You can assess AI for preparing bank reconciliation matches and assembling evidence for differences. Suggestions should distinguish direct, grouped, partial and ambiguous cases without forcing a match. Your team checks the outcome and authorises any accounting changes; matching a transaction does not initiate a payment.
A checkable suggestion is better than a forced match
Two transactions with the same amount give you a clue. If their reference, account, entity and period also agree, you have more evidence to examine. The work to prepare is that chain of evidence, together with a list of everything still unresolved.
Bank reconciliation compares bank transactions with entries in the accounting bank ledger. Microsoft describes this comparison, including linking one statement line to several ledger entries, in its Business Central documentation. The reference explains the method; it does not mean Digital Employee has access to that system or a bank. Source: Microsoft, bank account reconciliation.
Distinguishing bank-to-ledger comparison from allocating a receipt to customer invoices avoids a common misunderstanding. They are connected tasks, but clearing a customer item does not by itself establish that the whole bank account is reconciled.
Set the scope before looking for matches
Choose one account, entity, currency and time window. Identify the statement, ledger export, reference balances and outstanding items carried forward. Record the extraction date: if a source is updated after the comparison, the result must show which version it used.
You can assess authorised exports or read-only access, depending on what the systems permit. An import facility does not establish data completeness or compatibility with every bank. Use the ERP integration checklist for access, formats and technical limits.
Agree how to handle amount signs, transaction dates, value dates and truncated references. A file received twice needs to be recognised before it creates duplicate transactions. Keep the original source: standardising the format of a reference does not authorise changing its meaning.
Six transactions: suggestion, evidence and review
This is an entirely fictitious example. B rows come from statement E-05 and C rows from ledger export L-05 for the same EUR account. The sample covers 30 April to 5 May; positive amounts are receipts and negative amounts are outflows. It is a set of cases to examine, not a complete reconciliation with opening and closing balances.
Scroll the table horizontally to read every column.
| Bank transaction | Suggested ledger match | Available evidence | Review required |
|---|---|---|---|
| B01: +EUR 1,200.00 | C01: +EUR 1,200.00 | Reference RC101 and date agree. | Confirm that neither line is already used in another match. |
| B02: +EUR 2,500.00 | C02: +EUR 1,500.00 and C03: +EUR 1,000.00 | Remittance Z lists both lines; 1,500 + 1,000 = 2,500. | Check the grouping and that neither component is already matched. |
| B03: +EUR 600.00 | C04: +EUR 1,000.00, same reference | Difference: 1,000 − 600 = EUR 400.00. | Establish whether this is a partial payment or a posting needing correction. Do not automatically clear the difference. |
| B04: +EUR 990.00 | C05: +EUR 1,000.00, same reference | EUR 10.00 difference; charge details unavailable. | Request evidence. The difference does not establish that a fee was charged. |
| B05: +EUR 900.00, incomplete reference | C06 or C07: both +EUR 900.00 | Equal amounts; no reference distinguishes the two entries. | Leave the suggestion ambiguous and ask the owner for more details. |
| B06: −EUR 300.00 on 2 May | C08: −EUR 300.00 on 30 April | Same reference, different dates. | Check the timing difference and the item carried forward from the previous period. |
In B03, do not call EUR 400.00 a verified remaining balance before checking the original entry. In B04, do not create a EUR 10.00 posting simply to make the total agree. B05 should remain visible among unresolved items: picking the first candidate merely transfers uncertainty into the records.
Useful rules distinguish different kinds of exception
You can ask for the comparison to start with stronger references before suggesting permitted groupings. Each suggestion should explain which fields agree, which differ and which are missing. A label such as “reference and amount agree” is easier to check than a confidence percentage without a verified method behind it.
Handle grouped payments, partial payments, possible charges and timing differences separately. Agree which records can resolve each uncertainty and who should obtain them. Any amount or date tolerance needs an approved rationale; it should not conceal an unknown cause.
Every line used in a group must remain identifiable. A later run should not reuse it as though it were unallocated. This is a condition to design and verify for the chosen process, rather than a capability guaranteed regardless of the accounting system.
Who approves, and what should they receive?
The reconciliation owner receives the suggestions, sources, open differences and reasons for exclusions. They can confirm, reject or request information. Any postings or corrections follow the company's accounting permissions; the comparison should not turn into an implicit change to the ledger.
Before treating the work as complete, also check transaction coverage, balances, outstanding items and the controls required for that account. A list of successful matches or a total that agrees does not describe all the necessary work. Matching neither authorises nor initiates a payment.
Connect the result to the next task
An unresolved receipt may affect payment reminders: that workflow should know about the hold before contacting the customer. At month-end, retain the open items and their owners, as described in the guide to month-end preparation.
For a bounded assessment, choose samples with straightforward and ambiguous cases, then measure preparation, review and rework time. Our finance teams page includes this support among the processes to assess. The useful question is whether your team can reach decisions with better evidence and a manageable workload while keeping differences visible.