Preparing year-end accounts with AI: tidying up your records
With year-end approaching, AI can do the groundwork: sort bank entries, spot duplicate bookings and flag missing receipts, so you can hand over a clean list.

Preparing year-end accounts with AI often starts on 28 December: your bank export contains 900 lines, half the receipts are still in your email or in a drawer, and your accountant would like to have everything before 5 January.
AI does not enter the figures for you and it does not file the return.
What it can do is help you sort faster, find missing items and list any doubtful cases neatly.
After 31 October, year-end accounts come into view for many sole traders and SMEs, because for entrepreneurs who file VAT returns quarterly, the VAT for the fourth quarter of 2026 must be with the Dutch Tax Office no later than 31 January 2027, says the KVK.
How can you prepare your year-end accounts with AI?
You can prepare your year-end accounts with AI by letting AI do the groundwork. Think sorting, grouping and flagging questions, while you and your accountant keep control of the judgement.
AI is especially useful when you have lots of similar lines. A bank export often contains recurring costs, payments to the same suppliers and descriptions you can spot quickly. AI can turn that into a proposal.
That proposal is not accounting. It is a prepared list. You use that list to see faster what is right, what is missing and what your accountant should look at.
AI can, for example, help with:
- sorting bank transactions into categories;
- spotting duplicate or near-duplicate payments;
- flagging missing information;
- grouping recurring costs;
- making a summary of doubtful cases.
AI cannot decide what is tax-correct. It does not make VAT decisions, choose deductible items or submit a return. That remains work for you, your accountant or the Tax Office.
A workable order looks like this:
- Export: download your bank transactions from your bank or accounting package.
- Clean up: remove personal data and unnecessary columns.
- Let it sort: give AI your categories and let it propose a category for each line.
- Let it check: ask AI to flag duplicate payments, missing receipts and unusual amounts.
That turns a big pile of separate lines into a workable list. Your accountant then has less searching to do and can focus on the real judgement calls.
What do you need before you start?
You need an export of your bank transactions, for example as csv or Excel. You also need your own list of categories or chart of accounts, and an AI tool that can read files.
Start with an export from your bank or accounting package. Choose only the period you want to clean up, for example 2026. Always keep the original file next to the edited version.
You will also need categories. Preferably use the categories you already use in your bookkeeping. If you have an accountant, ask which layout they want to see.
Examples of categories are office, software, travel costs, turnover, private and unknown. Use your own list first and foremost. Without a category list, AI will invent something itself, and that makes the result less useful.
You can work with an AI tool that can read files, such as ChatGPT, Claude or Copilot in Excel. Check with the provider which settings apply to storage and training of data. Where possible, choose settings where your data is not used for training.
Checklist: before you upload
- names and IBANs of individuals have been removed or replaced;
- your category list is ready;
- you are working in batches, not all at once;
- you have agreed the format with your accountant;
- the original file has been saved.
Also keep an eye on your deadline. For entrepreneurs who file quarterly, the VAT return for the fourth quarter of 2026 must be with the tax authorities by 31 January 2027. If you file monthly for October 2026, the KVK gives 30 November 2026 as the final date. For the annual VAT return for 2026, the KVK gives 31 March 2027.
If you work in Belgium, VAT deadlines and rules may be different. In that case, ask your own accountant or the Federal Public Service Finance which date and approach apply to you.
Can you put bank details and invoices into ChatGPT or Claude?
That is only sensible if you remove personal data first or have clear agreements with the provider. If in doubt, it is better not to upload any data.
Bank transactions may look purely business-related. But they can still contain personal data. Think of names of clients, IBANs, descriptions with private matters or payments to private individuals.
The Dutch Data Protection Authority warns that sharing personal data with chatbots can lead to data breaches. According to reporting by Binnenlands Bestuur about the AP, most chatbot providers store the data entered and the AP asks organisations to make clear agreements about chatbots at work.
So remove sensitive data first. Replace client names with Client 1, Client 2 and Client 3. Leave out IBANs if you do not need them for sorting. Also remove descriptions that say something about someone’s private life.
Also watch out for business customers who have a sole proprietorship. In that case, a business name can still refer to a person. Then GDPR rules still matter.
You can often work with less data. For sorting, date, amount and a general description are usually enough. A line such as software subscription often says more than a full bank description with name and account number.
If you want to go deeper, also read how to use AI safely when dealing with bank data and privacy.
How do you let AI sort your bank transactions?
Give the AI your list of categories and ask for a category, a confidence level and a short reason for each line. Make it clear that it should flag uncertainty rather than guess.
Do not start with all 900 lines at once. Send a batch of, say, 100 lines instead. AI has a limited amount of text it can handle properly at once, also called the context window.
Ask for the result as a table. Then you can easily copy it into Excel. Choose columns you can also give to your accountant later.
A useful table might have these columns:
| Column | Why is it useful? |
|---|---|
| Date | You can search it again in your bank or accounting package. |
| Amount | You can see straight away which lines are large or unusual. |
| Description | You keep the source line visible. |
| Category | You can see AI’s suggestion. |
| Confidence | You can see what you need to check yourself. |
| Short reason | You understand why AI suggests this choice. |
How do you check the result?
Then check a sample yourself. Review at least ten random lines and all lines with low confidence. Also look at large amounts, unknown suppliers and lines where private and business use may be mixed.
So use AI as a sorting machine, not as the final boss. The best result is a list where many lines are placed logically and the doubtful cases remain clearly visible.
Practical example: Sanne sorts 640 bank transactions
Here is how it can work in practice: Sanne, a freelance photographer in Utrecht, has a csv file with 640 bank transactions from 2026. She uses AI only for pre-sorting and does not decide anything about VAT or deductibility.
Step one is cleaning. Sanne opens the csv in Excel and removes the column with counterparty names from private individuals. Where she still wants to keep the information, she replaces names with Client 1, Client 2 and Client 3.
She then uploads the cleaned file to an AI tool that can read files. She does not start with all 640 lines. First she asks it to process the first 100 lines.
This is her prompt:
The expected result is a table in which many lines have a clear category. Sanne also gets a short list of Unknown items and a list of possible duplicate payments.
The exact quality varies by model and by file. That is why checking is still needed. Sanne reviews samples and checks all lines with low confidence.
That way she keeps control of the process. AI speeds up the sorting, but Sanne and her accountant remain responsible for the choices.
How do you find missing receipts and invoices with AI?
Ask AI to flag expenses without a matching invoice. Also have it suggest a search query for each supplier that you can use in Outlook or Gmail.
You do not necessarily need to give AI access to your mailbox. A list of bank transactions is enough to create search terms. You then search your email yourself.
For example, ask AI to flag software, subscriptions, larger expenses and suppliers that come back often. Ask for a separate list of possible missing receipts.
A useful table could look like this:
| Date | Amount | Supplier or description | Why look for an invoice? | Search term for email |
|---|---|---|---|---|
| Date from bank line | Amount from bank line | General description | Expense with no known receipt | Description plus amount |
AI can also group recurring subscriptions. That helps you see faster whether there is an invoice for every month. If a month is missing, you know where to look.
If you work a lot with loose receipts, it can help to make a separate list with only missing items. Also read how you can use processing invoices with AI for missing receipts and invoices.
Which mistakes can AI spot in your accounts, and which not?
AI mainly spots striking patterns, such as duplicate payments, unusual amounts and missing months. AI does not know your tax situation.
When tidying up accounts, AI is useful as a second pair of eyes. It can compare lines and flag anomalies. That can be especially helpful with lots of recurring payments.
AI can, among other things, look for:
- duplicate bookings or nearly identical amounts;
- subscriptions that suddenly increase;
- private and business mixed together;
- unusual amounts;
- months in which a recurring payment is missing.
But there are also clear limits. AI does not automatically know whether VAT is deductible. It also does not know whether something is an investment, or how private use should be treated in your situation.
These questions are for you, your accountant or the Tax Office:
- can I reclaim the VAT?
- is this a business expense?
- should this be treated as an investment?
- how do I deal with private use?
- which data belongs in the return?
Also watch out for hallucinations. That means AI can write something with confidence even when it is wrong. Read more about AI hallucinations and mistakes AI misses.
Preparing your year-end accounts with AI: what do you hand to your accountant?
Give them one Excel file with three tabs: sorted transactions, doubtful list and missing receipts. Agree the format with your accountant in advance.
A tidy handover can save work. Your accountant then does not have to go through separate files, emails and unclear lists. They can immediately see what looks certain and what needs attention.
A good layout is:
- Tab 1: sorted transactions, with date, amount, description, category, confidence and reason.
- Tab 2: doubtful list, with all lines with low confidence or category Unknown.
- Tab 3: missing receipts, with amounts, descriptions and suggested search terms.
Say that AI was used for pre-sorting. Then your accountant knows the classification is a proposal. That makes checking easier.
Always keep the original file. Also save the edited version. That way you can later see what came from the bank and what you changed.
Use this comparison as your boundary:
| Task | Can AI do this? | Who decides? |
|---|---|---|
| Sorting bank transactions | Yes, as a suggestion | You |
| Finding duplicate bookings | Yes | You |
| Flagging missing receipts | Yes | You |
| Deciding VAT deduction | No | Accountant |
| Submitting the return | No | You or accountant |
Common mistakes when tidying up with AI
The biggest mistakes happen when you trust AI too much or give it too much data. Keep it small, safe and checkable.
Pay particular attention to these five mistakes:
- Uploading the entire bank export with names and IBANs. The solution is to anonymise first or leave out columns.
- Asking AI to decide the VAT return or deductible items. The solution is to let AI only pre-sort and put tax questions to your accountant or the Tax Office.
- Sending all 900 lines at once. The solution is to work in batches and then do a sample check.
- Taking over the AI result without checking it. The solution is to check at least ten random lines and review all lines with low confidence.
- Not providing a category list. The solution is to send your own categories or those of your accountant.
These mistakes may seem small, but they make the difference between a useful preparation and a messy list. AI works best when you clearly say what it may and may not do.
Anonymising Sources
Frequently asked questions
Can AI do my bookkeeping?
No. AI can pre-sort and flag issues, but posting entries, VAT and the return remain with you or your accountant.
Is it safe to put bank details into ChatGPT?
Only after anonymising and with suitable settings. The Dutch Data Protection Authority warns that sharing personal data with chatbots can lead to data breaches.
Which AI tool is best for bank transactions?
Any tool that can read csv or Excel can help. Choose a tool that your employer or accountant accepts and always check the result.
When does my VAT return for the fourth quarter have to be in?
For quarterly filing, the VAT return for the fourth quarter of 2026 must be in by 31 January 2027, according to the KVK.
Do I have to tell my accountant that I used AI?
Yes, that is sensible. Then they know which lines were pre-sorted and which lines need extra checking.
The next step is simple: make a list of your recurring costs and let AI pre-sort them each quarter. Then next year’s year-end accounts will be smaller and you can also start earlier with processing invoices with AI.
Was this article helpful?
How this article was made
- Checked on
- 2 sources used and listed
- Written with AI and then checked by a second AI for facts, sources and links. Articles that fail that check are not published. This English version was translated from the Dutch original.




