The AI matches your bank statement lines to your accounting entries, suggests the matching (and clears the customer or supplier account when a payment settles an invoice) and flags discrepancies. The accountant keeps the validation. It plugs into the ERP or accounting software you already use.
AI bank reconciliation is not another piece of software: it is an AI layer that plugs into your ERP or accounting software (Sage 100, EBP, Cegid, Odoo…). It suggests the matching (and clears the customer or supplier account when a payment settles an invoice), flags discrepancies, handles unreadable labels, grouped payments, fees and currencies; the accountant validates. Automatic if the ERP offers an API or an integration, semi-automatic otherwise. First step: a trial on one month of real statements.
Simple matching is quick. The time goes into the lines that match nothing at first sight. For each one: the problem as the accountant lives it, then the AI's suggestion, which they accept or correct.
The statement label is cut off, coded or carries no counterparty name. The accountant opens the bank portal, looks for the document, asks the customer or the supplier.
What the AI does
The AI reads the full label, recognizes the counterparty and the reference and suggests the matching entry; the decision stays with the accountant.
A customer pays three invoices in a single transfer, sometimes reduced by a credit note. The statement line matches no entry taken on its own.
What the AI does
The AI looks for the combination of invoices whose total equals the amount received and suggests the grouped matching, clearing the settled invoices in the customer account; the accountant validates.
A deposit or a partial payment leaves an open balance on the invoice. Without follow-up, that balance gets lost or is matched by mistake.
What the AI does
The AI links the payment to the invoice, computes the remaining balance and flags it; the accountant confirms the balance to follow up.
Account fees, transfer commissions, overdraft interest: these lines have no supporting document and sit in a suspense account until the close.
What the AI does
The AI recognizes them and suggests the adjusting entry; the accountant validates, nothing is booked automatically.
The bank credits card or terminal settlements net of commissions, sometimes grouped over several days. The amount matches no sale.
What the AI does
The AI rebuilds the gross amount, the commission and the settlement period, then suggests the matching; the accountant accepts or corrects it.
Rent, subscriptions, loan instalments, insurance: identical lines every month, whose invoice arrives late or not at all.
What the AI does
The AI recognizes the recurrence, suggests the usual posting and flags the instalment whose amount changes; the accountant decides.
An invoice in dollars settled in euros never matches to the cent. The difference has to be isolated and booked as an exchange gain or loss.
What the AI does
The AI matches on the currency amount, computes the exchange difference and suggests the entry; the accountant accepts or corrects it.
A cheque or cash deposit shows up as one global amount, without the detail of the payers. The accountant goes back to the deposit slip.
What the AI does
The AI matches the deposit slip with the open invoices and flags what remains unidentified; the accountant rules on each cheque.
Each bank delivers its statement in its own format and on its own schedule: PDF, CSV, CAMT.053, MT940, CFONB 120. The accountant consolidates before matching.
What the AI does
The AI reads each format, aligns the statements on a single basis and prepares the matching; the accountant validates.
The matching is done in a personal workbook, line by line, the day before the close. It is neither shared, nor traced, nor reused the following month.
What the AI does
The AI keeps the reconciliation up to date through the month, traces each decision and keeps the history; the accountant validates and finalizes the close.
The roles do not move: the AI proposes, classifies and traces; the accountant decides.
AI bank reconciliation is tied to no vendor: it reads the entries of whichever ERP or accounting software you use, and the statements of each of your banks. Examples of software on the market:
Automatic or semi-automatic? The answer lies in how your ERP and your banks deliver their data: an API or a scheduled exchange on one side, plain exports on the other. We settle it during scoping, ERP by ERP and bank by bank; no software is filed in advance in either category.
Automatic reconciliation
Your ERP or accounting software can be queried without intervention: API, native connector from the vendor, direct access to its database or a file exchange that runs on a schedule (a nightly SFTP drop, for instance). On the bank side, the CAMT.053, MT940 or CFONB 120 statement is received automatically, or a bank API is available.
Semi-automatic reconciliation
Neither the ERP nor the bank offers automatic access: the bank account ledger is exported by hand and the statement is downloaded from the bank's portal (CSV, CAMT.053, MT940, CFONB 120).
Whatever the mode, no connector comes off the shelf: the one linking your ERP and your banks (CAMT.053, MT940, CFONB 120 or CSV statements, bank API when the bank offers one) is built for your project, starting at scoping. Microsoft clients: our Dynamics 365 and Power Platform expertise. ERPNext or Odoo clients: our open source ERP expertise.
Three cases where we advise against an AI layer on bank reconciliation, and we say so from the first conversation.
A statement of a few dozen lines per month is matched quickly in your software; scoping a project would take longer than the matching itself.
With a single monthly statement and clean entries, the reconciliation feature of your software is often enough.
Many accounting packages match the simple lines (same amount, same date) and fetch the statements through a bank connection. If that built-in matching covers your cases, switch it on; an AI layer is only justified on the lines it leaves open.
Three steps, from the trial on your real data to the scoping of the project. Bank reconciliation is the last step of a purchase invoice's journey: if your blockers sit upstream (receipt, control, approval, posting), see the AI purchase invoice processing page.
You entrust us with one month of statements, for each bank, and the corresponding bank account ledger, anonymized if needed. The AI suggests the match; we compare it with the one you actually did.
Lines matched without correction, lines left open, discrepancies found: we read the result with the accountant who keeps the bank accounts.
Free initial scoping. There we set the mode (automatic or semi-automatic) according to your ERP, the list of banks and the format of each statement (CAMT.053, MT940, CFONB 120, CSV or API), the validation rules, the hosting and the access rights. The start-up lead time is fixed at this stage.
All of them: Sage 100, EBP, Cegid, Odoo, SAP Business One, Microsoft Dynamics 365 Business Central, ERPNext and the others. The rule is simple: an ERP with an API or another means of automatic integration (native connector, database access, scheduled file exchange) allows automatic reconciliation; an ERP that only allows manual file exports and imports gives a semi-automatic reconciliation. The mode is determined during scoping, depending on your ERP. The connector is built for your project.
CAMT.053, MT940 and CFONB 120 (the French statement standard) electronic statements, CSV exports from your bank, or a bank API when your bank offers one. The format retained and the corresponding connector are defined during scoping, bank by bank.
The accountant, always. The AI suggests the matches and the adjusting entries; nothing is matched or posted without their validation. Each validation is traced.
Statements and entries are stored in the Azure region of your choice, in Europe or in Canada, and do not leave it. The people and systems allowed to consult them are named in the contract. The team delivering the project is in Tunis; the contract itself is signed in France, by Vivantro SAS.
The lead time is set during scoping: it depends on the integration mode (automatic or semi-automatic), the number of banks and the statement formats. The trial on one month of real statements takes place before any commitment.
Bank reconciliation is the last step of a purchase invoice's journey: order, receipt, invoice arrival, control, approval, posting, payment, then matching against the bank statement. Upstream blockers (invoices received through several channels, discrepancies between order and invoice, pending approvals, posting and VAT) are covered on the AI purchase invoice processing page. The two layers are scoped together or separately.
Page reviewed and updated on — Access International, a Franco-Tunisian IT group founded in Tunis in 1999, with a subsidiary in Paris.
Send us one month of statements and the corresponding entries: we return the matching suggestions and the detected discrepancies, then we scope the project together.
Discuss your project →