Banking automation software: frequently asked questions

The Reconciliation Break Resolver is for sample data only, and the page says so above the input. Matching runs in your browser. When you press Run, the break summaries (reference, amount, currency, value date and the computed difference) are sent to the classification model so it can write the classification and the suggested action. Bankautomation does not store the rows you paste, and there is no account behind the demo. Do not paste production or customer data. The privacy policy spells this out.

The day-one path is file based, because that is what every bank can produce this week: camt.053 and camt.054, MT940, MT942 and MT950, pain.001 and pacs.008, card and ATM settlement files, core banking exports and CSV. Where an API exists on your side, a source can be configured to pull from it on a schedule instead. We do not publish a list of pre-built core banking connectors, because the honest answer is that it depends on what your core exposes.

No. Bankautomation is operations software: it matches, classifies, routes, records and evidences. Suggested classifications and actions are decision support. Approvals, compliance determinations, regulatory filings and the control framework itself stay with your institution and its people. Automation supports the control, it does not become the control.

The design intent is weeks rather than quarters, because the bank formats are already parsed and matching rules are configuration rather than development. That is a statement about how the product is built, not a measured customer average, and we will not pretend otherwise. What takes the time on your side is agreeing tolerances, routing and who approves what.

Your operations team, in the interface, under maker-checker. One role proposes a tolerance, a reference normalization rule or a routing target, another approves it, and the change is versioned against every item it later touches. No vendor change request, no specialist consultant, no release window.

Yes. ISO 20022 messages are first-class inputs, not a conversion step bolted on the side, and structured remittance and party data is used by matching rather than flattened into a memo field. See ISO 20022 payment automation for the detail and ISO 20022 explained for the background.

SAML and OIDC single sign-on, a role model with segregation of duties, maker-checker approvals, an immutable audit log and a data residency choice are available from the Platform plan up. The controls are described in full on the security and compliance page. We describe controls we implement, and we do not claim certifications we have not been through.

Four plans, published: Operations at $1,200 a month, Platform at $3,900, Institution at $9,500 and Enterprise by agreement. Annual billing is cheaper per month. There is no free plan and no standing trial. The resolver on this site is the free taste, and it works without an account. No card is required to create an account. See pricing.

Invoicing on Platform and above, purchase orders on Institution, and full procurement support on Enterprise: MSA and DPA, security questionnaire responses, architecture and data flow documentation, and a named contact through the review. Create an account and we will start that thread by email.

It ages, visibly. Every break carries an age in days, an owning queue and an escalation point, and aged items are the first thing on the worklist and in the daily summary. Nothing is closed silently: a write-off or a reversal is an approved action with a reason code attached to the item.

Data residency choice is available from the Platform plan, and private or VPC deployment is an Enterprise arrangement discussed during procurement. If a self-managed deployment is a hard requirement for your second line, say so in the first email and we will tell you plainly whether it fits.

A deterministic matching engine produces the break and the exact difference: amount, value date, reference, currency. A language model then writes the classification, the suggested action and the audit note from that structured difference. If the model is unavailable, a rule based classifier writes them instead and the interface says which one you are looking at. The numbers never come from the model.