How the system sends mail
The system's shared sending address sits in the vendor's domain. The SPF, DKIM and DMARC records of that domain remain unchanged, because only the display name of the sender and the Reply-To address change. Each client has an isolated instance with its own mail system, so one client's traffic does not mix with another's.
For the DNS administrator on the plant side, this means he does not add records to his own domain to make invitations work. His task is to make sure the incoming mail policy does not treat these messages as suspicious.
Practice 1: check the headers of test messages
After the installation goes live, send three test messages: invitation, password reset and MFA code. In the headers, check the SPF, DKIM and DMARC results and the From and Reply-To fields. Record the result in the mail acceptance protocol, because it will serve as evidence for any later reports.
Test messages after going live
Practice 2: understand why the filter sees spoofing
In one rollout, the client's mail flagged system messages as spoofing, because they came from an external domain and carried the client's company name as the display name. Current rules are different: the sender name shows who generated the message. The person appears under their own first name, surname and company, with a note that the message went through the system, and the system as System. First name and company are shortened to 34 characters.
Reply-To is set to the person's address only when the message has an actor and that actor has an address. System messages, such as password reset and MFA code, have no actor and no Reply-To. This way a reply to a code never reaches a random person.
Practice 3: set the trusted senders policy
If the plant's mail uses a list of trusted senders or anti-spoofing rules, add the system's sending domain in a way that complies with company policy. Do it before inviting the first group of users, not after waves of reports.
The footer of every message contains a sentence about the two domains with the address of the instance host and appears in 6 languages. It helps the user understand that the message comes from the plant's system, even though the vendor's domain sends it.
Practice 4: prepare the helpdesk
The helpdesk should have a short instruction for the question I did not get the code: check quarantine, check the spam folder, check the 10-minute validity of the code and generate a new one. It is also worth deciding who in IT can release a message from quarantine within minutes, because the code expires after 10 minutes anyway.
Checklist
Video
Frequently asked questions
Do I need to add records to the plant's domain?
Not in the described configuration. The sender is the vendor's domain, whose records remain unchanged.
Why did a message land in quarantine despite a correct DKIM?
The filter may also assess whether the display name matches the domain. Release the message and add the domain to trusted senders according to policy.
Do MFA codes have a Reply-To address?
No. System messages have no actor, so they have no Reply-To either.
Let's test mail delivery
We will send test messages and, together with the mail administrator, record the result in the mail acceptance protocol.
Let's talk