analysis or an unallocated-items list. Both are bounded and reversible, the AI only reads and prioritises, and a wrong call costs a second look rather than a posting. Leave the trial balance and the statutory returns for later.
Do I have to remove client names before uploading? Yes, treat that as the default. Replace names and any ID or bank account numbers with neutral labels before the file leaves the firm, because the AI does not need real identifiers to rank a balance or group a receipt, and POPIA makes you responsible for that data wherever it goes.
Is C-Suite a Sage or Pastel partner? No. C-Suite is not a Sage or Pastel partner and claims no certification or endorsement. It runs read-only alongside the Pastel your firm already uses.
Where to go next
- The deeper close workflow inside the books, including bank-feed reconciliation and VAT201 and EMP201 prep: AI for month-end close.
- How document chasing and handling client data shapes the whole filing season: Document chasing decides your filing season.
- New to working with AI at all: Getting comfortable with AI at work.
- The broader picture of where AI fits an SA practice: AI for accounting.
- To see how the chase and exception pass would run on your firm: book a free Roadmap Session.