Two Sources of Truth: Bank Feed vs Ledger

Reconciliation sounds like an accounting-specific ritual, but at its core it's a simple, familiar idea: you have two records of the same reality, and they should agree. Your bank feed says what actually moved through your account. Your internal ledger or accounting export says what your books believe happened. Reconciliation is the process of comparing those two and finding where they disagree.

Why two sources instead of trusting one. Neither your bank feed nor your internal ledger is inherently "more true" — they're built differently and fail differently. A bank feed reflects exactly what cleared, but timing can lag (a check written Tuesday might not clear until next week) and it has no idea about intent or categorization. Your internal ledger reflects what your team recorded, which can be wrong due to a typo, a duplicate entry, a missed transaction, or a timing difference where something was recorded before it actually cleared. When the two agree, you have real confidence. When they don't, that gap is exactly where errors, fraud, or simple mistakes tend to hide — which is why reconciliation matters even when everything is probably fine.

Common reasons two sources legitimately disagree — not every mismatch is an error:

• Timing differences. A transaction recorded in the ledger on the last day of the month might not clear the bank until the first few days of the next month. This is normal and expected, not a mistake — but it still needs to be identified and tracked, not ignored.
• Outstanding items. Checks written but not yet cashed, deposits made but not yet cleared. Normal, but they need to be listed explicitly so nobody mistakes "not yet cleared" for "missing."
• Bank fees or interest that hit the bank feed automatically but haven't been recorded in the internal ledger yet.
• Genuine errors — a transaction entered twice, an amount typo, a transaction recorded in the wrong period, or a transaction that's simply missing from one side entirely.

What reconciliation is actually for. It's not about producing a report that says "everything matches" — it's about producing a short, honest list of every place the two sources disagree, sorted by how much it matters, so a person can look at each one and decide: normal timing difference, needs a correcting entry, or something worth investigating further. The value isn't in confirming agreement — it's in surfacing disagreement fast, before it compounds across months.

Where AI fits, and where it stops. AI is well-suited to the comparison and pattern-recognition part of this — matching transactions across two messy datasets where dates, amounts, or descriptions don't line up perfectly, and summarizing what kind of discrepancy each mismatch looks like. AI does not decide which discrepancies are acceptable timing differences versus real problems, and it never corrects either source. That judgment call, and any resulting correction, belongs to a person — which is the entire subject of the next lesson.

▶️ Try this

Pull up (or imagine, if you don't have one handy) your own bank statement and your internal record of the same period — even a personal budget spreadsheet versus your bank app works as a stand-in. Pick five transactions and check: do they match exactly? Where they don't, is it a timing difference, a genuine error, or something you're not sure about? That three-way sort is exactly the judgment reconciliation asks a human to make.