It is the fourth working day after month end. One accountant is matching a long bank statement line by line, another is keying vendor bills from memory of how the last one was coded, and the internal auditor is preparing to test a sample of a few dozen entries out of thousands. None of this is wrong. It is how most finance teams in Cairo and Riyadh still close, and exactly where AI is now being sold to them.
This guide is the hub of a twelve-part MFT series on AI in accounting, audit and Odoo 20: what AI actually does in the accounting cycle, what it changes in your controls, and how to begin without trading your audit trail for speed.
What is AI in accounting?
AI in accounting is software that learns from your existing records, or reads documents and questions in natural language, to prepare, match, classify or test accounting data that a person then reviews and approves. That definition matters because it excludes two things often sold under the same label: fixed rules (a reconciliation model that always books bank charges to the same account is automation, not AI), and autonomous posting (software that books entries with nobody accountable is a control failure, whatever it is called).
Four capabilities sit under the label, each needing a different control:
- Prediction: suggesting the account, tax or analytic tag for a bill line from history. The risk is that it repeats yesterday’s coding error with confidence.
- Matching: pairing bank transactions with invoices, bills and payments. The risk is a wrong match that clears a balance which should have stayed open.
- Anomaly scoring: ranking every transaction by risk, such as duplicates, late round-amount manual entries or approvals outside the matrix. The risk is treating a flag as a finding.
- Language agents: answering questions about the ledger, reading documents and drafting records or entries. The risk is a fluent answer with no record behind it.
Where does AI help in the accounting cycle?
Read it process by process: what the machine does, and what a person must still own. The table maps the cycle to the article in this series that goes deeper.
| Process | What AI does | What a person still owns | Go deeper |
|---|---|---|---|
| Journal entries and bills | Predicts lines, extracts documents, drafts recurring and correcting entries | Approval, coding policy, reversal of errors | Automating journal entries and bank reconciliation |
| Bank reconciliation | Matches transactions to open items, proposes write-offs, flags suspected errors | Tolerance limits, unexplained differences, reviewer sign-off | AI bank matching and its controls |
| Fraud red flags | Scores duplicates, near-duplicate vendors, unusual timing and amounts | Investigation; a flag is never proof of fraud | AI-based financial fraud detection |
| Month-end close | Tracks open items, drafts accruals and variance commentary | Judgement on estimates, cut-off, final sign-off | AI in the monthly close and financial reporting |
| Audit approach | Monitors continuously instead of once a year | Scope, risk assessment, conclusions | Continuous auditing with AI vs traditional auditing |
| Audit testing | Tests the full population, not a sample | Defining the test and following up exceptions | From sampling to full-population testing |
| Access and duties | Runs as a system user with its own rights | Keeping the agent inside segregation of duties | AI, segregation of duties and internal controls |
| Evidence | Cites the records behind each answer | Rejecting any output with no source | AI provenance and the zero-hallucination principle |
| Odoo 20 features | Bill line prediction, financial insights, agents | Deciding which features to switch on | Odoo 20 AI features for accounting |
| Connecting AI to the ERP | Reads ledger data through the API or MCP | Access scope, data residency, approvals | Connecting AI to Odoo for continuous auditing |
| Purchasing | Drafts and routes purchase documents | Approval matrix and thresholds | Automating purchasing and approvals in Odoo 20 |
| Rollout | Runs in pilot on a closed period first | Gates between phases | Step-by-step AI accounting on Odoo 20 |
How many finance teams actually use AI in 2026?
Fewer than the marketing suggests, and the number has stopped climbing. Gartner’s 2025 AI in Finance Survey of 183 CFOs and senior finance leaders found that 59% report using AI in their finance function, against 58% in 2024 and 37% in 2023. The cited barriers were complexity, data and talent. Two cautions: it measures the finance function overall, not accounting specifically, and “using AI” covers anything from one pilot to a production process.
Our reading: the easy experiments have been run. What separates teams now is whether they can put AI inside a process with defined inputs, a reviewer and evidence, which is a controls question, not a technology one.
What does Odoo 20 add for accounting teams?
According to Odoo’s 20 release notes, the accounting items are concrete: bill line prediction that suggests product, account, tax and analytic values from bill history and the label; AI-powered financial insights that answer questions about receivables, cash flow or reconciliation issues based solely on the accounting data in your database; and duplicate detection on invoices and bills with a red level for highly likely duplicates and a yellow level for potential ones. Agents can now be triggered by automated and scheduled actions, and the database can be connected to outside assistants through MCP.
Two facts shape the business case. All AI features require credits bought through In-App Purchase, billed per request and not included in any plan. And which edition includes AI agents should be confirmed for your own edition and plan before you design around them. The governance of Odoo 20 AI agents is covered separately; the short version is that an agent which can create records is part of your control environment.
What can go wrong when AI touches the ledger?
Confident answers with no record behind them
A language model can produce a balance or a vendor name that reads correctly and exists nowhere in the books. The defence is structural: every figure cites its record, and anything that cites nothing is rejected.
History repeated as policy
Prediction learns from what was posted, not what should have been. If a recurring service was miscoded for a year, the model will propose the same account confidently. Review predicted lines against the coding policy, not against last month.
A new user nobody put in the access matrix
An agent that creates or updates records is a user. If it runs under an administrator’s account, it inherits rights no single person should have. Give it its own user, map it into your segregation of duties in Odoo, and make sure it never approves what it prepared.
Which controls should govern AI in accounting?
The controls are not new. COSO’s 2013 framework already asks for control activities (Principle 10), general controls over technology (Principle 11) and ongoing evaluations (Principle 16). What changes is that the preparer can now be software. We apply five rules:
- The machine drafts; a named, registered person approves. Nothing is posted by the AI on its own.
- Approval is tiered by your ISA 320 materiality threshold: above it, a second approver signs.
- Errors are corrected by reversal, never by deletion, so the trail shows what the AI proposed and what a person changed.
- Every output cites its source records; outputs with no source are rejected and kept visible as rejected.
- Rights, tolerances and prompts are owned by the controller and changed through the same approval as any other configuration.
If the machine cannot show the record, the number does not exist.
How does AI change the audit?
Traditional audit relies on sampling under ISA 530 because testing everything by hand is impossible. Once the ledger can be read programmatically, an auditor can test the full population: every manual entry, every payment, every vendor. ISA 240 already requires testing the appropriateness of journal entries; full-population analytics make that test complete rather than illustrative.
The case for continuous work is about time. The ACFE’s Occupational Fraud 2026 report, covering 2,402 cases from 143 countries and territories, found that the median fraud lasts 12 months before detection, that tips remain the top detection method at 43% of cases, and that proactive data monitoring is associated with lower losses and faster detection. An annual audit, as we have argued, misses eleven months. Standards are moving too: in August 2026 the IAASB issued exposure drafts revising ISA 500, ISA 330 and ISA 520 to address the growing use of technology, with comments due 15 December 2026. They are not yet final.
Where should a finance team start?
Start with one process and a measurement, not a platform decision:
- Pick one process with high volume and clear evidence. Bank reconciliation is usually the best candidate, because every match can be checked against a statement line.
- Baseline it on a closed month: count unmatched lines, days to reconcile and items older than thirty days.
- Clean the master data the AI will learn from, starting with duplicate and near-duplicate vendors.
- Write the approval rule before switching anything on: who reviews AI drafts, and what threshold needs a second approver.
- Run the AI on the closed month and compare its result with the human reconciliation, difference by difference.
- Go live only if the comparison passes, then extend to the next process through the gates in the step-by-step Odoo 20 rollout.
This is also how we work. Our seg-audit platform reads the books straight from the ERP, read-only, and scores every transaction by risk instead of sampling a few percent; a finding cannot be closed without a named approver. Our MFT Intelligence agents draft corrections that a registered person approves, and our internal audit team uses both on engagements. The right first step is a conversation about one process, not a demo.
Sources
- Gartner, "Gartner Survey Shows Finance AI Adoption Remains Steady in 2025" (press release, 18 Nov 2025)
- Odoo, "Odoo 20 Release Notes" (Sept 2026)
- ACFE, "Occupational Fraud 2026: A Report to the Nations" (press release, 19 May 2026)
- IAASB, "Project update: ISA 500, Audit Evidence" (exposure drafts, Aug 2026)