
Who Is the Vendor When the Buyer Is an Agent?
Your supplier master assumes a buyer chose a vendor and someone approved the relationship. Agent led buying inverts that order, and the master data is where it shows.
- Agent led purchasing can create a payable to a counterparty that never went through onboarding, because selection happened at the moment of need.
- Three way matching depends on a purchase order existing before the goods or service. Agent buying often produces the order and the consumption at once.
- Tax determination needs jurisdiction and registration data that an agent will not collect unless a rule forces it to.
- The practical defence is a closed counterparty list for anything an agent may transact with, refreshed deliberately rather than expanded automatically.
- ChatFin validates counterparties against the ERP supplier master before preparing anything, and routes anything unmatched to a person.
Vendor onboarding exists because someone decided, once, that this company was acceptable to trade with. Credit was checked, tax details were captured, bank details were verified, and a record was created that everything downstream depends on.
Agent led buying can arrive at a purchase without any of that happening, because selection is made at the moment of need from whatever is available. The commercial decision and the counterparty decision collapse into one event.
The result is not usually fraud. It is a payable to a counterparty your controls never assessed, and a set of downstream processes that silently assume the assessment happened.
The assumption chain nobody wrote down
Almost every payables control depends on a prior act of vendor approval. Duplicate checking keys on vendor identifiers. Bank detail change controls compare against a record that must already exist. Tax determination reads registration data captured at onboarding.
Remove the onboarding step and each of those controls does not fail loudly. It quietly has nothing to compare against, which is a much worse failure mode than an error message.
The chain is only visible when it breaks, which is why it is worth drawing before agent buying reaches production.
What breaks first
In practice the sequence of breakage is fairly predictable, and it starts with the controls that depend most on prior data.
These four are the ones that surface within the first month:
Closed list or open field
There are only two coherent designs. Everything in between produces a supplier master that grows faster than anyone reviews it.
The choice determines how much control work you carry afterwards.
Closed counterparty list
Agents may only transact with counterparties already onboarded and verified. Anything outside the list becomes a human request. Slower at the edge, and every existing control keeps working unchanged.
Open with post checks
Agents may transact with new counterparties, and controls run after the fact. Faster, and it requires monitoring, holdbacks and a cleanup process that somebody has to staff permanently.

The matching problem
Three way matching assumes an ordered sequence: a purchase order, then a receipt, then an invoice. It is a control precisely because the order exists before the commitment.
Agent purchasing frequently produces the commitment and the consumption simultaneously, especially for services and metered access. That does not remove the need for a match. It changes what you are matching against.
| Traditional | Agent purchase | The replacement control |
|---|---|---|
| Purchase order approved in advance | Authority granted in advance as a mandate | Match the transaction to the mandate scope, ceiling and counterparty list |
| Goods receipt confirms delivery | Service consumed as it is requested | Match against the usage record produced by the provider, retained as evidence |
| Invoice matched to both | Statement aggregating many events | Reconcile the statement to the usage record, then to the mandate, before payment |
The three way match survives. The three documents change identity.
Tax is the quiet one
Duplicate payments get found eventually. Tax determination errors accumulate quietly and surface during an audit two years later, by which time the amounts have compounded and the counterparties may not be contactable.
If an agent can create a payable, then either the counterparty carries verified tax data already or the transaction has to stop for a person. There is no safe middle setting where a rule guesses the treatment.
The control does not fail loudly. It quietly has nothing to compare against.
ChatFinPreparing the supplier master
This is master data work, which means it is unglamorous and it is the whole defence. Four steps, in order of value.
Flag counterparties as agent eligible
Add a single attribute to the supplier master. Agents transact only with records carrying it, and setting it is a deliberate act by a named person.
Verify tax and banking before the flag
Registration status, place of supply data and verified bank details are prerequisites for eligibility rather than things to chase afterwards.
Give unmatched counterparties a queue, not a workaround
When an agent meets a counterparty outside the list, it should stop and raise a request. If people can bypass that, the list is decorative.
Review the eligible list quarterly
Same discipline as a bank mandate review. Dormant records lose the flag, and changes to bank details reset it until reverified.

How ChatFin handles counterparty validation
ChatFin agents validate every counterparty against the supplier master in the ERP before preparing a transaction, using the master data already maintained in NetSuite, SAP Business One, JD Edwards, Acumatica, Dynamics 365 or Sage Intacct.
Where a document names a supplier that does not resolve to an existing record, the agent does not create one. It raises an exception with the extracted details attached so a person can decide whether onboarding is appropriate.
Bank details on an incoming document are compared to the master record rather than trusted, and any mismatch stops the item regardless of how routine the rest of the invoice looks.
AI Cost Estimator for the CFO Office
Four inputs, a first year cost range, a timeline, and three delivery models compared. No login required.
Frequently Asked Questions
Should agents ever be allowed to create supplier records?
As a default, no. Supplier creation is where onboarding controls, tax data capture and bank verification all attach. An agent can prepare a proposed record with the extracted details for a person to review and approve, which keeps the speed benefit without removing the control.
How do we handle genuinely one off purchases?
Route them through an existing mechanism designed for it, such as a corporate card or a one time payee process with its own controls, rather than expanding what agents may do. The exception should stay an exception rather than becoming the reason the list is open.
Does this apply to metered service consumption too?
Yes, and it is easier there because providers are few and stable. Onboard them properly once, mark them eligible, and the ongoing consumption flows against a verified counterparty with correct tax treatment.
What does ChatFin do when a supplier is not found?
It stops and raises an exception with the extracted supplier details, bank details and source document attached. It does not create the record, and it does not proceed on a partial match.
General information only. It is not legal, accounting or tax advice, and rules change. Check your own position with your advisers before acting on it.
Decide the counterparty list before an agent decides for you.
A single eligibility flag on the supplier master, set deliberately and reviewed quarterly, preserves duplicate detection, bank verification and tax determination at once. It is the cheapest control available here.
ChatFin validates against your existing supplier master and stops on anything it cannot resolve. If you want to see what that looks like against your own vendor data, we will run it with your records on screen.
Book a DemoSee the platform
Test agent counterparty validation against your own supplier master.
Bring a batch of supplier invoices, including a few awkward ones. We will show what resolves, what stops, and what evidence reaches the person who decides.