Automated bank reconciliation: how it works and how AI handles payment reconciliation

A practical guide to bank reconciliation: the manual process, the differences that always show up, why spreadsheets stop scaling, and what changes when AI reads payment receipts and reconciles them on its own.

Bank reconciliation is a task every finance team knows by heart and almost nobody enjoys. You compare what the bank says with what your books say, find out why they differ, and document every difference. This guide covers what it is, how to do it manually, the usual reconciling items, why spreadsheets stop scaling, and how automated bank reconciliation works, including the hardest case: customer payments that arrive as a screenshot in a chat.

What is bank reconciliation?

Bank reconciliation is the process of comparing the transactions and balance on a bank statement with the company's accounting records for the same period, identifying the differences, and explaining or correcting them. The result is a reconciled balance: confidence that the cash in your books is the cash in the account, or that every difference has a known cause.

The two balances rarely match at first, and not because something is wrong. Some transactions are already in your books but not yet processed by the bank, and the other way around: fees or debits the bank already posted that your team has not recorded yet. Reconciliation sorts those differences out.

Why reconcile?

  • Know how much cash you really have. Without reconciliation, the book balance is an estimate.
  • Catch errors early. A mistyped amount, a payment recorded twice, or a transaction posted to the wrong account.
  • Spot fraud and unauthorized charges. Debits nobody approved or fees that should not be there.
  • Control collections. Know which customers paid, how much, and what is still open.
  • Close the month with clean numbers. Up-to-date reconciliations make the month-end close and any audit much simpler.

How to reconcile a bank account, step by step

The manual process has the same structure regardless of company size.

  1. Pick the period. Monthly is the most common. Companies with high transaction volume reconcile weekly or daily.
  2. Gather the data. The bank statement for the period and the general ledger for the cash account over the same dates.
  3. Check opening balances. They should match the reconciled balance from the previous period. If they do not, fix that first.
  4. Match what agrees. Go through each statement line and find its counterpart in the ledger by amount, date, and description. Matched items get checked off.
  5. List what does not match. Everything left unchecked, on either side, is a reconciling item.
  6. Classify each difference. Is it a timing difference that will clear on its own, a transaction missing from the books, or an error?
  7. Post the adjustments. Record what the bank already processed (fees, interest, automatic debits) and correct your own errors.
  8. Build the reconciliation statement. Start from the bank balance, add and subtract the outstanding items, and arrive at the book balance. If the difference is zero, you are done.
  9. Document it. Who prepared it, when, what is still outstanding, and why. That record is what gets reviewed in an audit.

Common reconciling items

These show up in almost every reconciliation. Knowing them speeds up the work.

ItemWhat it isWhere it gets adjusted
Outstanding checksYou recorded the payment, but the payee has not cashed the check yetNo correction needed: clears when cashed
Deposits in transitYou recorded a deposit the bank has not credited yetNo correction needed: clears when credited
Unidentified incoming transfersThe bank credited a payment your team has not matched to a customerIn the books, once the payer is identified
Bank fees and chargesMaintenance, wire, or service fees debited by the bankIn the books
InterestInterest earned or charged by the bankIn the books
Automatic debitsRecurring charges debited automaticallyIn the books
ErrorsWrong amounts, duplicates, or entries posted to another accountWherever the error is, usually the books

In practice, the row that eats the most time is not a technical one. It is unidentified incoming transfers: the bank shows a deposit with a truncated name or an account number, and someone has to figure out which customer and which invoice it belongs to.

Types of bank reconciliation

By method, reconciliation usually falls into three types:

  • Manual. A person compares the statement against the ledger line by line. Fine for a handful of transactions, unmanageable as volume grows.
  • Semi-automated. A spreadsheet or an accounting module matches exact amount and date pairs, and the rest is handled by hand.
  • Automated. A system pulls transactions, matches them using rules and AI, records what agrees, and routes only exceptions to review.

By what is being compared, you will also hear about bank reconciliation proper (bank versus books), payment reconciliation (payments received versus customer invoices or balances), vendor payment reconciliation, credit card reconciliation, and intercompany reconciliation. In practice they overlap: reconciling the bank forces you to identify customer payments, and reconciling customer payments forces you to look at the bank.

Bank reconciliation in a spreadsheet, and why it stops scaling

Spreadsheets are the most common reconciliation tool, and they work to get started. The typical setup is one sheet with the exported bank statement, another with the ledger, a key column to match on (amount plus date, or a transaction reference), a lookup formula such as VLOOKUP, XLOOKUP, or MATCH to flag items with a counterpart, a filter for the ones without, and a summary that walks from the bank balance to the book balance.

It breaks down when real life shows up:

  • One transfer pays several invoices, or one invoice is paid in two installments. Matching by amount finds nothing.
  • Bank descriptions do not say who paid. Someone has to open each transaction and investigate.
  • Multiple accounts and digital wallets, each exporting in a different format that has to be normalized first.
  • Formulas break silently. Someone inserts a column or pastes data in another format, and matching stops working without anyone noticing.
  • No audit trail. The spreadsheet does not record who decided a difference was explained.

At that point the problem is no longer the tool, it is the process: reconciliation depends on one person who knows every case from memory.

How automated bank reconciliation works

Automated bank reconciliation replaces manual matching with a system that pulls bank transactions, compares them with accounting records, and records matches without human input. Off-the-shelf tools handle the standard case well: they connect bank accounts, match on amount, date, and description, and apply rules to recurring items such as fees and automatic debits.

The difference between rules-based and AI-driven bank reconciliation automation shows up in the exceptions. Rules work when data matches exactly. AI adds judgment: it recognizes that two differently written descriptions refer to the same party, understands a partial payment or one that covers several invoices, and explains why it proposed each match. Either way, the golden rule stays the same: anything that does not match is reviewed by a person, and every decision is logged.

The hard case: payment reconciliation with informal receipts

There is one scenario where standard tools fall short, and it is common in businesses that collect from many small customers: medical practices and clinics, schools, retailers, subscription services. The customer does not pay through a portal. They pay by bank transfer or from a digital wallet and send the receipt however they can: a screenshot, a phone photo, a PDF, often through a messaging app like WhatsApp.

That adds a step no bank feed solves: knowing what the receipt says and who it belongs to. Today, in most companies, someone opens every image, reads the amount, looks for the transaction in the bank, finds the customer's balance, and marks it paid.

How AI handles it

With intelligent document processing, that whole path is automated end to end:

  1. Read the receipt. AI understands the screenshot, photo, or PDF from any bank or wallet and extracts amount, date, payer, and transaction reference, each field with a confidence score.
  2. Find the transaction. It checks the extracted data against the bank statement or account to confirm the money actually arrived. A receipt without a matching transaction is never marked as paid.
  3. Reconcile against the invoice or balance. It identifies the customer and applies the payment to what they owe, even when they pay several invoices at once or only part of one.
  4. Calculate balances. It updates what the customer owes and flags partial, duplicate, or unidentified payments.
  5. Send only mismatches to review. An unreadable receipt, an amount that does not add up, or a missing transaction goes to a person, with the original image next to the extracted data.
  6. Report to finance. A summary of what was reconciled and what is pending, with a trail of every step.

For this to work, the solution has to be connected to where the balance lives: the ERP, the management system, or the company's own software. That is the automation and integration side: the receipt does not end up in a folder, it ends up recorded as a payment in the right system.

Real case: M.I.C.A with Integrando Salud

M.I.C.A is the AI agent Axlan built with Integrando Salud for clinics and medical practices, and it runs on WhatsApp. Among other tasks, it reads and reconciles payment receipts from any bank or wallet (photos, screenshots, and PDFs) that patients send by chat, calculates balances, and sends audit reports to the administration team. The team no longer opens every image: they review the report and handle only the exceptions.

It is the same problem any business faces when customers pay by transfer and send receipts by chat, which is why the approach carries over to other industries. What changes is the system where balances live and each business's rules.

When to automate reconciliation

  • When reconciliation falls behind. If the month is reconciled weeks later, it no longer helps you make decisions.
  • When most of the time goes into identifying who paid, not into matching amounts.
  • When payments arrive through many channels: several banks, digital wallets, receipts by chat or email.
  • When it depends on a single person who knows every case from memory.
  • When your accounting system's reconciliation only handles exact matches, or does not cover what you need.

If your case is standard, a good off-the-shelf reconciliation module may be enough. If you deal with informal receipts, custom rules, or a system without a ready-made connector, a custom solution is the better fit.

Frequently asked questions

What is automated bank reconciliation?

It is reconciliation done by a system: it pulls bank transactions, matches them against accounting records, records what agrees, and sends only what does not to human review. With AI, it also handles differently worded descriptions, partial payments, and payments that cover several invoices.

What is the difference between bank reconciliation and payment reconciliation?

Bank reconciliation compares the bank statement with your books. Payment reconciliation matches the payments you received against customer invoices or balances. They overlap: identifying an unexplained deposit in the bank is a payment reconciliation problem.

What are the three types of bank reconciliation?

By method: manual, semi-automated (a spreadsheet or system matches exact pairs and the rest is handled by hand), and automated (a system matches with rules and AI and routes only exceptions to review).

How often should you reconcile your bank accounts?

At least once a month, before closing. Businesses with high volume or many paying customers benefit from reconciling weekly or daily, because recent differences are easier to explain.

Can AI reconcile payment receipts sent by chat?

Yes. AI reads the screenshot, photo, or PDF, finds the transaction in the bank, applies it to the customer's invoice or balance, calculates the remaining balance, and sends only mismatches to review. That is what M.I.C.A does with patient payments.

Does automation replace the finance team?

No. It removes the repetitive part: opening receipts, finding transactions, and matching amounts. The team still decides on exceptions and reviews the reports, with more time for analysis.

Wondering how AI
can help your company?

Tell us about a process that eats up hours today. We'll tell you whether it can be automated, how, and with what.

Tell us about your process