Payment Reconciliation: why it's still manual and how to automate it
How does the matching currently work?
The easiest way to explain it is to go through a single transaction.
Take accounts receivable. A customer transfers 14,000 euros to your company bank account, they have five open invoices in your books, and you have to work out which one covers that amount. So you open the bank statement in one window and your open invoices in the other, and you manually go down the list until you find the one that matches. Accounts payable is the same job in reverse, a supplier charges your account, the invoice it relates to is sitting in someone's inbox, and at some point you have to go looking for that invoice and match it to the payment. Once that is done, you have to map it to the correct cost centre and account in your ledger.
That is the work, every transaction has to be reconciled, and the act of matching itself is done by hand, for recurring transactions and new ones alike. Your accounting software may already collect every invoice for you, forwarded to its own inbox and read out automatically, but it does not decide which payment settles which invoice. Even at 100 invoices a month, this means hours of someone's time spent on a low-value task.
Why reconciliation is still done manually
German bookkeeping rules (GoBD) work on the principle of no booking without a receipt (Belegprinzip), which means that every financial transaction must be recorded and directly linked to its underlying source document.
Reconciliation happens manually ultimately because your bank account does not hold a record of your existing invoices and, as mentioned above, your accounting system, even with a banking integration, an invoice inbox and OCR technology, still does not know which invoice a payment settles. The setup we see most often is companies that have 1, possibly 2 bank accounts, an expense management tool, an accounting tool that is sometimes managed externally, an invoicing tool embedded into a CRM and manually held spreadsheets that bring all of this together for reporting.
That's a total of at least 5 disconnected systems, and for a typical 50 to 60 person tech company in Germany, which processes roughly 300 to 700 transactions a month, that means every single one of them has to be worked through by hand. On top of that, missing receipts, payments that land in the wrong month and unmatched entries all hold up the month-end close. The immediate solution that comes to mind is to hire more people to get through it faster but that is not a scalable strategy.
What this process costs
Whether we talk about receivables or payables, the cost lies in having to work through transactions one by one. A recurring supplier payment takes roughly one to three minutes: find the invoice or receipt, match it to the payment, confirm the correct account booking. And you spend those minutes again every month, even on transactions that have been booked the same way a dozen times, you might move faster but the process treats every transaction the same. A new invoice takes closer to five minutes, because on top of the searching and matching, you are often working out which account it even belongs to.
At 300 to 700 transactions a month, that is roughly 10 to 45 hours, so up to a full working week, every month. And that is precisely the moment when companies look to either outsource or hire a dedicated person. Naturally, as the business grows so does the volume of transactions, which means even more time spent reconciling. A process where someone has to work through every transaction by hand does not scale.
How to automate the reconciliation process
Firstly, automating reconciliation means having all fragmented data coming in from your existing systems mapped and visible in one single ledger. Secondly, once everything sits in the same place, Agent F automatically matches transactions to open invoices as they come in, and posts them to the right account following your own cost centre logic, which it takes in as context from you, and the standard chart of accounts, such as SKR03 or SKR04 in Germany.
As the end user, your work then shifts from a time consuming matching exercise to an approval flow. Every transaction, recurring or new, comes in with the full booking suggested, the match of invoice to payment, the account allocation in your chart of accounts, and your internal cost centre logic, and each suggestion carries an AI confidence interval. As a finance manager, your job becomes simply approving or rejecting the matches with a click, and dealing with the few outliers: such as a specific expense that is missing a receipt or an overdue invoice that still has to be paid. If you reject a match, the system comes back with another suggestion or correction.
The result is that as your business grows and the volume of transactions grows with it, our ERP system learns your patterns and automates them. Your financial infrastructure scales without extra hires or extra hours of work.
Want to see this in practice? Learn how agent F works, and book a demo at agent-f.ai/demo.