AI in accounting
For in-house accounting teams and accountancy practices. The repetitive work with documents and reconciliations that AI can shorten, and a clear line: AI does not make tax decisions and does not replace the accountant’s review.
What we practise
Reading documents
Data from invoices, bank statements, contracts and scans into a table. How to phrase the request so you get the columns you need, and how to check the totals.
Reconciling two lists
Two lists that should match but do not: a bank statement against the ledger, a register against a client’s file. AI flags the differences, the participant decides what to do with them.
E-invoicing and tax reporting files in a spreadsheet
Exports from your national e-invoicing system and statutory tax reporting files (in Poland: KSeF and JPK, or your local equivalents) opened in Excel or Sheets: finding discrepancies, totals by counterparty, completeness checks.
Formulas and macros
The formula nobody remembers and the macro nobody wants to write. AI writes it, the participant reads it, tests it on a small sample and only then uses it.
Letters and replies
A draft letter to the tax office, a reply to a client’s question, a reminder about documents. In your voice, with room for what a human must check.
Summaries of regulatory changes
A long circular or ruling into three points of “what this changes for our clients”. With a duty to verify at the source that we practise rather than merely declare.
Limits
What AI may prepare and what the accountant must decide and sign. A list of cases from your practice where AI stays a tool, not an adviser.
Confidentiality
Client data and professional secrecy: how to work on specifics without pasting them into a tool. Anonymisation, a business plan, what to put in the engagement letter.
Example exercises
A batch of invoices into a table
Ten PDF invoices from different suppliers. Task: number, date, supplier, net, VAT, gross in one table, with a totals check. Then: where AI got it wrong and how to catch it.
Statement versus ledger
Two payment lists where some amounts only partly match. AI prepares a list of discrepancies, the participant judges each one.
A formula on request
Everyone brings a spreadsheet where they calculate something by hand. They describe the task in plain language, AI writes the formula, the participant checks it on three rows.
A letter in half an hour instead of two
A reply to a tax office request or a client’s question: a draft from AI, the participant’s corrections, a list of things to check before sending.
What changed
A notice about a regulatory change summarised for clients, then compared with the original point by point.
Who it is for
- An in-house accounting team that retypes and compares data by hand every month.
- An accountancy practice that serves many clients and worries about the confidentiality of their data.
- A head of accounting who wants the team to use AI, but under control and with rules.
- People who tried ChatGPT at work and gave up after the first mistake.
What you take away
- A set of tested prompts for reading documents, reconciling lists and writing formulas, adapted to your spreadsheets.
- A checklist: what to verify before AI output goes into the books or to a client.
- A rule for working with client data, ready to go into the practice’s procedures.
- A list of tasks from your department rated “you can do it yourselves with AI”, “needs a developer” (e.g. automatic data pulls and a recurring report) or “stays with a human”.
- A certificate of attendance with the programme.
- 30 days of questions by email and a session after a month.
Honest about scope. I am a developer, not an accountant. I teach how to work with the tool on your documents and how to verify the results. The tax substance stays with you. Before the training I ask to speak with someone from the team so the exercises match your practice.
Questions about this module
Can AI post an entry or advise on a VAT rate?
In the training we teach that it cannot. AI prepares data, compares, drafts. The tax decision is made and signed by the accountant. That is not caution for show; it is the line that protects you and your clients.
Do we paste client data into ChatGPT?
Not into a personal account. In the workshop we work on anonymised copies of your documents. I show how to anonymise in half a minute and how a business plan, where data is not used for training, differs.
Do we need Excel, or will Google Sheets do?
Either works. We practise on whatever you use day to day.
Is this module for accountancy practices?
Yes, first and foremost. Lots of repetitive work and high confidentiality requirements are exactly where AI helps most and where mistakes are easiest. For practices I extend the part on confidentiality and client agreements.
Can data pulls and reports be automated after the training?
Often, yes. That is usually a “needs a developer” item on the workshop list. I can do it as implementation after training.
Your staff already use AI. Let’s teach them to do it well and safely.
Tell me the department and the headcount. I reply within one business day with a proposed programme.
Ask about training for your team
Tell me the department and the headcount. I reply within one business day with questions or straight away with a proposed programme. Quoted after a call, depending on format and headcount.