ESP Migration to Klaviyo, Braze, Iterable or Customer.io

A migration fails on two files, not on the database: the suppression list and the consent record. We move the suppression list first, every hard bounce, complaint and global unsubscribe from every system the brand has ever sent from, including the platform abandoned in 2019 that nobody mentions in the kick-off. Nothing leaves the new ESP until it reconciles row for row. On the last insurance migration we ran, that recovered forty-one thousand addresses which had unsubscribed from a legacy system and would otherwise have been mailed again on day one. The consent record we insist on: A subscriber row travels with its timestamp, the source URL, the exact wording shown at the point of capture, the session or IP, the double opt-in confirmation where one was used, and for healthcare professional audiences the credential check that gated the record. A flag reading subscribed equals true is not a consent record, and we will not migrate on it. Addresses that cannot be evidenced to that standard go to a re-permission segment or they stay behind. That is an uncomfortable meeting and we have it before the contract, not after. Sending architecture: Marketing on its own subdomain, transactional on another, and the root domain left alone so a deliverability incident in the campaign programme cannot take password resets and policy documents down with it. Click tracking on a CNAME'd domain you own, so your links do not carry a shared vendor domain. SPF kept inside the ten-lookup limit, which usually means removing three vendors nobody remembers adding. DKIM at 2048 bits. DMARC moved from p=none through a percentage ramp on quarantine to p=reject, with aggregate reports going somewhere a person actually reads them. BIMI only after reject is live and a verified mark certificate is bought, and we will tell you when it is not worth the money. Warm-up: Dedicated IPs earn their keep above roughly one hundred thousand sends a month at a regular cadence. Below that, a cold dedicated IP performs worse than a well-run shared pool and we say so before you buy one. The ramp runs four to six weeks, opening with profiles engaged in the last thirty days, and Google Postmaster Tools and Microsoft SNDS are read daily rather than at the weekly status call. We hold complaints under 0.10 per cent through the ramp, against the 0.30 per cent threshold Gmail and Yahoo enforce. Platform choice: Klaviyo, Braze, Iterable and Customer.io. The first question we ask is who runs it on the Tuesday morning after we leave. Braze without an in-house engineer who writes Liquid and understands Connected Content is a platform your team will use at a tenth of its capability while paying in full for it. We have twice recommended a client stay where they were and cancel the migration, and both times that was the correct invoice not to send. Not included: CDP selection or implementation, Segment or RudderStack pipeline work, backend event instrumentation beyond writing the specification your engineers build to, creative production, and legal advice. We will prepare the processing detail your data protection officer needs; we are not your data protection officer. Who this is not for: Anyone who needs the migration finished inside a month. A file with unclear provenance takes six to ten weeks to move properly, and the compressed version is how brands land on a blocklist in week two. Also not for a team whose reason to move is that the current platform feels dated. That is a training problem, and migrating reproduces it on a larger invoice.