You click “Send invoice” in your accounting program. You have not opened the provider’s website or copied the figures into another form. Yet a moment later the program says that the provider has accepted the document. Something happened between the click and that message. Understanding those steps helps explain both the convenience of automatic sending and the questions to ask when it stops working.
A program is a set of instructions carried out by a computer. Your accounting program stores invoice details and shows buttons and forms so you can work with them. The provider also runs programs, on computers that receive work from customers over the internet. A computer doing that receiving work is often called a server. It is not necessarily one particular box; the service may use several computers. What matters here is the receiving role. 13
The click becomes a request
Suppose you have prepared an invoice for the bicycle shop. Your accounting program knows the recipient, the items, the amounts and the invoice number. When you click to send, it prepares a message using the provider’s agreed instructions. The message asks for an action, such as accepting this invoice for processing, and includes the information needed for that action.
That outgoing message is a request. It is not a person writing an email, and the receiving program is not reading the screen as you do. Both programs have been built to use an agreed arrangement: which actions can be requested, how the information must be organized, and what answers mean. This arrangement is an application programming interface, usually called an API. It lets one program ask another program for information or work. Documentation should explain the supported requests, answers, errors and access arrangements. 7
The accounting program sends its request to the provider’s configured address. The receiving software examines it. Does it recognize the connection? Is this connection allowed to submit an invoice for this company? Are the required details present in the form this service expects? These are different questions. A correctly arranged message may still come from a connection that has no permission to use it. Permission for an operation needs to be checked by the receiving service. 5
If the service accepts the request, it sends an answer back. That returning message is a response. For example, the answer might say that the document has been accepted for processing and give a reference for finding it later. Your accounting program reads that answer and turns it into a label you understand on screen. The label is the program’s presentation of the answer, not the answer of a person who has inspected the invoice.
Later, your program might send a different request: “What is the status of the document with this reference?” The provider returns a new response, and the display changes. Sending a document and asking about it are two distinct operations. These are illustrative possibilities; the supplier’s instructions must say what its actual service supports. 7
If a required recipient detail is missing, the response may instead explain why this submission was refused. Your program needs to show the reason so a person can correct the source invoice. Repeating the same request without correcting the missing detail does not solve that problem. A response confirming technical acceptance also does not establish that the invoice amount matches the agreed purchase. The service’s documentation must explain its actual errors and acceptance rules. 7
How does the provider recognize your program?
If anybody could submit instructions as your company, automatic sending would be unsafe. The connection therefore needs an appropriate way to establish which software client is asking. A software client means the program using the service, not necessarily the business customer as a person.
One possible arrangement gives the connection a value called an API key. The installer configures it in the accounting connection, which presents it as the provider requires. For a server-to-server connection that uses a confidential key, someone possessing it may be able to use the connection’s access, so it should not be pasted into emails or screenshots. Not every value called a key has the same role: a supplier must distinguish confidential credentials from identifiers intended to be public. The essential distinction here is that an API key can identify a software client; it must not be treated as proof of the identity of an individual employee. 2
For example, your program sends approved invoices overnight. The provider may record that your company’s accounting connection submitted the bicycle shop’s invoice. That tells you about the connection used. It does not, by itself, show whether Anna or Peter approved the document earlier. Personal approval needs its own record in the relevant business process. 2
Recognition also does not settle permission. Your program may be permitted to submit your invoices but not invite new administrators. An administrator is an account with authority to manage other accounts or important settings. There is no reason to give a sending connection that wider authority merely because it is recognized. A useful business instruction is: “This connection sends our invoices; it does not manage staff access.” 5
An answer can describe only one stage
Return to the bicycle shop’s invoice. An answer saying “accepted for processing” tells you about acceptance by this service. It should not be read as “the recipient has checked the invoice” or “the bicycle shop has paid”. Those events require later work and different evidence. Ask what each displayed status means and which event causes it to change.
If the invoice continues through Peppol, the receiving provider may then use a separate network service called an Access Point. That service exchanges messages with the recipient’s Access Point. The rules called the Peppol AS4 profile describe this exchange between Access Points, not automatically the customer API through which your accounting program first submitted the document. The same supplier can run both, but the two connections do different jobs. 10
A portal, where people sign in to view invoices, adds another way of working. An archive, if purchased, adds keeping records for later retrieval. Neither should be assumed to come with transmission alone. Knowing these stages helps you describe a problem: “Our program did not get an acceptance answer” is different from “The provider accepted the document but transport has not finished.”
Keep the invisible connection understandable
An automatic connection can carry on working after everybody forgets how it was set up. Then an accounting package changes, an employee leaves, or a provider changes its instructions. Someone needs to know which connection is used, which version of the interface it follows, and what data and permissions it uses. Keeping track of versions, integrations and documentation is part of API security guidance. 7
You need not learn to write these messages yourself. Ask the person responsible for your connection to walk through one ordinary invoice: what the click sends, what the provider answers, how the reference is retained, and how the final status is found. Use existing documentation or an expressly authorized demonstration with invented invoices, not a live experiment with customer documents. Once those steps are clear, the button on screen stops being a mystery without becoming your programming job.
Sources
13 — MDN: Overview of HTTP — client/server requests, responses and sessions
7 — OWASP API9:2023 Improper Inventory Management
5 — OWASP API5:2023 Broken Function Level Authorization
2 — OWASP API2:2023 Broken Authentication
10 — OpenPeppol AS4 Profile 2.0.3