AI agents for finance teams: the recurring work they take over
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.
What do AI agents do in a finance team?
An AI agent carries out a task in your systems, from start to finish. That is the difference from a chatbot or a reporting tool. The agent reads the bank line, finds the invoice, checks the amount, books the match and records why. What falls outside your rules comes back to you as a question.
For a finance team in a growing company that means concrete work that now lands at fixed points in the month. Matching hundreds of bank lines, coding purchase invoices, the accruals at month end, chasing invoices that are 21 days overdue. The work does not disappear, but it is done as it arrives, by agents that do not get tired on day three of the close.
Thred builds those agents as one team, each with its own job, on the accounting software you already use, starting with Exact Online.
Which work do agents take over today?
| Work | What the agent does | What you do |
|---|---|---|
| Bank reconciliation | Matches every bank line to an open item, including batch and partial payments | Decides on the lines the agent cannot explain |
| Purchase invoices | Reads, codes to account and VAT code, books with the document attached | Approves new suppliers and unusual amounts |
| Month-end close | Runs the checklist, prepares accruals and recurring journals, books depreciation | Reviews exceptions and signs off |
| Credit control | Sends reminders on schedule, records and checks payment promises | Handles disputes and decides on escalation |
| Payments | Checks for duplicates and changed bank details, prepares the payment run | Signs off the batch |
| Reporting | Explains movements against last month and budget | Discusses the story with management |
| Data quality | Scans for suspense balances, duplicates and wrong VAT codes | Approves the corrections |
How is an agent different from automation you already have?
Most finance teams already automate something. Bank rules, recurring entries, an invoice scanner, an RPA script someone built three years ago. All of these follow fixed instructions. They work until the input changes: a customer who pays two invoices together, a supplier with a new invoice layout, a bank description with a typo.
An agent works from goals and rules instead of fixed paths. It knows that it has to find the open item for this payment, and it uses everything available to do so: the amount, the reference, the payer's name, earlier payments from the same customer. When it finds the answer, it acts within your limits. When it does not, it asks, with its reasoning attached. That is why agents handle the exceptions that rules leave behind, and the exceptions are where the time goes.
How do you stay in control of what agents do?
- Start read-only. Let the agent analyse and propose for a few weeks, and compare its proposals with what your team would have done.
- Approve each write at first. Every entry shows the administration, accounts, amounts and VAT before it is posted.
- Set rules for routine work. Once a kind of entry has been right a hundred times, allow it to go through on its own, within an amount limit.
- Keep the audit trail. Every action records what the agent did, on which source and who approved it.
- Review the exceptions, not everything. Your time goes to the items the agent flags.
The EU AI Act sets transparency and oversight duties for AI systems; for finance work the practical translation is simple. A person decides on anything that needs judgement, and every automated action can be traced.
What results can a finance team expect?
On Thred Ledger, Thred's own ledger, 99.5% of 36,963 work units in production were completed without a human change, measured on 19 September 2026. Those units include bank classifications, journal entries, purchase and sales documents, reconciliation links and VAT determinations. That is the kind of work agents do well: high volume, rule-bound and checkable against the ledger itself.
For a team on Exact Online the effect shows first in the close. The bank is reconciled every day, purchase invoices are booked on arrival, and the close starts with a short list instead of a backlog. See how to speed up the month-end close.
What do Thred's agents do for a finance team?
Thred is the finance function of one or none: a team of agents that takes the recurring finance work off your desk, on the ledger you already use, under the rules your team sets. The Bookkeeper keeps the record up to date. The Collector manages your receivables. The Payer manages your payables. The Accountant handles filings and optimises for taxes. The Treasurer manages your cash. The Analyst monitors performance and variance. The Connector finds context in your connected tools, from email to Google Drive. And The CFO is your finance lead when you have none.
On Exact Online the agents read and write directly: journal entries, purchase and sales entries, relations, payment matching and fixed assets with depreciation, with two-way sync. The Close agent runs the period close per administration, and the Bookkeeping Quality agent scans for errors before they reach the report. Agents for Xero and Moneybird work alongside. Every write is an approval card until you set a rule, and every action is recorded. Your team decides; the agents do the work. Read what this means for the controller's role or see Thred Layer.
Frequently asked questions
What is an AI agent in finance?
An AI agent is software that carries out a task from start to finish inside your systems, rather than only answering a question. In finance that means, for example, reading a bank line, finding the right invoice, explaining the difference and proposing or booking the match, within rules you set.
What is the difference between an AI agent and RPA?
RPA follows fixed click paths and breaks as soon as a description or amount differs. An agent reads the context, such as a payment reference with a typo or a partial payment, and chooses the right step itself. Where it is unsure, it puts the question to you instead of failing silently.
Can an AI agent post entries in Exact Online?
Yes. Thred writes journal entries, purchase and sales entries, relations, matches and fixed assets in Exact Online. Every write appears as a proposal you approve first, unless you have set a rule that this kind of entry may go through without asking.
Will AI agents replace the finance team?
No. Agents take over the processing work: matching, coding, reminding, preparing. The team keeps estimates, the interpretation of variances and the conversation with management. What changes is where the time goes: less catching up, more analysis, and a team that can support a larger business without growing at the same rate.
Is it safe to give AI agents access to the books?
That depends on how it is set up. With Thred you start with read access, decide per administration whether an agent may write, and every entry is traceable to its source and its approval. Data is hosted in the EU, in AWS Frankfurt, with a data processing agreement in every contract.
Which accounting software do Thred's agents work with?
Thred works on Exact Online, with two-way sync, and has agents for Xero and Moneybird. Smaller companies can also use Thred's own ledger, Thred Ledger, which captures bank, Stripe, Shopify, PayPal and Mollie data and books it automatically.
Sources
- Thred, method behind the 99.5% figure (measured 19 September 2026)
- Thred Layer, AI agents on your existing accounting software
- Exact, Exact Online product information (UK)
- EU AI Act, Regulation (EU) 2024/1689
Read next
- A finance automation playbook for growing SMEs: what to automate first, second and never
- Copilot vs Claude for finance: a factual comparison for finance teams, and when each is better
- What AI tools really cost a finance team, and how to justify the spend
- What makes finance software an agent
- How we measure bookkeeping without a human: the 99.5% method
Where Thred comes in
Finance agents running on top of your existing ledger. You set the rules, approve the work and handle the exceptions. No migration.
