As the annual tax return deadline approaches, desks at accounting firms and tax consultancies fill up with tax invoices, withholding slips, bank statements, and WhatsApp messages from clients asking the same questions again and again. Many staff have already tried ChatGPT or Claude to speed this work up. Some found it helpful. Others ended up with more work, because every result had to be checked from scratch.
AI for accountants and tax consultants works best when it is used on the right tasks, with client data protected and a disciplined review habit in place. This article covers where AI helps, where it needs close supervision, and how a small or mid-size firm can start using it safely.
Work that AI handles well
Extracting data from documents
This is where the benefit shows up fastest. AI combined with OCR can read tax invoices, withholding slips, supplier invoices, and bank statements, whether they arrive as PDFs or phone photos, and turn them into a table: date, document number, counterparty, tax base, VAT, and total. Staff no longer retype hundreds of rows every month. Their job shifts to checking the output, especially for blurry scans and handwritten documents.
Support for reconciliation
AI can help match bank transactions against sales records or the general ledger and flag entries that have no counterpart. A typical case is an incoming transfer from a distributor that is slightly short because of a bank fee, or one payment that settles three invoices at once. Keep one thing in mind. Language models are not reliable at arithmetic. Totals and differences should still be calculated by a spreadsheet or your accounting system, while AI reads the transaction descriptions and suggests likely matches.
Drafting replies to client questions
Questions like "which documents do I need to send for this year's return" or "why do I owe more tax" come in constantly. AI can prepare draft replies based on your firm's own templates and procedures, so the wording stays consistent across staff. Someone still reads and adjusts each draft before it goes out. Tax rules change often, and AI can easily answer from an outdated rule while sounding completely sure of itself.
Protecting client data before it reaches AI
Accounting files are full of personal data: NIK (national ID numbers), NPWP (tax ID numbers), bank account numbers, home addresses, and even the salary details of a client's employees. Most of the tasks above do not need any of these numbers. You can summarise a profit and loss statement or match bank entries just as well with the NIK and NPWP masked.
Make masking NIK and NPWP a standard step before AI reads a document. Use AI accounts managed by the firm, not staff members' personal accounts, so access can be revoked when someone leaves. Limit AI access per client as well, so a staff member working on one company does not open another client's archive. Indonesia's Personal Data Protection Law (UU PDP, Law No. 27 of 2022) broadly expects organisations to keep the personal data they process secure and to use it only for its intended purpose. Add professional confidentiality on top of that, and your firm needs to be able to show that client data is handled with care.
Review discipline for AI in accounting work
Responsibility for a report stays with the accountant or consultant who signs it. AI speeds the work up, but a small error in the numbers can lead to corrections, administrative penalties for the client, or a loss of trust. Review should be a written routine that does not depend on each person's memory. The checklist below is a reasonable place to start:
- Check that the number of extracted rows matches the number of source documents.
- Recalculate totals and differences in a spreadsheet or accounting system instead of trusting figures produced by AI.
- Manually check every document flagged as blurry, cropped, or in an unusual format.
- Ask the AI to point to the part of the document its answer is based on, then open that document.
- Compare any reference to a tax rule with the regulation currently in force.
- Note in the working papers which parts were prepared with AI and who reviewed them.
This looks like extra steps, but in practice checking is far quicker than retyping from zero. What matters is that everyone in the firm follows the same steps. When a mistake does slip through, you can trace the stage where it got past and fix the process.
Start with one type of work
Firms that get good results rarely use AI for everything at once. Pick one recurring task that is easy to measure, such as compiling input tax invoices for a building supply store or a distributor with a large volume of documents. Record how long the task normally takes and how many corrections come up. Run it with AI for one or two reporting periods, then compare.
Involve the staff who do this work every day from the beginning. They know which documents tend to cause trouble and which formats are confusing. Once one task runs well, extend the same rules to the next one.
A document workspace for accounting firms
Folders is built for tax consultants, accountants, notaries, and law firms whose staff already use AI. You decide which client folders Claude or ChatGPT may read over MCP, every access is recorded in an audit log, and NIK and NPWP are masked before they reach the AI. Its OCR reads documents in Indonesian and English, and answers drawn from your documents come with citations to the source.
