Email Deliverability

Email Provider Migration Checklist: 40 Items Before You Switch

A 40-item email provider migration checklist organised into seven phases from authentication and DNS through to sign-off, with the three highest-risk items marked

Migrations rarely fail on a missing step. They fail on a step nobody owned.

Someone assumed the DNS change was handled because someone else mentioned DNS in a meeting. The suppression list was exported, but only for the primary domain. The rollback plan existed, in the sense that somebody said, “We can always switch back,” out loud.

This is the list of 40 checks, grouped into the seven phases they belong to, with the reason each one matters. Work through it with a name against every line.

Quick Answer: The Seven Phases

PhaseItemsWhen
Authentication and DNS1–8Before any traffic moves
Data and suppression9–14Before the first production send
Integrations and code15–22Before the first production send
Content and templates23–27Before the first production send
Monitoring and measurement28–33Before you ramp past 5%
Rollback readiness34–37Before you ramp past 5%
Sign-off38–40Before stage 1

Three items cause most incidents: item 2 (SPF lookup count), item 12 (suppressions imported before the first send) and item 18 (no duplicate send path). If you only verify three things, verify those.

How to Use This

Timeline diagram showing which email migration checklist phases must be clear before the first production send and which are re-verified at each rollout stage.
Four phases gate the first send. Two gate the ramp. Sign-off comes before any of it.

Two rules make the difference between a checklist that works and one that gets pasted into a document and forgotten.

Every item needs a named owner, not a team. “Platform team” is not an owner. An item assigned to a group is an item assigned to nobody, and those are the ones that turn up unfinished at 50% rollout.

This is a gate, not a one-time pass. Items 28–37 get re-checked at every ramp stage, because what was true at 5% can stop being true at 50% — alerts that never fired, a rollback path that quietly broke when someone cleaned up an unused environment variable.

Common Mistake

Treating the checklist as pre-flight only. Teams work through all 40 items carefully, start the ramp, and never look at the list again — which is exactly when the monitoring and rollback items start to decay. The rollback was verified in week one; by week three somebody has rotated the old provider’s credentials during a routine security sweep, and the rollback that everyone believes in no longer exists. Re-run phases 5 and 6 at the start of every ramp stage. It takes ten minutes and it is the difference between having a rollback and believing you have one.

This checklist covers what to verify. It deliberately does not cover the order in which to move traffic — that is in our guide to migrating email providers without downtime, and the measurement method is in running two email providers during a migration.

Phase 1: Authentication and DNS (Items 1–8)

Authentication failures are one of the most common causes of email delivery problems, and during a migration they are the most likely thing to break both providers at once rather than just the new one.

#CheckWhy it matters
1New provider’s include added to SPF, with the old one still presentBoth must be authorised during the parallel run
2Total SPF DNS lookups counted and confirmed under tenExceeding it returns permerror and fails authentication for both providers simultaneously
3DKIM published on a new selector, independent of the old provider’sIndependent selectors let both sign in parallel with no conflict
4The DKIM signature was verified on a real send, not just a published recordA record that resolves is not proof the signature validates
5DMARC record present with a policy of at least p=noneGoogle requires a DMARC record from bulk senders
6From-domain alignment confirmed on SPF or DKIMOnly one needs to align, but one must
7TTLs lowered at least one full old-TTL period before cutoverLowering a TTL only takes effect after the old one expires everywhere
8Any sending subdomain has an A or MX record in public DNSSome receivers treat a subdomain with neither with more suspicion

Item 2 is the one that bites hardest. The ten-term limit is specified in RFC 7208, and it counts include, a, mx, ptr, exists and redirect — but not ip4, ip6 or all. Adding one more provider’s include is often what tips a long-standing record over the edge. Verify with MXToolbox and the free email deliverability checker, and see our guide to SPF, DKIM and DMARC for the mechanics.

Phase 2: Data and Suppression (Items 9–14)

#CheckWhy it matters
9Suppressions exported from every list type the old provider keepsSome providers split them into several separate lists
10Exported per sending domain, and per stream where the provider has themSuppression data is often scoped below the account level
11Export stored in your own system, not only uploaded to the new providerThis is your data; the next migration should not need an export at all
12Hard bounces and complaints are imported before the first production sendA new sending identity with no bounce history gets judged hardest on its first sends
13Explicit decision on which categories not to importProvider-side blocks reflect that provider’s infrastructure, not your recipients
14Final re-export scheduled for the last day of the parallel runAddresses that bounced during the ramp exist only on the old provider

Item 13 catches people out. Not every suppression is about the recipient — some record the old provider’s own delivery decisions, and importing those carries another company’s infrastructure history into your new setup. Our guide to reducing bounce rate covers why the first sends from a new identity matter disproportionately.

Phase 3: Integrations and Code (Items 15–22)

#CheckWhy it matters
15Every send site found by grepping for the old provider’s hostname and SDKA forgotten cron job or admin tool keeps sending after cutover
16All old credentials inventoried, including any in CI and in stagingCredentials in a pipeline are the ones nobody remembers
17Routing by recipient hash implemented, percentage in configurationA rollback that needs a deploy is not a rollback
18One send site, one decision, one provider — no duplicate pathAdding a second call rather than routing causes duplicate delivery
19New provider’s webhook endpoint deployed and publicly reachableEvents you cannot receive are events you cannot compare
20Webhook signature verification implemented and tested against a real eventSigning schemes differ entirely between providers
21Every payload shape the new provider sends is handledSome providers post more than one shape depending on the event source
22Inbound mail checked — MX records and any routing rulesIf the old provider also receives mail, that is a second migration

Item 22 is the one most likely to be missed entirely, because inbound routing is configured once and then never touched. Its failure is silent in both directions: senders get a bounce you never see, and your application simply stops receiving the posts it used to. See migrating from Mailgun for the reimplementation pattern, and the SMTP relay API reference for the sending side. PhotonConsole issues SMTP credentials on signup, so items 17 to 21 can be built and tested against a live endpoint before any production traffic moves.

Quick Fix

The Five Items That Cause Most Incidents

  • Item 2 — SPF lookups over ten returns permerror and breaks both providers at once
  • Item 12 — suppressions imported after the first send instead of before
  • Item 18 — a second provider call added rather than traffic routed, causing duplicate delivery
  • Item 22 — inbound mail forgotten, failing silently for days
  • Item 36 — old credentials rotated mid-migration, destroying the rollback nobody re-tested

Phase 4: Content and Templates (Items 23–27)

#CheckWhy it matters
23Templates exported into version control, not into another provider’s storeVersion-controlled templates make the next change a config edit
24Layouts or shared wrappers flattened or reproducedA template exported without its wrapper looks almost right
25Every template rendered against missing and null variablesRenderers differ on undefined values: blank, literal placeholder, or error
26Marketing mail carries a one-click unsubscribe with an HTTPS URLRequired by Google for bulk marketing mail, per RFC 8058
27Tracking and link domains updated where the provider rewrites URLsLinks pointing at a decommissioned tracking domain break after teardown

On item 26, Google’s bulk sender requirements ask for one-click unsubscribe on marketing and promotional messages, implemented with List-Unsubscribe headers under RFC 8058, including one HTTPS URL in the header. Transactional mail is exempt. A mailto link or a link in the body does not satisfy it, and Google recommends fulfilling requests within 48 hours.

Phase 5: Monitoring and Measurement (Items 28–33)

A migration checklist ledger with team placeholders and vague criteria compared with one carrying a single named owner and a numeric criterion for each item
An item owned by a team is an item owned by nobody. Names, numbers and a date.
#CheckWhy it matters
28Both providers’ events normalised into one vocabulary of your ownTwo dashboards both reading 99.2% are not measuring the same thing
29Unmapped event types logged and counted, never silently discardedA dropped bounce category makes a provider look better than it is
30Rates reported segmented by receiving domain, not only as an averageA problem at Gmail and none at Outlook is a finding an average hides
31Google Postmaster Tools access confirmed for every sending domainProvider-reported delivery is acceptance, not inbox placement
32Seed inbox tests scheduled on both pathsShows where mail lands rather than whether it was accepted
33Alerts configured on bounce rate and complaint rate thresholdsNobody watches a dashboard at 3am on the third day of a ramp

For item 31, Postmaster Tools reports spam rate and domain reputation from Google’s side. Google asks senders to stay below 0.1% user-reported spam and to prevent the rate from ever reaching 0.3%; a sender above 0.3% is ineligible for mitigation until it has stayed below that for seven consecutive days. Those are the numbers to put in item 39. Mail Tester covers item 32 for individual messages.

Phase 6: Rollback Readiness (Items 34–37)

#CheckWhy it matters
34Rollback to 0% achievable without a deployIf reverting needs a release, you cannot abort quickly
35Old provider still fully functional — not downgraded, paused or cancelledA downgraded plan may not handle 100% of your volume any more
36Old provider’s credentials still valid and excluded from rotationA routine security sweep silently destroys the rollback
37Rollback executed once, in production, at a low percentageAn untested rollback is a plan, not a capability

Item 37 is the one people resist and the one worth insisting on. Take the rollout from 5% to 0%, confirm mail still flows entirely through the old provider, then put it back. Ten minutes, during business hours, while the stakes are low. The alternative is discovering at 50% that the revert path has a bug.

Phase 7: Sign-Off (Items 38–40)

#CheckWhy it matters
38A named individual against every item aboveAn item owned by a team is owned by nobody
39Numeric exit criteria agreed in writing before stage 1A ramp without criteria ends when somebody gets bored
40A decommissioning date set, with the teardown order written downMigrations that are never finished leave credentials live indefinitely

Quick Fix

The Day Before Cutover

  • Re-verify SPF resolves and the lookup count is still under ten
  • Confirm the old provider’s credentials are still valid — test an actual send
  • Re-export suppressions from the old provider and import them
  • Check that alerts from item 33 have fired at least once in testing
  • Confirm nothing else ships today: no template release, no DNS edit, no campaign

Note

Items 39 and 40 look like paperwork and are the two that most often decide whether the migration ends. Without agreed numbers, a single hard bounce in a small cohort becomes an argument rather than a data point, and the ramp stalls at 5% indefinitely. Without a decommissioning date, the old provider stays live for months with valid credentials and an open invoice, which is both a cost and a security exposure. Set both before stage 1, when nobody is under pressure and the discussion is cheap.

Pro Tips

  • Put a name, not a team, against every line. This single change catches more than any individual item.
  • Count your SPF lookups before adding anything. It is the one mistake that breaks both providers at once.
  • Import suppressions before the first send, never after. First impressions on a new sending identity are expensive.
  • Grep for the provider’s hostname, not its SDK name. The hostname appears in config files the SDK search misses.
  • Test the rollback for real, at 5%. Ten minutes to turn a plan into a capability.
  • Exclude the old provider’s credentials from rotation until teardown, and tell whoever runs the sweep.
  • Re-run phases 5 and 6 at every ramp stage. Monitoring and rollback decay quietly.
  • Change one thing at a time. A template release during a ramp invalidates the week’s data.

Related Reading

This checklist is the hub for our migration cluster. Each article below covers one part in depth:

Frequently Asked Questions

What is the most commonly missed item in an email provider migration?

Inbound mail. If the old provider also receives mail through an MX record, that is a second migration with its own cutover, and its failure is silent in both directions.

Which item breaks migrations most often?

The SPF lookup count. Adding a second provider’s include can push the record past the ten-term limit, which returns permerror and fails authentication for both providers at the same time.

Should I import every suppression list from my old provider?

No. Import hard bounces and complaints, which are facts about recipients. Decide deliberately about provider-side blocks, which reflect that provider’s infrastructure rather than your recipients.

When should suppression data be imported?

Before the first production send. A new sending identity has no bounce history, so mailing previously-bounced addresses from it teaches receivers that this identity sends to addresses that do not exist.

How do I avoid sending duplicate emails during a migration?

Route rather than add. One send site makes one decision and calls one provider. Adding a second provider call alongside the existing one causes duplicate delivery and corrupts your comparison data.

What spam rate should trigger a rollback?

Google asks senders to stay below 0.1% and to prevent the rate from ever reaching 0.3%. At or above 0.3% a sender is ineligible for mitigation until seven consecutive days below it, which makes 0.3% a sensible hard stop.

Do I really need to test the rollback?

Yes. Take the rollout from 5% to 0% in production, confirm mail flows through the old provider, then restore it. An untested rollback is a plan rather than a capability.

How long should I keep the old provider after reaching 100%?

Around two weeks, fully functional and with valid credentials. Set the decommissioning date in advance so the account does not stay live for months.

Why does every item need an individual owner?

Because an item assigned to a team is assigned to nobody. Items owned by groups are the ones that turn up unfinished halfway through a ramp.

Conclusion

Forty items sounds like a lot until you consider that most of them take under five minutes and three of them prevent the incidents that make migrations memorable.

If you do nothing else from this list, do these: count your SPF lookups before adding a provider, import suppressions before the first send rather than after, and confirm you have one send site making one decision rather than two calls running side by side. Those three cover the failures that produce duplicate invoices, authentication outages affecting both providers, and a burst of bounces on a brand-new sending identity.

Then do the two that feel like administration. Put a person’s name against every line, and agree the numbers that mean proceed, hold and roll back before the first percent of traffic moves. Checklists do not fail because items are missing. They fail because nobody owned the one that mattered, and because without agreed criteria a single bounce becomes an argument instead of a data point.

Done properly, what you keep is worth more than the switch itself: templates in version control, a suppression list you own, and an event vocabulary that is not shaped around one vendor’s API. At that point the next migration is a configuration change. If you are still choosing a destination, our comparison page sets the options side by side, and PhotonConsole issues credentials on signup so the new path can be live and testable while you work through phase 1.

Read More

phoconadmin

About Author

Leave a comment

Your email address will not be published. Required fields are marked *

You may also like

Email Deliverability

Why Emails Go to Spam in Gmail: 7 Real Reasons + Fixes (2026)

Struggling with emails going to spam in Gmail? Discover the real reasons behind poor deliverability and learn step-by-step fixes to
Email Deliverability

SMTP Not Working? 10 Common Errors & How to Fix Them (Step-by-Step Guide)

Your SMTP stopped working. Emails aren’t going out. OTPs are failing. Users are complaining they never got their verification link.