Thred

AI for financial controllers: what to hand over and what to keep

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 does AI change for a financial controller?

A controller is hired for judgement and spends most of the month on processing. Matching bank lines, chasing missing invoices, typing the same recurring journals, tying out receivables to the ledger, rebuilding the same variance table in a spreadsheet. The judgement work, estimates, unusual transactions, the story behind the numbers, gets squeezed into the last two days of the close.

AI agents change that split. They take over the processing, so the controller's time goes where it was always meant to go. The question is not whether to use AI, but which parts to hand over, how to stay in control and where to start.

What do you hand over and what do you keep?

Controller tasks split between AI agents and the controller
TaskHand over to agentsKeep
Bank reconciliationMatching, including batch and partial paymentsLines the agent cannot explain
Purchase invoicesReading, coding, bookingNew suppliers, disputed invoices
AccrualsRecurring items, expected invoices, reversalsProvisions and estimates that need judgement
Recurring journalsRent, insurance, allocations on scheduleChanges to allocation keys
Tie-outsReceivables, payables, VAT and fixed assets against the ledgerInvestigating differences that matter
Variance analysisFirst pass against last month, last year and budgetWhich explanation holds, what management hears
Data qualitySuspense balances, duplicates, VAT codesApproving corrections
Period closeRunning the checklist with evidenceSign-off

Why keep judgement with the controller?

There are three good reasons, and none of them is that AI is not capable enough.

  • Responsibility. The controller signs off the period and answers to the finance director, the auditor and sometimes the bank. That responsibility cannot be delegated to software.
  • Context outside the ledger. A provision for a disputed contract depends on a conversation with legal that the agent was not part of.
  • Policy. How the company treats development costs, when revenue is recognised on a multi-year contract: those are choices under FRS 102 or IFRS, made once and applied consistently.

What AI does is make those decisions better informed. The agent finds the entries behind a variance in seconds; the controller decides what they mean.

How do you stay in control?

  1. Write down the rules in plain language. Materiality threshold, which accounts need approval, which suppliers are routine. An agent follows them to the letter.
  2. Approve every write at first, and look at the detail: accounts, amounts, VAT, period.
  3. Allow routine entries to go through on their own once they have been right consistently, within an amount limit.
  4. Review the log weekly. Which entries were made automatically, which were approved, which were rejected and why.
  5. Keep segregation of duties. The agent prepares, a person approves payments and material entries.

The EU AI Act asks for human oversight and transparency of AI systems. In a finance team that is exactly what approval cards, rules and an action log deliver.

How do you start in four weeks?

Week one, pick one process, ideally bank reconciliation, and give the agent read access. Weeks two and three, let it propose matches every day and compare them with what your team books. Note where it is right, where it asks and where it gets it wrong. Week four, allow writes with approval on each entry, and set the first rules for routine cases. Then widen the scope: purchase invoice coding, recurring journals, the close checklist.

Controllers who start this way know within a month where an agent beats a manual process and where it still needs a person. That is more useful than any demo.

What does the controller's month look like afterwards?

The bank is reconciled every morning. Purchase invoices are booked on arrival. On the last day of the month, accruals and recurring journals are proposed with their basis, and the tie-outs have already run. The close starts on day one with a list of exceptions and a first-pass variance analysis, not with a backlog. That leaves days two and three for review, the estimates and a proper conversation with management. Read more in how to speed up the month-end close.

What do Thred's agents do for a controller?

Thred is a team of AI agents that takes the recurring finance work off your desk, on the accounting software you already use. The Bookkeeper reconciles the bank and books purchase and sales entries as they arrive. The Close agent runs the monthly or quarterly close per administration and marks each step passed, needs you or failed, with the evidence attached; its step agents prepare accruals and recurring journals. The Bookkeeping Quality agent scans for suspense balances, duplicates, wrong VAT codes and entries in closed periods. The Analyst explains movements against last month and budget. The Collector and The Payer keep receivables and payables moving, and The Treasurer keeps an eye on cash.

On Exact Online the agents read and write directly, with two-way sync, and Thred books depreciation while updating the asset card. Every write arrives as an approval card until you set a rule for it, and every action is logged. On Thred Ledger, 99.5% of 36,963 work units were completed without a human change, measured on 19 September 2026. You keep the judgement and the sign-off. See Thred Layer.

Frequently asked questions

Will AI replace financial controllers?

No. AI replaces the processing part of the job, not the judgement. Estimates, provisions, the treatment of unusual transactions, internal control design and explaining the numbers to management stay with the controller. What AI removes is the catch-up work that keeps controllers from doing that well.

What should a controller not hand over to AI?

Anything that needs judgement or carries responsibility: significant estimates such as provisions and bonus accruals, decisions on accounting policy, approval of material entries, the final sign-off of the period and conversations with auditors, banks and management about what the figures mean.

How does a controller keep an audit trail when AI posts entries?

Make sure every entry records its source document, the reasoning and who approved it, and that the agent only writes within rules you set. With approval cards and a log of every action, the audit trail is often more complete than with manual entries, which rarely record why something was booked.

What is the first process a controller should automate with AI?

Bank reconciliation. It has the highest volume, the clearest rules and an immediate effect on the close. Results are easy to check, because a correct match is either right or wrong. After that, purchase invoice coding and recurring journals are natural next steps.

Can AI do variance analysis?

AI can do the first pass very well: compare every account with last month, last year and budget, flag movements above your threshold and find the entries behind them. The controller decides which explanations hold, what is a timing difference and what management needs to hear.

How long does it take to introduce AI agents in a finance team?

A first process can run read-only within a week. Plan four weeks of side-by-side comparison before you allow the agent to write, then widen the scope one process at a time. Most teams have reconciliation, invoice coding and recurring journals running within two to three months.

Sources

  1. Thred Layer, AI agents on your existing accounting software
  2. Thred, method behind the 99.5% figure (measured 19 September 2026)
  3. EU AI Act, Regulation (EU) 2024/1689
  4. FRC, FRS 102 The Financial Reporting Standard applicable in the UK and Republic of Ireland

Read next

Where Thred comes in

Thred Layer

Finance agents running on top of your existing ledger. You set the rules, approve the work and handle the exceptions. No migration.

Finance, handled.

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

Book a demo