Payments coming in via online payment providers (think Mollie, Stripe, Adyen) need to be matched against sales invoices and against the bank transactions on which the money ultimately appears. This is a matching job: laying amounts, references and dates side by side to establish which invoice belongs to which payment and which payment belongs to which bank transaction. Typical work for bookkeepers and administrative staff, with a payment provider link and the accounting system as tools.
Three axes are decisive here: structuredness, volume and cost of errors.
The task is quite structured (4 of 5). A payment has an amount, a date, a reference or transaction ID. An invoice has the same kind of characteristics. Matching is at its core an arithmetic and logical comparison, not a task that requires interpretation of a situation. Volume is high (5 of 5): for a webshop or service provider with many small transactions, this quickly adds up to hundreds or thousands of lines per month, and that is exactly the kind of repetition where automated matching shows its value.
The cost of errors, however, is on the low side (2 of 5): a wrong match means an invoice being incorrectly written off as paid, or a bank transaction being attributed to the wrong customer. That kind of error stands out, but often only at the next closing or when a customer complains about receiving a reminder for an invoice that has already been paid. That is why full automation without oversight is not a good idea, even though the task itself is easy to structure.
Discretion (2) is limited but not zero: the vast majority of matches are unambiguous, a small share requires a decision (partial payment, currency difference, duplicate transaction). Compliance (3) plays a role because reconciliation is part of a reliable financial administration, with requirements for auditability that do not follow from a link itself but from the way a company must be able to account for its bookkeeping. Customer contact (5), physical work (5) and creativity (5) do not come into play here; this task takes place entirely within systems, not at a counter or in a conversation.
An example: a webshop receives hundreds of payments per week via a payment provider, each with a transaction reference that can be found in the accounting system. With an active link and clear matching rules (amount plus reference, or amount plus date plus customer name if the reference is missing), an RPA process can independently complete the largest part of these matches. What remains is a list of exceptions: deviating amounts, missing references, duplicate entries. That list is placed by an employee alongside the original documents and assessed with an approval or rejection, with a reason. That is the form that fits here: AI performs the matching, a human maintains oversight of what does not add up on its own.
At another company, this looks different. Without an active link with the payment provider, or with payments arriving via separate spreadsheets, the structure on which matching depends is missing, and the work shifts back to manual sorting out. At a company with many partial payments, discounts and credit notes, the degree of discretion increases and the share that can proceed without oversight decreases. The boundary conditions are therefore not decorative: without a link and without clear matching rules, this is a task that largely remains human work, with or without ambition to automate.
Reconciliation does not stand on its own. Companies that have already largely automated drawing up and sending sales invoices often find that matching on the other side becomes easier by itself, because references are more consistent. The same applies on the purchasing side: recording purchase invoices in the accounting system and monitoring due dates of purchase invoices for timely payment are similar matching tasks with comparable pros and cons. Companies that look at these tasks side by side often see a consistent picture: structured, high-volume work shifts to systems with oversight of exceptions, while work with a lot of discretion or customer contact, as seen in what AI can take over in hospitality or in answering and forwarding phone calls, moves along more slowly. Drawing up an outstanding items list for accounts receivable and payable also relies on the same matching logic and the same dependence on clean links.
This is not personnel advice and not a reason to revise job roles. Whether and how an organisation redistributes the freed-up hours is a choice for the employer itself; decisions affecting personnel are subject to their own legal requirements, separate from what is established here about the task. What is stated here is an assessment of the work, not of the people currently doing it.
Whether reconciliation at your own company can largely or only partly proceed automatically depends on the link you have with your payment provider and on how unambiguous your matching rules are. The free quickscan, twelve questions, no account required, gives an indication of what share of the hours in this profile can be taken over by AI today. The full work scan, which breaks down your company's work down to the level of tasks, is still under construction.
Vraag maar. Ik ken de kennisbank van deze site; wat ik niet weet, zeg ik erbij.
Answers come from this site’s knowledge base. Not tailored advice, and not a scan of your company.