About this case. It describes a common situation and a sensible approach, written without naming any customer. It reports no measured results, so set your own baseline and targets and measure your own outcomes.
The situation
Consider an online marketplace for handmade and regional goods. Thousands of independent sellers list products; buyers browse and order. The platform sends a lot of email on behalf of both sides: order confirmations to buyers, new-order alerts to sellers, messages between buyer and seller, review requests, dispute updates, payout notices and promotional digests.
Growth brought a deliverability problem that had two faces. First, the platform's bounce rate had crept up because buyers and sellers had signed up with mistyped or abandoned addresses, and the system kept sending to them. Second, seller-to-buyer messages, which sellers write themselves, sometimes contained content that recipients reported as spam, and promotional digests went to many users who had not visited in more than a year. Mailbox providers responded with slower delivery and more spam placement. Sellers complained that order alerts were late. Buyers said they never got their confirmations.
A marketplace is a special case
A marketplace sends mail from the platform's own domain but is partly written, triggered or targeted by third parties. That creates three risks that a single-brand sender does not face:
- User-generated content. Messages between members may contain links, requests to pay outside the platform or other content that looks like abuse.
- Uneven list quality. Addresses come from thousands of independent sign-ups, with different levels of care.
- Mixed expectations. A buyer expects an order confirmation within moments but may not welcome a promotion at all.
The platform's task was to protect the reputation of its domain from those risks, while keeping the essential messages fast.
Step 1: classify every message
The team began with an inventory of all the email types and sorted them into three classes, each with its own rules and its own sending identity.
| Class | Examples | Rules |
|---|---|---|
| Essential transactional | Order confirmation, new-order alert, payout notice, security notices | Own subdomain; never throttled by other classes; no promotional content |
| Member-to-member | Buyer and seller messages, dispute replies | Own subdomain; content checks; reply-to relay so addresses stay private; rate limits |
| Promotional and engagement | Digests, recommendations, review requests | Own subdomain; consent and engagement rules; one-click unsubscribe |
Step 2: stop the bounce bleed
Hard bounces, where an address does not exist, are the quickest way to damage a sender's name, because they suggest a list that nobody maintains. The team attacked the problem at three points:
- At sign-up. Validate the address format and check whether the domain can receive mail. A confirmation message with a link prevents typos from becoming permanent records.
- On every send. Process bounce events automatically: a hard bounce suppresses the address immediately, and repeated soft bounces suppress it after a set number of attempts.
- On the existing base. Run the full user list through an address-validation pass, suppress the invalid ones and flag risky ones for a re-confirmation email before they receive more mail.
Bounce and complaint events arrive through webhooks, so the application can update a user record the moment something happens, not at the end of a batch. We describe the monitoring side in our spam troubleshooting checklist.
Step 3: make member-to-member mail safe
Messages between buyers and sellers are the most unpredictable stream. The team added a set of controls, in the product more than in email settings:
- Relay addresses. Members never see each other's real email. The platform sends the message from its own authenticated domain with a reply-to relay, which also keeps contact details private and prevents off-platform payment requests from being sent directly.
- Content screening. Messages that contain certain patterns, such as external payment requests or suspicious links, are held for review or stripped of the link.
- Rate limits. New accounts can send only a small number of messages per hour, and the limit rises with reputation.
- A report button. A prominent "report this message" action gives recipients a better option than the spam button, and gives the platform a signal it controls.
Step 4: promotional mail on engagement
For digests and recommendations, the team shifted from "send to all users" to "send to users who show interest". They defined engagement by clicks and purchases, and not by opens, because machine-generated opens inflate that metric; we discuss the problem in our analysis of the open rate. Users with no clicks or orders over a long period were moved to a quarterly summary, then to a re-permission message, and then suppressed if they did not respond. Every promotional message carried a one-click unsubscribe, as the large mailbox providers now expect for bulk mail; see our guide to provider requirements.
Step 5: authentication and reputation monitoring
Each of the three subdomains has its own SPF, DKIM and DMARC records, aligned with the visible From domain. The team started DMARC at p=none with reporting and moved to enforcement after the reports showed no legitimate source failing. They added Postmaster Tools for each domain and set alerts for complaint rate and bounce rate on each class, so a problem in promotions could never again hide behind good numbers on order mail.
A week in the new world
Here is what an ordinary Wednesday looks like once the system is running. A buyer mistypes an address at sign-up; the confirmation link is never clicked and the account stays unverified, so no further mail is sent. A seller receives a new-order alert within seconds, on the essential subdomain, with no promotional content. A new seller tries to send a message with an external payment link; the link is removed and the message flagged. A digest goes out only to users with recent clicks; users who do not engage for a long period slide into the quarterly summary. In the monitoring dashboard, complaint and bounce rates are shown separately for the three classes, and one class turns amber on a Thursday morning. The team investigates the cause, finds a batch of old addresses added by a data import, suppresses them and moves on within the hour. Nothing that matters to buyers or sellers is affected.
How the team would measure success
| Measure | Why it matters |
|---|---|
| Hard bounce rate by class | Shows whether list hygiene is working |
| Complaint rate by class | Gmail guidance aims for under 0.1% and treats 0.3% as a ceiling |
| Order alert delivery time | Sellers depend on speed |
| Share of members who are engaged | Measures the quality of the promotional audience |
| Reports of abusive member messages | Indicates whether the safety controls are catching the right things |
We do not state numeric outcomes. A real marketplace should record its starting bounce and complaint rates and measure the change.
What the team learned
- Classify before you configure. Deciding what each message is for makes the technical choices obvious.
- Hygiene is cheaper than recovery. Blocking a bad address at sign-up costs nothing; repairing a reputation takes weeks.
- Safety is part of deliverability. Controls on member content protect the platform's name as much as DNS records do.
- Offer a better button. A visible "report" action reduces spam-button clicks.
Running a marketplace or platform? Tell us about your message types and we will help you map them to streams.
Related reading
Frequently asked questions
Does this case name a customer?
No. It describes a common situation and approach without naming any customer, and it reports no measured results.
How do I reduce bounces on a platform with many sign-ups?
Validate addresses at sign-up with a confirmation link, suppress hard bounces immediately through webhooks and clean the existing base before sending more.
How do I keep buyer and seller messages safe?
Relay messages through your authenticated domain, screen content, rate-limit new accounts and provide a report button.
Who should receive promotional digests?
Users who show recent engagement through clicks or purchases, with a sunset path and one-click unsubscribe for the rest.