Computing and AI
AI for accountants in Morocco: practical uses, checks and limits
Excel formulas, invoice extraction, plain-language explanations, engagement letters: what AI does well for an accountant, what you must check and what you never hand over.
For an accountant in Morocco, AI is first a way to save time on everything around the numbers: Excel formulas, pulling data out of invoices and bank statements, explaining things to clients, letters, commentary on monthly reports. It does not replace checking, professional judgement or responsibility. The rule fits in one line: AI drafts, you check, you sign.
This guide is for chartered accountants (experts-comptables), certified accountants (comptables agréés), accounting firm staff, in-house accountants and payroll officers.
What AI changes for an accountant, and what it does not
An assistant such as ChatGPT, Claude or Gemini runs on a language model: it produces plausible text, word by word. It is very good at rewording, structuring, summarising and suggesting a formula. It is much less reliable at arithmetic, at quoting a specific text or at knowing the latest finance law, and it does not warn you when it gets something wrong.
For an accountant, that draws a clear line. Anything that is drafting, formatting or a first version can be sped up. Anything that commits you to a figure, a rate, a deadline or a tax position stays your job, backed by your own sources.
Six practical uses in a working week
1. Excel formulas and checks
This is often where the time savings are biggest. Describe your sheet (the columns and what they hold) and what you want: a SUMIFS by account and month, an XLOOKUP between the general ledger and the trial balance, conditional formatting that flags differences.
Ask for checking formulas too: a gap between total debits and total credits, duplicate invoice numbers, dates outside the period. The AI suggests the formula and explains how it works. You test it on a few rows where you already know the answer before you copy it across the whole file.
2. Pulling data from invoices and bank statements
A scanned invoice or a PDF statement can become a table: supplier, ICE, number, date, amount before tax, VAT and total for an invoice, or date, description, debit and credit for a statement. The time saved on data entry is real, but extraction is never guaranteed: a misread digit, a skipped line, a misplaced decimal point.
So check by hand, every time:
- Net amount plus VAT equals the total on each invoice, and the VAT rate applied line by line.
- Opening balance plus movements equals the closing balance on the statement.
- The number of rows extracted matches the number of lines in the document.
- The totals in your table match the totals printed on the document.
3. Explaining a tax point to a client in plain words
A client does not understand why tax was withheld at source on their rent, why their VAT return shows a credit, or how corporate tax (IS) instalments work. AI is very good at turning a technical explanation into a clear message, in French, Arabic or simple darija.
The method that works: you first write the correct points in a few lines, then ask the AI to reword them for a non-specialist. You start from your substance, not its own.
And if the AI cites an article of the CGI (the General Tax Code), check it in the current edition. Assistants regularly invent article numbers or rely on an old version: the CGI is amended every year by the finance law, and the DGI, Morocco's tax authority, publishes the up-to-date edition on tax.gov.ma. One misquoted article in a client letter, and your credibility pays for it.
4. Client letters: reminders and engagement letters
A reminder about unpaid fees, a request for missing documents before year-end, a heads-up about a deadline: give the context, the tone (firm but polite), the language and anything that must not be said, and you get a draft in seconds.
For an engagement letter, AI can suggest a structure: parties, nature and scope of the work, each side's obligations, fees, duration. But the content has to reflect the actual engagement and the standards of your professional body, which the draft knows nothing about. The safest route: start from your own approved template and ask the AI to adapt it to a new client.
5. Commentary on monthly reports
You have a reviewed table: revenue, expenses, margin, cash, compared with last month and with the budget. AI can write a few paragraphs of commentary for the business owner: what moved, which variances need explaining, which questions to ask.
Two traps. It sometimes recalculates percentages and gets them wrong. It can also "explain" a variance with a made-up cause. Give it only figures you have already checked, tell it not to do any calculations and to present causes as hypotheses to confirm, then check every number in the text against the table.
6. Payroll: explain, do not calculate
In payroll, AI can help draft a note to staff, explain how a payslip is laid out or prepare a month-end checklist. Do not let it calculate CNSS contributions, AMO health insurance or income tax (IR): rates, ceilings and brackets change, and it may use outdated values. Your payroll software and the official sources (CNSS, DGI) are the reference.
Summary table: what AI does, what you check
| Task | What AI does | What you must check |
|---|---|---|
| Excel formulas | Suggests and explains a formula | Test on rows with a known result |
| Scanned invoices | Extracts fields into a table | Net + VAT = total, rate, ICE, number of lines |
| Bank statements | Sorts the movements | Opening balance + movements = closing balance |
| Tax explanation | Rewords in plain language | Every cited article, in the current CGI |
| Reminders and letters | Writes a draft | Tone, amounts, dates, recipient |
| Engagement letter | Suggests a structure | Actual scope, professional standards |
| Monthly report | Writes the commentary | Every figure against the reviewed table |
| Payroll | Explains a payslip | No calculation used without checking |
Limits you need to know
Invented figures and articles
A "hallucination" is a wrong answer delivered with confidence. In accounting it takes three forms: a total added up wrongly, a rate or threshold from another year, an article number that does not exist or does not say what it is claimed to say. The more precise and local the question, the higher the risk. The fix: ask for the reference, then open the text yourself.
AI never has the final word on tax
A tax position commits your client and your own responsibility. AI can help you structure research or list the questions to ask yourself, not settle the matter. For a doubtful case, the references remain the current CGI, the DGI's circulars and, where needed, the tax administration itself or a colleague's opinion.
Professional secrecy and Law 09-08
Chartered accountants practise under Law 15-89, which set up the Ordre des experts-comptables, and certified accountants under Law 127-12. Professional secrecy and the confidentiality you owe clients do not stop at the door of an AI assistant: pasting a general ledger, a tax return package or a payslip into a public tool means handing that information to a third-party service, often hosted abroad.
On top of that comes Law 09-08 on personal data protection, as soon as you handle names, CIN numbers, salaries or bank details of individuals. The CNDP, Morocco's data protection authority, stated in a press release of 18 March 2025 (in French) that AI processing which uses personal data falls under this law. The good habits are set out in our article on ChatGPT at work and Law 09-08.
In practice:
- Anonymise: "Client A", "Employee 3", rounded amounts when the detail is not needed.
- Remove CIN numbers, bank details, addresses and phone numbers before pasting.
- In free versions, check the settings to see whether your conversations are used for training, and turn it off where you can.
- For regular use in the firm, prefer a business plan with written commitments on data.
These pointers are not legal advice and do not replace the rules of your professional body.
What not to hand over to AI
- Approving a tax or social security return, and filing it.
- Choosing a tax option or position for a client.
- The final calculation of tax, contributions or payroll.
- Signing, attesting or certifying anything.
- Sending any document to a client without a human review.
- Processing client data that has not been anonymised in a public tool.
Where to start this week
- Pick a repetitive, low-risk task: a checking formula, a reminder letter.
- Write a complete prompt: context, role, format, constraints.
- Test it on a case where you know the right answer.
- Note the time saved and the mistakes you caught.
- Keep the prompts that work in a shared document for the firm.
For ready-to-adapt templates, see our AI prompts for office work. Lawyers face the same questions about secrecy and sources: see AI for lawyers in Morocco.
In Tiznit, ApexTiz runs AI at work, four on-site mornings, Saturday and Sunday from 9:00 to 13:00 over two weekends in a row. The first weekend sets the rules (confidentiality, reliable sources, checking), the method and AI in the browser; the second moves on to connectors, small agents and then a final project, shared by several jobs or specific to yours, for example a VAT assistant that cites the tax code article for checking. Paid tool accounts are provided by the centre, one per pair. You leave with your project, a prompt kit and a certificate of attendance.
Frequently asked questions
Can ChatGPT keep a company's books?
No. It helps with drafting, formulas and data extraction, but bookkeeping, checking and filing remain a professional's job, done with accounting software.
Does AI know the Moroccan General Tax Code?
Partly, and often in an old version. It also invents article numbers. Always check in the current edition published by the DGI.
Can I paste a client's bank statement into ChatGPT?
Not as it is. Remove names, bank details and anything that identifies someone, keeping professional secrecy and Law 09-08 in mind. For regular use, prefer a business plan with commitments on data.
What is the best first use?
Excel formulas and draft letters: the time saved shows quickly and mistakes are easy to spot.
Do I need a paid subscription?
Not to learn: the free versions are enough. For use in a firm on sensitive documents, a business plan with written data guarantees is the safer choice.
Want advice for your case?
Put it into practice in four mornings
The AI at work course: 16 hours at the ApexTiz centre in Tiznit, over two weekends, paid tools provided, connectors, agents and a final project on your own job.