Accounting intelligence
How AI changes reconciliation
The shift from matching every transaction to reviewing exceptions, evidence, and decisions.
For years, the central question in reconciliation was: Which record matches this transaction? As software becomes better at finding likely matches, a more useful question moves to the front: What still needs a person’s attention, and what evidence would let them decide?
The shift
Traditional bank reconciliation compares bank activity with accounting records, then investigates what does not line up. Rules and software already handle some predictable entries. AI extends that assistance by using more context to propose or complete matches, while leaving uncertain items for review.
Xero’s account of its own automatic bank reconciliation makes the change concrete: it uses matches, rules, past behaviour, and predictions; it reconciles when confidence is high and leaves uncertain transactions to a person. Its review screen shows the method used and brings supporting documents into view. [1] Xero: What’s new in Auto Bank Reconciliation (opens in a new tab)
The familiar workflow
Find a transaction. Search for a record. Compare details. Resolve the difference. Repeat.
The emerging workflow
Collect context. Suggest a match. Surface uncertainty. Review evidence. Record the decision.
From match to meaning
A matched amount is a start, not the whole answer. Two records can agree numerically while still pointing to the wrong customer, period, account, or supporting document. The useful system tells a reviewer why a match was proposed and which source records support it.
That moves the work from repetitive comparison toward interpretation. KPMG describes an “intelligent close” built around trusted transactions, exception-based accounting, real-time reporting, and a modern workforce. It specifically describes real-time reconciliation that matches transactions and identifies discrepancies. [2] KPMG: From digital to intelligent close (opens in a new tab)
The exception is the work
An unmatched bank line could mean a late invoice, a bank fee, a duplicate, a timing difference, or a genuine error. The label “unmatched” does not tell the accountant which of those is true. A useful exception view should show the transaction, candidate records, missing evidence, and a clear next action.
That is also why a queue should be prioritised by consequence, not just age. A low-value recurring item with clear history is different from a material payment without a source record. AI can help sort and explain those cases, but the criteria for escalation need to be visible to the practice.
Evidence before confidence
Confidence is useful only when a reviewer can inspect what sits beneath it. A suggested match should keep the bank line, ledger entry, invoice or receipt, and any relevant context together. If evidence is absent, the system should say so plainly and make it easy to ask for it.
This is a design principle, not a claim that every AI prediction is reliable. KPMG places trusted transaction data and controls at the foundation of its intelligent close model. In the adjacent field of audit, ICAEW says safeguards and human oversight are essential, alongside policies for confidentiality and responsible use. [2] KPMG: From digital to intelligent close (opens in a new tab) [3] ICAEW: Artificial intelligence in audit work: managing the risks (opens in a new tab)
The human decision
The goal is not to ask an accountant to rubber-stamp an algorithm. It is to give them a shorter, better explained list of decisions: approve a well-supported match, correct a wrong suggestion, request a missing document, or investigate an anomaly.
For BLVNK, that is the direction of accounting intelligence: keep the source of truth in the accounting platform, bring exceptions and evidence into a reviewable workflow, and leave the decision with the practitioner. As the tools improve, the measure of quality should be more than how many transactions they process. It should include whether people can understand, challenge, and trace the outcome.
Sources
These sources informed this article. The workflow and priorities described for BLVNK are our interpretation.
- 01Xero — What’s new in Auto Bank Reconciliation
How automated matching, confidence, supporting documents, and review work together.
- 02KPMG — From digital to intelligent close
The case for continuous reconciliation, exception-based workflows, and controls.
- 03ICAEW — Artificial intelligence in audit work: managing the risks
Professional guidance on human oversight, policies, and confidentiality in audit work.