First, find out whether the message was sent, refused or filed as junk
Start from one customer and one message. Ask for the order or booking reference, the address the customer gave and roughly when they expected the email, then work out which system sends that message. A business that sells online often has several: the shop sends order confirmations and receipts, a booking tool sends booking confirmations, the account or app system sends password resets, and some of these hand their mail to the marketing platform. Each one keeps a record of what it sent, and that record answers the first question.
The troubleshooting collection and the deliverability guide both separate the same three outcomes, and so does this page.
| What the sending system's record shows | What happened | Causes to check first |
|---|---|---|
| No message at all for that order, booking or account | It was never sent | The event no longer reaches the sender; the address suppressed for all mail |
| A message to an address that differs from the customer's real one | It was sent to the wrong place | An address mistyped at checkout, booking or registration |
| A bounce, with a reason naming an unknown or invalid address | It was refused because the address is wrong | An address mistyped at checkout, booking or registration |
| A bounce, with a reason naming authentication or policy | It was refused because the mail failed a check | A system that never authenticated; a DMARC policy tightened too soon |
| Delivered, but the customer cannot find it | It was accepted and filed somewhere else | Service mail sharing a domain or system with marketing mail |
For the last row the sending system can tell you no more, because acceptance is the end of what it sees. Ask the customer to check their junk and spam folders, and send the same message to a few addresses you monitor at the main mailbox providers. The deliverability guide describes these monitored addresses as a smoke test rather than a measurement, and that is the right weight to give them: they show whether order mail is landing in junk at all, not how often.
When several customers report the same thing at once, look for a change. A new checkout, a shop or booking platform update, a tightened DMARC policy or a move of marketing platform usually sits just before the first complaint, and it narrows the table below to one or two rows.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Order or booking mail refused with authentication or policy reasons, from one system only | Whether that system is in SPF and signs with DKIM on your own domain | A shop, booking or app system sending as the domain without its own authentication | Set up DKIM signing with your domain for that system |
| Refusals from several systems began on the day the DMARC policy changed | The aggregate reports for every legitimate source before and after the change | A DMARC policy tightened before every sending system passed and aligned | Step the policy back one stage until every legitimate source passes |
| Receipts and resets land in junk, often after a large promotional send | Whether service and marketing mail share a domain, subdomain or sending system | Order and account mail sharing a domain, subdomain or system with marketing mail | Separate the streams by subdomain and, where possible, by system |
| No message in the sender's record for recent orders or resets, since a site or platform change | The events the sending system now receives against the one the message listens for | The order or account event no longer reaching the system that sends the message | Reconnect the message to the event as it now arrives |
| Single customers, a bounce naming an unknown address, often a misspelt domain | The address on the order or account against the one the customer gives | An address mistyped at checkout, booking or registration | Correct the address with the customer and resend; prompt for common misspellings |
| No message sent to customers who unsubscribed from the newsletter or once bounced | The address's suppression status and its scope in each sending system | The address suppressed for all mail after a marketing opt-out or an old bounce | Limit marketing opt-outs to marketing; review the bounce behind the suppression |
Fix, verify and prevent each cause
A system that sends as the domain without authenticating
The shop, the booking tool and the account system were often set up by different people at different times, and each one puts the business's domain in the From address. The deliverability guide is explicit that every system which sends as the domain must be in SPF and signing with DKIM, or DMARC enforcement will eventually block it. The DMARC failure diagnosis, citing DMARC.org, explains why this goes unnoticed for so long: under a monitoring policy the mail still arrives, so nobody notices until enforcement starts refusing it. That page also covers the near miss, a system that signs with its own platform's domain rather than yours, which passes DKIM for a domain DMARC ignores.
Fix: in the sending system, set up DKIM signing with your own domain, or with a subdomain kept for that system's mail, publishing the records it provides.
Verify: the headers of a test order confirmation show DKIM passing for your domain, and the next aggregate reports show the source passing with an aligned domain.
Prevent: make authentication part of switching on any new tool that sends mail, and run the sending domain setup checklist whenever a shop, booking or account system is added or replaced.
A DMARC policy tightened before every sender passed
DMARC adds a policy and alignment on top of SPF and DKIM, and the policy is meant to tighten in stages. The deliverability guide sets the order: start at monitoring, read the aggregate reports until every legitimate source is authenticated and aligned, then move to quarantine and finally to reject. Google's email sender guidelines, as that guide summarises them, document authentication for all senders and, for bulk senders, SPF and DKIM, a DMARC policy and alignment of the From domain, which is why many businesses tighten the policy in a hurry, sometimes after a spoofing scare. Tightened early, the policy refuses or quarantines the order mail of any system that was still failing.
Fix: step the policy back one stage while the failing systems are authenticated, and record the decision, since the step back also relaxes protection against impostors.
Verify: the aggregate reports show every legitimate source passing and aligned, including the systems that send only occasionally.
Prevent: follow the DMARC page's staged approach, holding each stage long enough to see a full business cycle of senders, such as month-end invoicing or a seasonal booking system, before moving on.
Service mail sharing an identity with marketing mail
When receipts and campaigns leave from the same domain and the same system, they share one reputation and one queue. The promotional versus transactional comparison puts marketing and transactional emailAn email triggered by a person's own action or account, such as a receipt, password reset or shipping notice, rather than a marketing send. on separate subdomains, each with its own authenticationRecords and signatures, SPF, DKIM and DMARC, that let mailbox providers check that a message comes from the domain it claims. records, and on separate systems where possible, so that a campaign's complaints cannot drag receipts into junk and a campaign queue cannot delay a password reset. The diagnosis for emails going to spam names the same fault from the marketing side: receipts and password resets landing in spam alongside campaigns.
Fix: move service mail onto its own authenticated subdomain and, where the business can, onto a system that sends only service mail.
Verify: service mail holds its place in the monitored addresses through the next large promotional send.
Prevent: decide the stream for every new message type before it is built, and keep receipts, booking confirmations and resets free of promotion, so they stay service messages in content as well as in routing.
The event no longer reaching the sending system
A confirmation or a reset is triggered by an event: an order placed, a booking made, a reset requested. When a site redesign, a new checkout, a shop plugin or a move of platform sends that moment under a different name, or stops sending it, the message waits for an event that never comes, and the sender's record shows nothing at all. The diagnosis for a flow that is not sending covers this trigger fault in marketing flows, and the same check applies to service messages.
Fix: compare the events the sending system now receives with the one the message listens for, then reconnect it to the new event or restore the old one at the source.
Verify: a real test order, booking and password reset each produce a message in the sender's record and in a test inbox.
Prevent: add "which messages listen for this event?" to the release checklist for every change to checkout, booking or accounts, and test each service message after the release.
An address mistyped at checkout, booking or registration
Some missing confirmations were never going to arrive: the customer typed the address wrong. The bounce rate diagnosis explains that mistyped addresses bounce on the very first message, and a hard bounceAn email rejected permanently by the receiving server, usually because the address does not exist or the domain cannot receive mail. naming an unknown address or a misspelt domain is the sign. Occasionally the typo produces a different real address, and the confirmation goes to a stranger.
Fix: correct the address on the order or account with the customer, through support rather than by email, and resend the confirmation. Add a gentle prompt at checkout, booking and registration for well-known domain misspellings, which that diagnosis recommends.
Verify: the resent message is accepted, and bounces on order and account mail fall after the prompt goes live.
Prevent: read bounces on service mail regularly, not only on campaigns. The single versus double opt-in comparison treats the order confirmation as the proof that a checkout address works, so an address whose confirmation bounced should not join the marketing list either.
An address suppressed for all mail
A suppression listA list of addresses or contacts excluded from every send, such as unsubscribes, hard bounces and spam complaints. is supposed to have a scope. The promotional versus transactional comparison sets it out: an unsubscribe from marketing suppresses marketing only, and a hard bounce suppresses both. The suppression list entry adds that suppressions are synced between the email platform, the CRM and other sending tools, which is exactly where the scope gets lost. When service mail goes out through the marketing platform, or a sync copies marketing opt-outs into a suppression the shop or account system also honours, a customer who left the newsletter stops receiving their receipts. An old hard bounce can do the same long after the mailbox was fixed.
Fix: limit marketing opt-outs to marketing in every system, and for an address suppressed after a bounce, read the recorded reason before deciding whether it still applies.
Verify: a test address unsubscribed from marketing still receives a test order confirmation and a password reset.
Prevent: keep that test address in every sending system and check it after any change to syncs or suppression rules. The opposite fault, a person who opted out and still receives marketing, is diagnosed in unsubscribed contacts still receiving email.
Prevention: a sender register with one owner per system
Every cause above comes from a change made by someone who did not know an order or account email depended on it: a new tool switched on, a policy tightened, a checkout rebuilt, a sync rule edited. The DMARC failure diagnosis recommends a register of every system that sends as the domain or a subdomain; for service mail, extend each line with the messages that system sends, the event that triggers each one, and the suppressions it honours. The register only works if each line has an owner. The RACI matrix fits because transactional templates sit between teams: the comparison describes a handover between marketing and engineering whenever a transactional template changes, and a handover with no single accountable person is where messages fall through.
| Change | Accountable | Consulted | Check before it ships |
|---|---|---|---|
| A new system that sends as the domain | The domain owner | The team adding the system | SPF or DKIM on your domain, a line in the register |
| Tightening the DMARC policy | The domain owner | Every system owner in the register | Every legitimate source passing and aligned in the reports |
| A change to checkout, booking or accounts | The product or engineering owner | Whoever owns the service messages | Each service message still fires from a test |
| A new or edited service template | Whoever owns the service messages | Marketing, for design only | No promotion; the right stream and subdomain |
| A change to syncs or suppression rules | The marketing operations owner | The service message owner | A marketing opt-out still receives service mail |
Then watch the stream for what it is for. The comparison measures service mail on delivery, delay, bounce reasons and whether the customer completed the task, not on clicks, so put those into the monthly report as indicators (KPIA metric tied directly to an objective.) next to the deliverability measures. Treat support tickets as part of the monitoring: the DMARC page notes that the first sign of a broken sender is often a finance team asking why customers did not receive mail, and a support team asking about receipts is the same signal. When a fault has already left a known group of customers without their confirmations, the guide to proactive service messages covers telling them what happened and what they need to do, in a message that stays a service message.
A plant nursery's missing order emails (illustrative)
Mirravolt Plants
Mirravolt Plants, an invented online plant nursery, moves its DMARC policy from monitoring straight to reject after a spoofing scare. Within four days support logs 63 tickets from customers who never received an order confirmation, and 11 who never received a password reset. The shop's records show the confirmations as sent and bounced, with reasons citing DMARC: the shop platform sends as the nursery's domain but was never set up to sign with the nursery's DKIM, which went unnoticed for months under monitoring. The resets tell a second story. The account system hands them to the marketing platform on the main domain, which had suppressed all mail to 2,140 customers who unsubscribed from the newsletter. The nursery steps the policy back to quarantine, sets up DKIM for the shop on an orders subdomain, moves account mail off the marketing platform, limits marketing opt-outs to marketing, resends 410 confirmations for orders not yet delivered, and records an owner for each of its four sending systems. After six weeks of reports with every legitimate source passing and aligned, it moves to reject again, and order-related tickets in the following month fall to 3.
Common mistakes
- Treating every missing email as a spam problem. Read the sender's record first; a message never sent or refused needs a different fix from one filed as junk.
- Tightening DMARC to stop spoofing before every sender passes. The impostors stop, and so do the receipts.
- Authenticating only the marketing platform. The shop, booking and account systems send as the domain too.
- Letting a marketing unsubscribe stop service mail. The person who left the newsletter still needs the password reset.
- Adding offers to receipts to make them earn their keep. A receipt with promotions becomes marketing and loses the reasons it was kept apart.
- Testing only the marketing flows after a release. Place a real test order, booking and reset after every change to checkout or accounts.
