# Best practices for Bank Statements Import > A few habits keep your reconciliation fast, accurate and safe (Apps: Bank Statements Import) - Page: https://dashboard.nizu.io/knowledge_base/view/246-best-practices-for-bank-statements-import - Part of: [Knowledge base](https://dashboard.nizu.io/knowledge_base/llms.txt) on NIZU — https://dashboard.nizu.io/llms.txt - Updated: 2026-11-13T15:10:04Z ## Best practices for Statements Import A few habits keep your reconciliation fast, accurate and safe. ### Build one template per bank export, and reuse it Bank exports rarely change format. Invest a couple of minutes to build a clean map template from a real sample once; every future import from that bank is then one click. If a bank offers several export formats, prefer the one that includes the structured communication and the counterparty IBAN. ### Always attach the IBAN to the bank account Configuring the account IBAN lets the app warn you if you upload a statement from the wrong account. It also keeps reconciled income tied to the correct payment method for reporting. ### Start with manual review; raise auto-confirm gradually Keep the auto-confirm threshold at 0 at first and watch how the confidence scores line up with reality for your data. Once you trust high-confidence matches (for example structured-communication matches with the exact amount), raise the threshold so those reconcile automatically and you only review the rest. ### Import in order and don't fear overlaps Imports are idempotent: the app fingerprints each line and skips anything it has already imported, and it refuses a file whose exact contents were imported before. Overlapping date ranges will not create duplicate payments. ### Encourage customers to use the structured communication The single biggest accuracy win is customers paying with the structured communication (OGM/VCS) printed on the invoice. Its check digit makes the match all but certain. Where you can, put it prominently on your invoices. ### Reconcile promptly and use Ignore deliberately Review each import soon after uploading, while the context is fresh. Mark genuine non-business lines (internal transfers, bank fees, reversals) as **Ignore** so your reconciled/outstanding counts stay meaningful. ### Reverse instead of editing If you reconcile the wrong invoice, use **Reverse** it removes the created payment and refreshes the invoice status, then re-match correctly. This keeps a clean audit trail. ### Control who can reconcile Reconciliation writes financial records. Grant the *reconcile* permission only to trusted staff; you can still let others upload and review without the ability to post payments. ### Use AI as a helper, not an autopilot Turn on AI reconciliation for the awkward, human-worded payments the deterministic engine misses not as a replacement for it. Let the exact matches (structured communication, invoice number, exact amount) be handled deterministically first, then run AI on what's left. Always review AI suggestions before confirming: they are marked *AI (provider): …* so they are easy to spot. Remember each AI suggestion is a paid API call and sends transaction details plus candidate invoice data to your chosen provider enable it only where that is acceptable, and prefer a provider you already trust with your data.