Thred

AI in accountancy: what works today in a compilation practice

Manoj Tutika · Published 16 November 2026 · Last checked 16 November 2026

Working on this yourself? Talk to Mark, our co-founder. We’re here to solve it with you.

Book a demo

What can AI actually do in a compilation practice today?

Most of the time on a compilation engagement does not go to judgement. It goes to everything before it: checking bank lines, clearing suspense accounts, checking VAT codes, updating depreciation, collecting questions for the client. In 2026 AI can take over exactly that work, because it repeats, it lives in the ledger and it can be tested against rules.

What AI does not take over is responsibility. The Dutch professional bodies NBA and NOREA say it plainly in their AI guide 2: AI Toegepast of 4 June 2026: AI applications support the accountant, who keeps the responsibility. That is also the useful line for your firm. Let AI do the groundwork, and make every result traceable to the books, so you or your team can check it in seconds.

So the question is not whether AI replaces an accountant. The question is which part of the groundwork your team still does by hand, when an agent could do it every week for every client.

Which tasks are ready for AI now?

Not every task is equally ready. The more a task can be checked inside the ledger itself, the better AI does it. The more it depends on knowledge that only sits in the client's head, the more often the agent returns a question instead of an answer.

Compilation tasks and what AI does with them today
TaskWhat the agent doesWhat you do
Suspense and query accountsWorks out every balance, with the oldest open item and its ageDecides on the items the agent cannot explain
VAT codesTests every purchase and sale against the counterparty's country and VAT number, flags missing reverse chargeApproves the correcting entry
Duplicate entriesFinds the same invoice twice, also when a bank statement was imported into two bank journalsKeeps one
DepreciationBooks the period's depreciation and keeps the asset card up to dateChecks useful life and residual value on new assets
Analytical review before the client meetingCompares with earlier periods and flags movements above your thresholdDiscusses the explanation with the client
Questions for the clientWrites up the open questions per client, short and specificSends them, or edits them

The NBA guide lists similar uses in section 3.11 (compiling the annual accounts): analytical reviews with preset thresholds, for example a movement of more than 20%, and consistency checks on the treatment of cash, VAT rates and cut-off.

What is the difference between a chatbot and an agent?

A chatbot answers a question about text you paste into it. An agent works inside the books. It reads journal entries, counterparties and VAT codes straight from the accounting software, runs a series of checks and records what it checked and what it found.

For a firm that matters in three ways.

  • You do not copy client data into a public AI model. The agent works in a closed environment on the connection you set up.
  • The work runs on a schedule. Every Monday, or on the fifth working day after the period, for all clients at once.
  • Every finding points to the entries it is about. That is what the guide asks for: results you can verify and a record you can trace.

What do the professional bodies say about AI?

The NBA has two guides. AI in Control (2024) covers controlling AI systems. AI guide 2 (June 2026, with NOREA) covers using AI and is explicitly written for SME accountants too. Its core: validation and human judgement stay the basis, black box AI cannot be justified, and client and personal data does not belong in public AI models such as the free versions of ChatGPT and Gemini.

SRA, the Dutch network of accountancy firms, published a series of four AI white papers in 2026. The last part, AI in beleid (24 July 2026), covers the EU AI Act, the DPIA, shadow AI, a step-by-step route to an AI policy and the questions to ask a supplier.

How do you start without risking quality?

  1. Pick one recurring check your team does by hand today, for example the weekly check of suspense accounts and VAT codes.
  2. Let the agent read only at first. For four weeks, compare its findings with what your team finds.
  3. Write down your firm's rules in plain language: which threshold, which accounts, what you want to see and what you do not. An agent follows those rules to the letter.
  4. Turn on writing per client only, and have every entry approved first.
  5. Record in the file which check the agent ran and who reviewed the result.

Firms that start this way see within a month where the agent beats a manual sample, and where it still needs a person. That insight is worth more than any demo.

What do Thred's agents do with this work?

Thred is a team of AI agents that takes the recurring finance work off your desk, on the accounting software you already use. For accounting firms it works from one portal on the Exact Online administrations of all your clients, with agents for Xero and Moneybird alongside.

The Bookkeeping Quality agent reviews each client's new entries every week across nine kinds of error, from suspense accounts and duplicate invoices to VAT codes, private expenses and closed periods. The Close agent runs the period close per client and marks every item passed, needs you or failed, with the evidence attached. Corrections arrive as entries you approve. Per client you decide whether agents only read or also write.

Behind them stands the full team. The Bookkeeper keeps the record up to date, The Accountant handles filings, The Analyst explains the numbers, and The Collector and The Payer manage receivables and payables. Your firm grows without the groundwork growing with it. Read more on the page for accounting firms.

Frequently asked questions

Will AI replace accountants?

No. AI takes over the groundwork, such as reconciling, reviewing and flagging. Professional judgement, advice and sign-off stay with the accountant. The Dutch NBA guide on applying AI says explicitly that the accountant keeps the responsibility.

Which AI tools do accountants use?

Firms usually combine a business chatbot for technical questions and drafting, the AI features in the client's accounting package, and agents that do the recurring groundwork across all administrations. The difference is whether a tool works inside the books and across all clients at once.

Can I put client data into ChatGPT?

Not into the free version. The Dutch NBA guide names the free versions of ChatGPT and Gemini as places where client and personal data does not belong. Use a closed environment with a data processing agreement.

How do I start with AI in my firm?

Pick one recurring check, let an agent run it read-only next to your team for four weeks and compare the results. Write your firm's rules in plain language, turn on writing per client only, and record in the file what the agent did.

What is the difference between a chatbot and an AI agent?

A chatbot answers questions about text you paste in. An agent works inside the books, reads entries straight from the accounting software, runs checks on a schedule and records what it checked and found.

Sources

  1. NBA and NOREA, AI guide 2: AI Toegepast (4 June 2026, Dutch)
  2. Accountant.nl, NBA and NOREA publish new guide on applying AI (5 June 2026, Dutch)
  3. NBA, AI in Control guide (2024, Dutch)
  4. SRA, White paper AI: in beleid (24 July 2026, Dutch)

Read next

Where Thred comes in

Accounting Firms

Make your firm more efficient with AI, so you have more time for relationships, new customers and advisory work.

Finance, handled.

Book a demo, or request a callback and we will call you.

Book a demo