Comparison
Should you build or buy vendor payment automation?
Buy vendor payment automation, and check first if your ERP already does it. The real risk sits in who can change a vendor's bank details.
By the Foxnut team · Updated
Should you automate vendor payments yourself?
Buy vendor payment automation, and check first whether you already own it. Most requests to automate vendor payments describe the payment run, and the payment run is the one part of this process that is already finished.
We are Foxnut Studios, a senior-only practice in Bengaluru and Paris. We build and hand over systems that move money, and on this subject we mostly tell people not to build. There’s a link to talk to us at the end of this article. Let’s first work out what is actually left to automate.
Selecting which approved invoices are due, grouping them by vendor and bank account, applying the discount dates, producing a file the bank will accept and posting the result back to the ledger is a shipped feature of every serious accounting package and every ERP. For euro transfers inside the Union, even the file format is not a vendor’s choice. The law requires the ISO 20022 XML message format wherever a business that is not a microenterprise initiates credit transfers bundled together for transmission.
So the question is not whether the run can be automated. It is what is left once it has been, and what is left is not a payments problem at all. It is the vendor master record, who is allowed to change an account number, and a cross-border leg where the cost is in the currency conversion and not in the transfer.
Because we build these systems, we want to be plain about the transfer standard for anything that moves money before we agree to build one that does not need building.
What are your options to automate supplier payments?
You have four options to automate supplier payments, and they are not four levels of sophistication. They differ in where the payment decision is made and in who carries the loss when it is made wrongly. Those are separate questions from how much of the keying is automated.
| Axis | A person keys each payment at the bank | The accounting system’s own payment run | A dedicated payables or payments product | An integration built for this company |
|---|---|---|---|---|
| What it produces | Individual transfers, entered from an approved invoice list | A proposal, then a bank file, then a posted settlement against each invoice | The same file, plus vendor onboarding, bank-detail capture and multi-bank routing | A file assembled from the ledger by code the company owns |
| What decides who gets paid | The person, reading the aged payables list | Selection criteria set once: due date, discount date, method, currency, entity, payment blocks | The same criteria, restated in the product’s own vocabulary | Whatever the specification said, in code |
| Where the exceptions go | Back to the person, immediately and visibly | Onto a proposal the run reports and a person edits before release | Into the product’s own workflow, with its own audit record | Into whatever was built for them, which is the part specifications routinely omit |
| What it does about a wrong bank account | Nothing, beyond the person recognising the vendor | Nothing by default, though the change control over the vendor record is configurable | This is the substantive difference: supplier-maintained bank details with verification | Nothing that was not specified, and it is rarely specified |
| Who owns it after go-live | Accounts payable, and nobody else | The finance system administrator, using a documented licensed feature | The product vendor, with the criteria owned internally | The team that inherits the code, if it was handed over properly |
Read the fourth column twice. In the first three arrangements the payment run is somebody else’s maintained product and you own only your criteria. In the fourth you own a payment file generator, and a payment file generator is a piece of software whose failure mode is sending money to the wrong place.
When does a person at the bank portal still win?
A person at the bank portal wins where there are few enough vendors that the person recognises each one, where the payables list is short enough to read, and where there is one bank relationship.
Failure mode: not slowness, but a control that lives entirely in one person’s memory of who the suppliers are. A convincing request to change an account number has nothing to fail against. The upgrade worth making at that size is a written rule about how a bank detail changes, not a system.
When does the accounting system’s own payment run win?
It wins in almost every case above that size, and it is under-configured far more often than it is outgrown. The selection criteria in a payment proposal are your payment policy written down: pay by due date or by discount date, within this window, for this method and this currency, excluding blocked items, not exceeding this total, and check first that the vendor is not sitting in a debit balance. One vendor’s documentation goes further and offers a recurring schedule that runs the proposal automatically, which is all that most briefs ask a new system to do.
Failure mode: criteria set at implementation by somebody who has left, against a supplier base that has since changed, with payment blocks piling up on invoices nobody revisits.
When does a dedicated payables or payments product win?
It wins on the two things the ERP does not do well: onboarding suppliers, including collecting and verifying bank details from the supplier and not from an email, and routing across several banks, currencies and countries from one place. That is a real category with a real job. Notice that the job is supplier data and bank connectivity, not the payment run.
Failure mode: buying it for the run, then discovering that the onboarding module is the part that had to be implemented properly and the part nobody scoped.
When does an integration built for your company win?
It wins where the payment decision itself is genuinely unusual, and the honest examples are narrow. One is a marketplace or platform paying out to a large population of counterparties on rules that are the product and not an administrative function. The other is a treasury structure where which entity pays, in which currency, from which account is a decision with money in it and no packaged product models it.
Outside those, a build here is a bank file generator with a support burden, competing against something you are already licensed for.
What decides whether vendor payments are worth automating?
Six structural dimensions decide whether a process is worth automating. Vendor payments read unusually on four of them, and the four point the same way. The process is standardised almost to the point of being a utility, and the risk that remains sits at one cheap, non-technical control point.
| Dimension | What it reads on vendor payments | Source |
|---|---|---|
| Exception variance | Low, and unusually for this pillar the exceptions are enumerated in law rather than discovered in the field. Union law defines an R-transaction as a payment that cannot be properly executed or that results in exception processing, and lists the causes: a lack of funds, revocation, a wrong amount or a wrong date, a lack of mandate, or a wrong or closed account. The residual variance is name matching, which the same legislature describes as diacritics, transliterations between alphabets, and differences between habitual and formal names, and screening against restrictive measures, where transaction-by-transaction checking is recorded as flagging a very high number of transfers of which the large majority turn out on verification to involve nobody restricted. Both of those exception loads sit with the payment service provider, not with the payer. | Regulation (EU) No 260/2012, Article 2(25); Regulation (EU) 2024/886, recitals 21 and 25 |
| Volume | Not the invoice count, and this is the row most business cases get wrong. A payment file is one file whether it carries ten lines or ten thousand, and for euro transfers inside the Union its format is fixed by law rather than by bank, so a larger run is not a harder run. What scales is the number of banks, currencies, legal entities and jurisdictions, which is why the incumbent's parameters are shaped around exactly those axes: paying invoices held in other legal entities, proposing a separate payment per legal entity or one combined payment per vendor, a payment currency per run, and country-specific behaviour where a jurisdiction requires it. The payback condition follows as a condition rather than a threshold: automation earns its place where a run touches more than one bank, more than one currency or more than one entity, and not where the count of lines is simply large. | Regulation (EU) No 260/2012, Article 5(1)(d); Microsoft, Dynamics 365 Finance documentation |
| Cost of an error | High in both directions, and paid to different people. Paying the wrong party is the expensive one: business email compromise, defined by the United States federal crime reporting as a scam targeting businesses regularly performing wire transfer payments, drew 21,442 complaints and 2,770,151,146 dollars of reported losses in 2024, the second largest loss category of the year behind investment fraud. Paying late is priced by statute rather than by relationship: in European commercial transactions the creditor is entitled to interest without a reminder at the reference rate plus at least eight percentage points, plus a fixed sum of forty euro per debt as compensation for recovery costs, also without a reminder. | FBI IC3, 2024 Internet Crime Report; Directive 2011/7/EU, Articles 2(6), 3 and 6 |
| Reversibility | Effectively none, and the legal position is blunter than the operational instinct. Under the European payment services rules, a payment order executed in accordance with the unique identifier is deemed to have been executed correctly with regard to the payee that identifier specifies, and where the identifier supplied was itself incorrect the payment service provider is not liable and owes only reasonable efforts to recover the funds. The domestic United States position is a narrow window rather than a right: a reversing entry must reach the receiving institution within five banking days of the erroneous entry's settlement date and only for a closed list of reasons, and an improper reversal can itself be returned. Recovery once money has moved is an intervention rather than a correction: the federal recovery team invoked its kill chain on 3,020 complaints in 2024 covering 848.4 million dollars of attempted theft, and reports a 66 percent success rate. | Directive (EU) 2015/2366, Article 88; Nacha, ACH Network Rules: Reversals and Enforcement; FBI IC3, 2024 Internet Crime Report |
| Regulatory exposure | High, and almost all of it lands on the payment service provider rather than on the payer, with one clause that lands on an automated run specifically. From 9 October 2025 providers in euro-area member states must offer verification of the payee before a credit transfer can be authorised, and must screen their own users against targeted financial restrictive measures at least once every calendar day rather than transaction by transaction. The clause that matters to a payment file is the opt-out: users who are not consumers and who submit multiple payment orders as a package must be given the means to decline that verification, and the provider must inform them of the implications for liability and refund rights of doing so. An automated run is the exact thing that opt-out was written for. | Regulation (EU) 2024/886, Articles 5c and 5d |
| Vendor market maturity | The most mature process in this pillar, on three separate readings. The message format is mandated by legislation rather than chosen by a supplier. The rails are utilities: the United States automated clearing house alone carried 8.08 billion business-to-business payments worth 63.11 trillion dollars in 2025, out of 35.19 billion payments and 93 trillion dollars across the whole network. And the application layer is shipped and documented, down to selection by due date or cash discount date, filters by vendor, method, currency and legal entity, a vendor balance check before payment, and a recurring schedule that automates the proposal run itself. No public measure of concentration in vendor payment software specifically was retrieved, so none is quoted here. | Microsoft, Dynamics 365 Finance documentation; Nacha, ACH Network volume and value statistics, 2025; Regulation (EU) No 260/2012, Article 5(1)(d) |
Three of those rows agree, which is rare. The exception row says the hard cases are enumerated and mostly somebody else’s. The volume row says the work does not grow with the thing a business case usually counts. The maturity row says the whole capability is licensed, documented and legally standardised.
A process with that profile does not reward a build. Saying otherwise would be arguing against three sourced rows to reach a more comfortable conclusion.
The reversibility row redirects the effort instead of reducing it. A transfer executed on the account number supplied is deemed correct against whoever holds that account, so the whole consequence of this process is decided by the accuracy of one field in the vendor master and by who may change it.
That is not automation work. It is a change-control question with a documented answer. The same product documentation that describes the automated run also describes a configuration that stops bank account edits made during a proposal from flowing into the payment lines, so that payments cannot be redirected without oversight and the audit trail survives.
How do you automate vendor payments across countries?
You automate the decision about which entity pays, in which currency and by when, because that is what is left to decide when you pay vendors in other countries. It is not the part that the word automation usually points at.
Important note: the measurements below are as of March 2025.
The first fact reorders the whole question. For a business-to-business cross-border payment of twenty thousand dollars, the average total cost was 1.6 percent. Of that, 1.4 percentage points was the foreign exchange margin and 0.2 percentage points was the fee. The currency conversion is 87.1 percent of what a cross-border vendor payment costs.
Automating the run touches the 0.2. Deciding which entity pays, in which currency, from which account, and whether two flows in opposite directions can be netted before either crosses a border, touches the 1.4. A company that has automated its cross-border payment run and has not looked at its currency and routing policy has automated the cheap part of the bill.
The second fact is that the corporate cross-border experience is materially worse than the wholesale one, and worse than it was two years ago. Wholesale payments, meaning those of one hundred thousand dollars or more, ran above ninety percent credited within one business day, with more than half credited within the hour.
Business-to-business retail payments over the same period ran at 39.6 percent credited within one business day, down from 54 percent in 2023, and 2.2 percent within one hour. Ten percent of business-to-business corridors cost more than three percent. Only 43.2 percent of business-to-business payment services published both their cost and their speed to the sender, against 73.7 percent for person-to-person services.
So the operational problem in cross-border payables is not that the run is slow to prepare. The payment leaves, and its arrival date and final cost are frequently not knowable in advance. That is a forecasting and vendor-communication problem, not a processing one.
The third fact is that the standardisation that makes domestic vendor payments a solved problem stops at the edge of the currency area. For euro transfers inside the Union the file format is fixed by law, in euro-area member states the payee-verification service is now a legal requirement, and screening obligations sit on the provider.
Outside that perimeter, formats, name-verification services and cut-off conventions are per-country and per-bank. The migration of the international messaging network to ISO 20022 is still in progress, and speed improves once both ends of a payment have completed it.
So the specification is different from the domestic one. The cross-border work worth automating is the decision layer: which of the group’s entities and accounts should settle a given supplier, in which currency, on which rail, and by when, in order to hit the invoice’s due date given a settlement window that is not reliably one day. The mechanical submission is, again, something the bank or the payments product already does.
Important note: payee verification is the practical safety difference between a domestic and a foreign payment. Inside the euro area, a mismatch between the account number and the supplier’s name will now be surfaced before authorisation. Outside it, in most corridors, nothing will check that the name and the number belong together at all.
What do people get wrong when they automate payments?
The first error is counting invoices. A business case built on the number of payments per month measures the one variable that does not drive the cost, because the file is one file at any line count and its format is fixed. What does drive the cost is the number of banking relationships, currencies, legal entities and countries in scope. A case that does not name those four has not been costed.
The second error is treating fraud as a payments control. Business email compromise is a scam about a message, not about a transfer. The fraudulent instruction to change a bank account arrives by email, a person accepts it, and whatever system you run then executes it perfectly. A faster and more automated payment process makes a successful redirection settle sooner and with fewer people looking at it.
The control that works is upstream and administrative: who may amend a vendor bank record, and what evidence they need. The recovery numbers show why prevention is the only economical option. Even the federal recovery team, working with law enforcement and banks, freezes funds in about two cases in three when it is invoked at all.
The third error is reading the payee-verification opt-out as a performance setting. It is a liability setting. The Regulation deliberately allows a business submitting payment orders as a package to decline the name check, and just as deliberately requires the provider to explain what declining it does to liability and refund rights.
Any automated run large enough to trip the check is exactly the run for which somebody will propose turning it off. Whoever carries the loss should make that proposal, not whoever owns the schedule.
The fourth error is treating late payment as a courtesy. In European commercial transactions the interest runs without a reminder, and the fixed forty-euro compensation is payable per debt without a reminder. Business-to-business payment terms are limited to thirty days by default and sixty as a general ceiling, unless a longer period is expressly agreed and not grossly unfair.
A payment calendar that quietly stretches terms is not a working-capital strategy in that jurisdiction. It is an accruing liability that nobody is invoicing yet, and it is an argument for a correctly configured run, not against one.
There is one prior question this page does not answer. Whether the invoice should have been approved for payment at all, and whether the amount matches what was ordered and received, is a different control with a different failure mode. This page begins after that decision has been made. A payment system built on approvals nobody trusts will simply pay disputed amounts on time.
What is our verdict on vendor payment automation?
Buy it, and check first whether it is already bought. On the evidence in the profile above, vendor payment automation is a configuration project and not a build. The burden of proof sits on anyone proposing otherwise.
The exceptions are a closed list defined in legislation and largely handled by the payment service provider. The effort does not grow with the volume a business case usually counts. The message format is mandated. The application layer, down to a schedule that runs the proposal without anyone starting it, is documented in the product most companies in this position already license.
No plausible reading of those rows produces a build. Manufacturing one to look even-handed would be a worse failure than a recommendation that is commercially inconvenient.
What survives is smaller than a project and more important than the run. The reversibility row says a payment executed against the account number the system supplied is final as against the person who actually holds that account. So all of the consequence in this process is decided by the accuracy of one field and by the rule governing who may change it.
The regulatory row says the one control that catches a wrong account in the euro area can be lawfully switched off for exactly the batch files an automated run produces, and that switching it off moves liability.
Software fixes neither of those. A written change-control rule for vendor bank details fixes them, along with a separation between the person who edits that record and the person who releases the run, and a decision, made by name, about whether the payee check stays on.
The cross-border leg is the only place a build has an argument, and the argument is about currency and routing, not payments. Where 87.1 percent of the cost of a cross-border vendor payment is the foreign exchange margin, and where fewer than half of such payments arrive within a business day, the thing worth automating is which entity settles which supplier, in which currency and when, so that a due date is met given a settlement window nobody can promise. That is a decision system, and it sits next to a payment run, not inside one.
Your useful first move costs nothing. Open the payment proposal in the system you already have and read its selection criteria against the payment policy as it is actually practised. Then take the last twelve months of vendor bank-detail changes and try to establish, for each one, who requested it, who approved it and what evidence was checked.
The first exercise usually finds that the automation was licensed years ago and never configured. The second usually finds the actual exposure, and it is never in the payment run.
Foxnut Studios is useful here because we would sooner have you configure what you own than sell you a build. The people who sign the proposal do the work, there is one tier, and we designed, made and launched our own products first. Where a system really is worth building, we build it and hand it over running.
The first step is a free expert audit: a senior practitioner reviews your setup, and there is no RFP and no pitch deck. You can read about the five disciplines at Foxnut Consulting first. If you have not run the second exercise on your own vendor-master changes recently, do it this week, then tell us what you find and we will help you read it.
The part most pages leave out
When not to choose Foxnut Studios
Situations where another option is the better call, and where we say so in the first conversation rather than the fourth.
- Your accounting package or ERP already has a payment proposal and a payment run, and nobody has configured the selection criteria, the payment blocks or the bank file export. The brief is to build something that assembles a payment file. That software, and we say so. One major vendor's own product documentation describes selecting invoices by due date or cash discount date, filtering by vendor range, method of payment, currency and legal entity, checking that a vendor is not in a debit balance before paying, previewing or suppressing the proposal, and defining a recurring schedule that runs the proposal automatically. That is the whole brief, already shipped and supported. Paying to rebuild it is the most common thing found underneath a vendor payment automation request, and configuring the licensed one takes a week, not a project.
- The design removes the human release step from the payment run, on the reasoning that a reviewed proposal is the bottleneck and the file is correct by the time it gets there. A human control point, or a simpler deterministic rule. The imbalance is in the law, not in the software. Under the European payment services rules, a payment order executed in accordance with the unique identifier is deemed to have been executed correctly with regard to the payee that identifier names. Where the identifier supplied was wrong, the payment service provider is not liable and owes only reasonable efforts to recover. Sending is the last moment at which a payment is still a decision. Automating the assembly and keeping a person on the release costs a signature. Automating the release costs whatever the file contained.
- Your payments are euro-area credit transfers submitted as a package, and the design opts the file out of the payee-verification service so that a run of several thousand lines completes without interruption. The incumbent certified vendor, whose attestation is the product. The opt-out is lawful and deliberate: the Regulation gives users who are not consumers the means to opt out when they submit multiple payment orders as a package. But the same article requires the provider to state the implications for liability and refund rights of ignoring or declining the check, and the refund obligation it creates runs to a provider that failed to perform the check, not to a payer who declined it. A bank or scheme participant that runs the verification, screens its users daily against restrictive measures and carries the scheme membership is selling exactly the attestation the opt-out gives away.
- Your case rests on the time the payment run takes, and one person currently clears the month's run in a morning from a proposal the accounting system already produces. The person already doing it, unchanged. A payment file is one file whether it carries ten lines or ten thousand, so the effort does not grow with the invoice count. It grows with the number of banks, currencies, legal entities and exceptions. Where those are one, one, one and few, the whole saving is the morning. The useful automation at that size sits upstream of the run, in the approval that decides what enters it, and it is usually already licensed.
- Your vendor master's bank details are maintained from emailed remittance forms and PDF letterheads, with no record of who changed an account number, when, or on whose authority. Your organisation, doing the documentation first. A payment run is only ever as correct as the account numbers it reads, and automating it gets a wrong account paid faster and on a schedule. United States federal crime reporting defines business email compromise by precisely this population, businesses regularly performing wire transfer payments. The same ERP documentation that describes the automated run also describes a configuration that stops bank account edits made during a proposal from reaching the payment lines, so that payments cannot be redirected without oversight. The control is already in the box. The record of who may change an account number is not.
Sources
- Regulation (EU) 2024/886 amending Regulations (EU) No 260/2012 and (EU) 2021/1230 as regards instant credit transfers in euro, Official Journal 19 March 2024 - Article 5c on verification of the payee, including paragraph 6 on the opt-out available to non-consumer users submitting multiple payment orders as a package and paragraph 8 on liability and refund, Article 5d on at least daily screening of payment service users against targeted financial restrictive measures, and recitals 21, 24 and 25. Read as the Official Journal full text on EUR-Lex Retrieved
- Directive (EU) 2015/2366 on payment services in the internal market (PSD2) - Article 88 on incorrect unique identifiers, paragraphs 1 to 3, and Article 89 on payment service provider liability for non-execution or defective execution. Read as the consolidated directive text on EUR-Lex, CELEX 32015L2366 Retrieved
- Regulation (EU) No 260/2012 establishing technical and business requirements for credit transfers and direct debits in euro - Article 2(25) defining an R-transaction, Article 5(1)(d) requiring the ISO 20022 XML message format where a non-consumer, non-microenterprise user initiates credit transfers bundled together for transmission, and recital 14 on mandatory standards. Read as the full text on EUR-Lex, CELEX 32012R0260 Retrieved
- Directive 2011/7/EU on combating late payment in commercial transactions - Article 2(6) defining statutory interest as the reference rate plus at least eight percentage points, Article 3(3) on the thirty-day period and Article 3(5) on the sixty-day ceiling for business-to-business terms, and Article 6(1) on the fixed sum of EUR 40 payable without a reminder. Read as the full text on EUR-Lex, CELEX 32011L0007 Retrieved
- Federal Bureau of Investigation, Internet Crime Complaint Center, '2024 Internet Crime Report' - the business email compromise complaint and loss figures, the definition of business email compromise in Appendix B, and the Recovery Asset Team's 2024 Financial Fraud Kill Chain results. Read as text extracted from the Bureau-hosted PDF Retrieved
- Financial Stability Board, 'G20 Roadmap for Enhancing Cross-border Payments: Consolidated progress report for 2025', published 9 October 2025 - section 3.1 on wholesale speed, and Tables 7, 8, 9, 12 and 15 on the cost, foreign exchange component, corridor spread, speed and transparency of business-to-business retail cross-border payments. Read as text extracted from the FSB-hosted PDF Retrieved
- Nacha, 'ACH Network Rules: Reversals and Enforcement' - the five-banking-day window within which a Reversal must be made available to the receiving institution, the closed list of permissible reasons for a Reversing Entry, the formatting requirements, and the R11 and R17 return codes for an improper Reversal Retrieved
- Nacha, 'ACH Network Volume and Value Statistics' - full-year 2025 network volume and value, and the business-to-business payment count and value for 2025 Retrieved
- Microsoft, 'Create vendor payments by using a payment proposal', Dynamics 365 Finance product documentation, article dated 19 November 2025 - the payment proposal selection criteria, the advanced parameters including the vendor balance check and the multi-legal-entity options, the vendor payment proposal automation scheduling feature, and the note on vendor bank account changes at proposal time Retrieved
Foxnut Studios takes on work like this from Bengaluru and Paris. If you want to see how we do it before you talk to anyone, here is what an AI engagement covers and what you keep.