The customer was sending 50M emails a month through SendGrid across four product lines. They wanted to move — better pricing, dedicated CSM, open-source platform — but had every reason to fear a botched cutover.
Here's what worked.
Day 1-2: Parallel domains. We provisioned the same three CNAMEs they had on SendGrid, but on subdomains (mail2., track2., s2._domainkey.). Both vendors live, no DNS conflict.
Day 3-5: Mirror traffic at 1%. Used a feature flag on their send pipeline to route 1% of every product line through Inbox OSS. Compared inbox-placement scorecards side by side.
Day 6-8: Ramp to 25%, then 50%. Engagement metrics matched SendGrid within 0.4 points. Bounce rate was 0.3 points better (validation enabled).
Day 9-10: 100% cutover. SendGrid kept warm for fast rollback.
Day 11: Decommission SendGrid. No rollback needed. Spam-folder rate dropped from 3.4% to 0.8% in week one.
Lesson: parallel-traffic migration with explicit metric gates beats a flag day every time.