# The Hard-Bounce Read: Re-Address, Follow the Mover, Fix Sender, or Retire

*Grade any hard-bounced record from its full server reply to one of four verdicts, with a named next action and a CRM write-back two operators would agree on.*

- Canonical URL: https://www.refolk.ai/guides/hard-bounce-read-four-verdicts
- Pillar: Process, data, and compliance
- Format: Framework
- Published: 2026-11-07
- Last reviewed: 2026-11-07
- Reading time: 21 min
- Keywords: what to do with hard bounced contacts, does 550 5.1.1 mean the person left the company, 550 5.4.1 recipient address rejected access denied cold email, hard bounce vs policy bounce suppress or not, bounced email job change find new contact, bounce recovery process CRM

## Key takeaways

- A hard-bounce code alone cannot separate a departure from a domain move: 550 5.4.1 carries both "Recipient address rejected" for one address and "Relay Access Denied" for a whole domain.
- Any 5.7.x authentication or policy reply, such as Outlook.com's 550 5.7.515, is a sender fault and goes to the deliverability owner, never to the suppression list.
- Microsoft 365 keeps a deleted mailbox for 30 days, or for as long as a hold lasts, so the date of a hard bounce can trail the person's real departure by weeks.
- Follow the mover only on logged public evidence with a URL and a date; an address bouncing is never enough to auto-swap a lookalike contact into the sequence.
- Google publishes no bounce-rate limit, only a spam-complaint ceiling of 0.30%, so resending to dead addresses costs reputation rather than tripping a fixed number.
- Refolk's index holds 342 US and 28 UK Revenue Operations titles with at most 2 at any top employer, so the bounce read usually rests on one person and must be written down to be repeatable.

A hard bounce tells you a message failed permanently. It does not tell you whether the person left, the address was wrong, the company moved its mail, or your own sending setup got rejected. This guide is for the recruiting ops or revenue ops owner holding this week's bounced prospect and candidate records. It scores five dimensions from the full server reply and lands each record on one of four verdicts, each with a named next action and a CRM write-back that a second operator would reproduce.

## What are the four hard-bounce verdicts?

Every bounced record ends on exactly one of four verdicts: re-address, follow the mover, fix the sender, or retire. Each verdict names who acts next and what gets written to the CRM, so nothing gets suppressed, deleted, or re-sequenced just because it was marked "bounced".

| Verdict | What it means | Next action |
|---|---|---|
| Re-address | Person likely still there; address or domain is the fault | Correct the address, check the domain, or hold for a dated recheck |
| Follow the mover | Logged public evidence shows they left | Route to the departure process with URL and date |
| Fix the sender | The reply is about your sending setup | Quarantine; deliverability owner fixes; never suppress |
| Retire | No live path worth pursuing | Minimal suppression entry on the address; stop |

Three terms used throughout:

- **DSN (delivery status notification):** the bounce message the receiving system sends back, carrying the codes and the reply text.
- **Enhanced status code:** the three-part code defined in RFC 3463, such as 5.1.1, that follows the basic three-digit reply such as 550.
- **Reporting host:** the server that generated the rejection, for example a `*.mail.protection.outlook.com` host for Microsoft 365 or a `gsmtp` host for Google.

Public answers to "what to do with hard bounced contacts" split into two camps. Code glossaries stop at "5xx means delete". One widely copied glossary says 550 5.1.1 means suppress immediately and 550 5.2.1 means suppress and log separately. Automation templates go the other way. They assume every bounce is a job change and swap a lookalike contact at the same company into the sequence. Neither camp reads the reply text, the receiving host, the pattern across the domain, and public evidence about the person together. Reading all four together is the whole method.

## Which bounce codes point at the person and which at the sender?

Codes 5.1.1, 5.2.1 and 5.4.1 point at the recipient side. 5.7.x authentication and policy replies point at you. The wording matters as much as the code, because the same 5.4.1 can mean one dead address or a whole domain that no longer accepts mail.

| Code | Reply wording | Points at | First move |
|---|---|---|---|
| 550 5.1.1 | Account "does not exist" (Gmail) | The address | Check provenance |
| 550 5.2.1 | Account "is inactive" (Gmail) | The person; account exists | Check host and evidence |
| 550 5.4.1 | "Recipient address rejected: Access denied" | One address in the tenant | Check the rest of the domain |
| 550 5.4.1 | "Relay Access Denied" | The whole domain | Domain-move check |
| 550 5.7.515 | Sending domain fails required authentication level | Your sending domain | Fix the sender |
| 550 5.7.1 | "likely unsolicited mail" (Gmail) | Your reputation or content | Fix the sender |

What each family proves, and how it misleads:

- **5.1.1 and 5.2.1.** Google gives 550 5.1.1 the wording "The email account that you tried to reach does not exist" and 550 5.2.1 "The email account that you tried to reach is inactive." Both point at the person's account, but "inactive" means the account still exists. On a Workspace domain that can be a suspension, which can be reversed.
- **5.4.1.** Microsoft says 5.4.1 "is generated when directory-based edge blocking in Microsoft Exchange Online blocks an incoming email message because the recipient's email address is invalid." So the failure is on the recipient side. The trap is that Microsoft's own fix list also covers hybrid on-premises mailboxes, public folders and dynamic distribution groups. A hybrid Exchange environment that has not synced properly can reject a live employee's real mailbox as invalid.
- **5.7.515.** Outlook.com requires domains sending more than 5,000 messages a day to Outlook.com, Hotmail.com or Live.com to pass SPF, DKIM and DMARC. Failures come back as "550; 5.7.515 Access denied, sending domain [SendingDomain] does not meet the required authentication level." The reply names your domain, not the recipient. A practitioner note on this code says the usual culprit is one sending source that was never configured for custom-domain DKIM.

#### Where a hard bounce can originate

1. **Your sending setup** - SPF, DKIM, DMARC on the From domain; 5.7.515 and 5.7.1 live here
2. **Receiving domain** - MX and relay; "Relay Access Denied" means the domain refuses mail
3. **Recipient directory** - Tenant edge blocking; 5.4.1 "Recipient address rejected", including hybrid sync gaps
4. **Mailbox account** - 5.1.1 does not exist, 5.2.1 inactive or suspended
5. **The person** - Still employed, departed, or never at that address

*Read the reply from the outermost layer inward, because a failure at an outer layer makes every record behind it look dead.*

> **Rule:** Never suppress a 5.7.x reply
>
> Any authentication or policy rejection goes to the deliverability owner as fix-the-sender, even when the DSN lists a single recipient. Suppressing it throws away a good record and hides the fault that caused the bounce.

Why is the sender gate so strict? Google's sender guidelines contain no numeric bounce requirement. Google's only published threshold is for spam complaints: stay below 0.10% and never reach 0.30% or higher. Klaviyo calls keeping bounces below 1.0% best practice, but that is a vendor norm, not a provider limit. Nobody has published how many resends to a dead address it takes to get blocked. Dead addresses cost you reputation gradually, not by tripping a fixed number. Outlook does reject at the domain level, though, so one misconfigured sending tool can make an entire list look dead in a single batch.

## How do you score the five dimensions?

Give each dimension one letter grade, then apply the verdict rules in order and stop at the first match. The grades are categories, not points, on purpose. If the grades were added up, a high-value record could outvote a sender-side reply, and that is exactly the mistake this read exists to prevent.

### 1. Reply class

- **S (sender-side):** any 5.7.x authentication or policy reply, or any reply that names your sending domain.
- **A (address-level):** 5.1.1, or 5.4.1 with "Recipient address rejected".
- **I (inactive):** 5.2.1.
- **R (relay):** "Relay Access Denied" or similar wording that refuses the whole domain.
- **U (unreadable):** no raw reply stored. You cannot grade U; go back to capture.

What it proves: which layer rejected the message. How it lies: some vendor blogs describe 5.4.1 as an access policy against the sender, which contradicts Microsoft. Grade from the wording and from Microsoft's own documentation, not from a glossary.

### 2. Domain-wide pattern

- **1 (single):** only this address bounced at the domain in this batch.
- **M (multi, others delivering):** two or more distinct addresses bounced, but other addresses at the domain still deliver.
- **W (whole domain):** every address you sent to there failed, or any reply carried relay wording.

Microsoft's triage step is to "determine whether the issue affects only one recipient or everyone in the recipient's domain." There is no published threshold for when a cluster means a migration. My convention is two or more distinct addresses at one domain in the same batch. What it proves: whether the problem is the person or the company's mail. How it lies: a few real departures can land in the same week by coincidence. To check for a move, compare MX records, follow the website redirect, and look for a company name change in the registry. That is a desk inference. I found no source documenting it, so treat it as corroboration, not proof.

### 3. Departure evidence

- **L (logged):** a public profile or company page showing a new employer or an end date, saved with URL and date.
- **W (weak):** ambiguous, stale, or unconfirmed.
- **N (none found):** looked, nothing there.

What it proves: that the person left, which is what follow-the-mover requires. How it lies: profiles go stale, two people can share a name, and a title change inside the same company can look like an exit. Checking public profiles one site at a time is what makes this step slow. [Refolk](/) cuts that down to one plain-English ask that searches LinkedIn, GitHub and the open web together and returns dated profiles you can log.

Do not use the bounce date as the departure date. Mail platforms keep a deprovisioned mailbox around for a while, so the bounce can trail the real exit by weeks.

| Platform state | Recoverable window | Source |
|---|---|---|
| M365 soft-deleted | 30 days | learn.microsoft.com |
| M365 on hold (inactive) | Duration of hold | learn.microsoft.com |
| M365 hold removed | 30 days (was 183) | practical365.com |
| Workspace deleted | 20 days, then address removed | toriihq.com citing Google |
| Workspace suspended | Data kept, mail bounces | backupvault.co.uk |

**30 days - How long Microsoft 365 keeps a deleted user's mailbox as soft-deleted**

With a hold in place it becomes an inactive mailbox and stays until the hold is removed, so public departure evidence often predates the bounce.

Two limits apply here. Nobody has published which SMTP reply a sender actually sees at each of these stages. Whether a soft-deleted mailbox still accepts mail is also unverified. The table tells you the timing can be off. It does not tell you which code goes with which stage.

> The bounce date is when the mailbox died, not when the person left.

### 4. Address provenance

- **P (person-supplied):** they gave it to you, on an application or in a reply.
- **V (verified):** checked by a verification step before sending.
- **G (pattern-guessed):** built from a firstname.lastname style pattern.
- **X (unknown):** treat as G.

What it proves: whether the address ever existed. A 5.1.1 on a guessed address is most likely a bad guess, not a departure. How it lies: "verified" only describes the moment it was checked. A verified address that bounces months later carries little information about the guess and more about the person.

### 5. Record value

- **H:** active candidate in process, or contact on an open opportunity.
- **M:** engaged before but nothing currently open.
- **C:** cold.

What it proves: how much effort the record deserves, which decides whether a sourcer spends the 5 to 15 minutes on evidence. How it lies: owners grade their own records generously. Hold H to the strict definition above.

### Verdict rules, first match wins

1. Reply class S: **Fix the sender.** Quarantine and do not suppress.
2. Reply class R, or domain pattern W: **Re-address (domain check).** Hold every record at the domain. Retire none of them individually until the check is done.
3. Departure evidence L with value H or M: **Follow the mover.**
4. Departure evidence L with value C: **Retire**, with the new employer noted.
5. Provenance G or X: **Re-address (correct the address)** for H or M. **Retire** for C. No mover search.
6. Everything else (class A or I, provenance P or V, domain 1 or M with others delivering, evidence N or W): **Re-address (hold)** with a dated recheck. I use two weeks. A second identical bounce with still no evidence means retire, except for H records, which get one attempt through another channel first.

#### After the sender and domain gates

Horizontal axis runs from No departure evidence to Logged evidence with URL and date. Vertical axis runs from Cold record to Active candidate or open opportunity.

| Quadrant | What it means |
| --- | --- |
| Hold and verify | Re-address or dated recheck; sourcer spends 5 to 15 minutes looking |
| Follow the mover | Route to the departure process with URL and date |
| Retire on second bounce | Minimal suppression on the address; no search |
| Retire with mover note | Log the new employer; no re-sequence |

*Once sender-side and domain-wide bounces are routed out, departure evidence and record value decide between following, holding and retiring.*

**Hard-bounce scoring sheet**

```
Record ID:
Raw reply (verbatim):
Reporting host:
Bounce date:

1. Reply class      [S / A / I / R / U]
2. Domain pattern   [1 / M / W]   other addresses at domain delivering? [Y / N]
3. Departure evid.  [L / W / N]   URL:            date seen:
4. Provenance       [P / V / G / X]
5. Record value     [H / M / C]

Rules (first match wins):
R1  class S                      -> FIX THE SENDER (never suppress)
R2  class R or pattern W         -> RE-ADDRESS: domain check, hold all at domain
R3  evidence L, value H or M     -> FOLLOW THE MOVER
R4  evidence L, value C          -> RETIRE, note new employer
R5  provenance G or X            -> RE-ADDRESS (H/M) or RETIRE (C), no mover search
R6  otherwise                    -> RE-ADDRESS: hold, recheck on (date)

Verdict:
Rule fired:
Graded by:
```

*Paste one block per record; fill every field before applying the rules, and stop at the first rule that matches.*

## The bounce read, step by step

Run the read as a weekly batch in seven steps: capture, sender gate, domain grouping, provenance, evidence, value, write-back. The order matters. Code glossaries jump straight from the code to suppression. Microsoft's own NDR guidance puts the spelling and domain-wide checks before any action, and this procedure follows Microsoft.

#### Weekly hard-bounce read

1. **Capture the full server reply** - Pull status code, enhanced code, reply text, reporting host and Final-Recipient for every bounce. Done when every record stores the raw reply, not just "hard bounce".
2. **Run the sender-side gate** - Route every 5.7.x authentication or policy reply, including 5.7.515, 5.7.26 and Gmail's 5.7.1 "unsolicited", to the deliverability owner. Done when those records are quarantined from the person workflow and none is suppressed.
3. **Group the batch by domain** - Ask whether each failure affects one recipient or everyone at the domain. Done when each domain is labelled single-record or multi-record, and multi-record or relay-denied domains are held for a domain check.
4. **Record address provenance** - Write down whether the address was person-supplied, verified, or pattern-guessed, treating unknown as guessed. Done when provenance is a field on the record.
5. **Log public departure evidence** - For high-value records, check public profiles and company pages, 5 to 15 minutes each. Done when evidence is logged with URL and date, or marked "none found".
6. **Score record value** - Grade H for an active candidate or open opportunity, M for previously engaged, C for cold. Done when the owner has set the grade against those definitions.
7. **Apply the verdict and write it back** - Apply the rules in order and stop at the first match. Done when code, reply text, bounce date, grades and verdict are written to the CRM.

The per-step times are my desk estimates and unverified, except where noted. Capture takes minutes per batch if your sending tool exposes the raw DSN. If it only exports a "hard bounce" flag, fixing that export is the first job, because without the reply text every later grade is a guess.

## How do you follow a mover from a bounced email?

Follow the mover only when departure evidence is logged with a URL and date, and then split the work into two separate tasks: find the person at their new employer, and find a successor at the old account. Never let automation treat a bounce as a job change and swap in a lookalike contact.

The two tasks have different evidence and different owners:

- **The person at the new employer.** This creates a new record at a new company. It needs its own address, its own provenance grade and its own check against your suppression rules. The old bounce does not carry over as permission.
- **The successor at the old account.** This is a separate search with its own evidence. Someone sharing a title at the old company is not proof they inherited the relationship.

Finding where a named person went is usually the slow part, because the evidence is spread across profile sites, company pages and code hosts. With Refolk you ask in plain English and get back people with their previous role and current employer, each tied to a public profile you can log as the evidence for rule R3.

Ask me this: `People who were in revenue operations at 6sense and have changed employer in the last six months` - [run the search](https://www.refolk.ai/start?q=People%20who%20were%20in%20revenue%20operations%20at%206sense%20and%20have%20changed%20employer%20in%20the%20last%20six%20months).

*Returns named people with their earlier 6sense revenue operations role and their current employer, each with a dated public profile you can log as departure evidence.*

If your buyers are operations people, the follow-the-mover handoff tends to point at a single person rather than a team. Refolk's index shows how thinly these roles are spread.

| Title | Country | Profiles | Share in top metro (n=25 sample) |
|---|---|---|---|
| Revenue Operations | US | 342 | 24% SF area |
| Revenue Operations | UK | 28 | 60% London |
| Recruiting Operations | US | 101 | 44% SF area |

These are exact-title matches, which undercount, so treat every figure as a floor. The metro shares come from 25-profile samples.

**12.2x - More Revenue Operations titles in the US than the UK in Refolk's index**

342 against 28, and the top-10 employer lists in both countries show at most 2 people per company.

Two consequences follow from the table:

| Comparison | Ratio |
|---|---|
| RevOps US vs UK | 12.2x (derived) |
| RevOps vs Recruiting Ops, US | 3.4:1 (derived) |
| Max people per employer, top-10 lists | 1 to 2 |

First, when a UK revenue operations contact moves, the new employer usually has one or two people in the role, so the follow-up has a single named owner. Second, on your own side, the person running this read is often the only one doing it. That is why the scoring sheet has to be written down. If the read lives in one operator's head, nobody else can check it or cover for them.

## How does the hard-bounce read go wrong?

The read fails in eight predictable ways, and almost all of them come from trusting the code over the wording, the person over the domain, or automation over evidence. Each failure below gives the wrong verdict it produces and the check that catches it.

1. **5.4.1 misread as a policy block.** Some vendor blogs describe 5.4.1 as an access policy against the sender, which contradicts Microsoft. False verdict: fix the sender, on an address that is actually dead. Check: Microsoft's own documentation, and whether other addresses at the same tenant still deliver.
2. **5.4.1 caused by the recipient's own sync gap.** The person still works there, but a hybrid sync gap rejects their mailbox. False verdict: departed, or retire. Check: no departure evidence and the rest of the domain delivering means hold (rule R6), not retire.
3. **5.2.1 treated like 5.1.1.** "Inactive" means the account exists. On a Workspace domain it can be a suspension that may be reversed. False verdict: immediate retire. Check: the reporting host, and departure evidence.
4. **5.7.515 suppressed as a hard bounce.** The DSN lists one person, but every Outlook.com recipient from that From domain is failing. False verdict: retire, repeated across the list. Check: does the reply name your sending domain?
5. **A cluster read as individual departures.** Five bounces at one domain in one week get retired one at a time when the company has actually migrated. False verdict: five retires. Check: domain grouping in step 3, then compare MX records and the website redirect. That comparison is a desk method and unverified.
6. **A pattern-guessed address read as a departure.** A 5.1.1 on an address that never existed starts a "find the new employer" search for someone who never left. False verdict: follow the mover. Check: the provenance field, which rule R5 reads before rule R6.
7. **Auto-swapping a lookalike contact.** Automation templates assume every bounce is a job change and re-sequence someone else at the account. False verdict: an unearned follow the mover, aimed at the wrong person. Check: require logged public evidence before any re-sequence.
8. **Over-retention.** Keeping a full profile on a bounce-only record goes beyond the ICO's minimum-data guidance. False verdict: retire in name only. Check: the retired record holds only the fields listed in the next section.

> **Watch out:** A code-only lookup cannot tell a departure from a domain move
>
> "Recipient address rejected" means one address and "Relay Access Denied" means the whole domain, yet both can carry 5.4.1. If your tooling stores only the code, failures 1, 2 and 5 are invisible until you start storing the reply text.

Most teams have one more blind spot: they never measure their own mix. Nobody has published an audit of what share of hard bounces are departures, guessed addresses, domain moves or policy blocks. If your guessed-address share is high, the fix is upstream in how addresses are built, not in the bounce queue.

## What does a retired record keep?

A retired record keeps a minimal suppression entry: enough to stop you contacting the dead address again, and no more. Deleting the record entirely removes the protection, and keeping the full profile goes beyond what the ICO calls the minimum.

The ICO's direct marketing guidance says that when someone opts out you "should put their details onto a suppression or 'do not contact' list, instead of deleting them," and that you should "only keep the minimum amount of information needed." The ICO also says it does not consider a suppression list to be used for direct marketing, so "there is no automatic right for people to have their information on such a list deleted."

That guidance covers objections. It does not cover bounces. Nobody has published guidance on whether a bounce-only suppression should block contact at a newly found address.

> **Note:** Bounce suppression and objection suppression are different
>
> I treat a bounce-only suppression as blocking the dead address, and an objection as blocking the person at every address. Store which kind each entry is, because published guidance does not settle the bounce case; confirm your own position with counsel.

**CRM write-back fields for a graded bounce**

```
bounce_code: 550
bounce_enhanced_code: 5.4.1
bounce_reply_text: <verbatim reply line>
bounce_reporting_host: <host from the DSN>
bounce_date: <date of the bounce>
grade_reply_class: A
grade_domain_pattern: 1
grade_departure_evidence: N
grade_provenance: V
grade_value: H
verdict: RE-ADDRESS (hold)
verdict_rule: R6
recheck_date: <date two weeks out>
departure_evidence_url: <blank if none found>
departure_evidence_seen: <date, blank if none found>
suppression_type: <none / bounce-address / objection-person>
graded_by: <operator name>
```

*Map each line to a field in your CRM or ATS; leave the evidence fields empty rather than inventing values when nothing was found.*

For a retired record, keep the address, the suppression type, the bounce code, the reply text and the date. Drop the rest of the profile. Re-read the ICO guidance on a schedule. Its pages carry dated change notes, and the definitions that sit under it have been amended by legislation before.

## How do you keep the bounce read current?

Keep it current by measuring your own bounce mix every quarter and rechecking the few provider rules the framework depends on, because every threshold in it is either a provider policy that can change or a convention of mine that your data may overrule.

Measure your own split. The most-cited decay figure counts records, not causes. MarketingSherpa found B2B contact data decaying at roughly 2.1% a month, and HubSpot's database decay simulation turns that into 22.5% a year. Those are one finding, not two independent sources agreeing, and neither says whether a bounce is a departure, a guess or a migration. The US Bureau of Labor Statistics median employee tenure of 3.9 years gives you a rough sense of turnover, but it says nothing about how your addresses fail. Tally your verdicts by rule fired each quarter. That count is the only local evidence you will have.

Recheck the mechanisms, not just the current values:

- **Mailbox retention windows.** Microsoft has shortened the post-hold recovery period before, from 183 days to 30. Re-read the Purview inactive-mailbox page when departure timing looks off.
- **Outlook sender requirements.** The 5,000-a-day authentication rule and the 5.7.515 wording come from Microsoft's published requirements for high-volume senders. Check them whenever you add a sending tool.
- **Google's sender guidelines.** Confirm that there is still no numeric bounce requirement and that the spam thresholds remain 0.10% and 0.30%.
- **Your own conventions.** The two-address domain trigger and the two-week recheck are mine. If your quarterly tally shows domain checks rarely finding migrations, raise the trigger.

#### Before you close this week's bounce batch

- [ ] Every bounced record stores the verbatim reply text and reporting host, not just a bounce flag.
- [ ] No record with a 5.7.x reply has been suppressed, and each one is with the deliverability owner.
- [ ] Every domain with two or more bounces, or any relay wording, has a domain-check result logged before any record there was retired.
- [ ] Every follow-the-mover verdict has a departure evidence URL and the date it was seen.
- [ ] No contact was re-sequenced at the old account on the strength of a bounce alone.
- [ ] Every retired record holds only address, suppression type, code, reply text and date.
- [ ] Every hold has a recheck date in the CRM.
- [ ] The rule that fired is recorded on every record, so a second operator could reproduce the verdict.

## Frequently asked questions

### Does 550 5.1.1 mean the person left the company?

Not on its own. Google's wording for 550 5.1.1 is that the account you tried to reach does not exist. That is true when someone left and their mailbox was removed, and equally true when the address never existed because it was pattern-guessed. Check provenance first: a guessed address returning 5.1.1 is a re-address case, not a departure. Only logged public evidence with a URL and a date supports a follow-the-mover verdict.

### Should I suppress a 550 5.4.1 "Recipient address rejected: Access denied" bounce from cold email?

Not until you have run two checks. Microsoft says 5.4.1 comes from directory-based edge blocking because the recipient address is invalid, so it is a recipient-side failure and not a policy block against you. But a hybrid Exchange sync gap can reject someone who still works there. If other addresses at that domain still deliver and there is no departure evidence, hold the record with a dated recheck. If the wording is "Relay Access Denied", check the whole domain instead.

### Hard bounce vs policy bounce: should I suppress or not?

Suppress true address failures only after the read, and never suppress policy or authentication rejections. A 5.7.515 reply names your sending domain and fails every Outlook.com recipient from it. Gmail's 5.7.1 "likely unsolicited mail" is about your reputation or content. Both go to the deliverability owner as fix-the-sender. Suppressing them throws away good records and hides the fault that caused the bounce.

### What bounce rate gets a sending domain blocked?

No major mailbox provider publishes one. Google's sender guidelines contain no numeric bounce requirement, and its only published threshold is for spam complaints: stay below 0.10% and never reach 0.30%. Klaviyo calls staying under 1.0% best practice, which is a vendor norm, not a provider limit. Outlook.com enforces authentication for senders above 5,000 messages a day. Resending to dead addresses hurts through reputation, and how many resends it takes is not publicly established.

### Can I contact a bounced person at their new company?

Published guidance does not settle this. The ICO says to put people who object on a suppression list rather than deleting them, but that guidance covers objections only. Nobody has published whether a bounce-only suppression should block a newly found address. I treat a bounce suppression as blocking the dead address, and an objection as blocking the person everywhere. Record which kind each suppression is, and confirm your own position with counsel.

### How many bounces at one domain mean the company migrated its email?

No published threshold exists. Microsoft's own triage asks whether the issue affects one recipient or everyone at the domain. My working convention is to run a domain check before retiring anyone when two or more distinct addresses at one domain hard-bounce in the same batch, or when any reply says "Relay Access Denied". Comparing MX records, the website redirect and any company name change is a sensible desk method, but no source documents it.

---

*From the Refolk guide library. I revise these guides rather than replacing them, so the current version is always at https://www.refolk.ai/guides/hard-bounce-read-four-verdicts*
