When a replatform is planned, attention goes to products and customers. That part usually goes fine — it is visible, countable and testable.
What causes trouble is everything running quietly behind the old system that nobody documented and nobody rebuilt.
Data that looks migrated but is not
Customer passwords never transfer. They cannot, from any platform to any other, because passwords are stored irreversibly. Customers have to reset on the new store, and if that is not communicated they contact support instead.
Product reviews usually live inside an app rather than on the product. If the new store uses a different review app, that transfer needs planning on its own — and reviews carry both social proof and search value.
Order history can move, but often read-only. Decide up front how returns and exchanges against old orders will be handled.
Custom fields. Every extra piece of information attached to a product in the old system — size chart, material, warranty period — has to be rebuilt as metafields. Skip this at inventory stage and product pages go live incomplete.
The SEO losses are the expensive ones
URLs change, and without a redirect map the history is simply deleted. The one-line version: build the map before the store, not after.
Page titles and meta descriptions get replaced with template defaults during a migration. A category that ranked on a carefully written title stops ranking on a generated one.
Structured data is rebuilt by the new theme and may simply not be there. Internal links inside body content still point at old URLs and should be rewritten rather than left to redirect.
Operations: the first week
Shipping integration. The label flow has to be tested with a real order — settings that look right are not the same as a flow that works.
Invoicing. Whether an invoice is generated automatically when an order lands, and where it fails if it does not, should be discovered on a test order rather than a customer's.
Email notifications. Order confirmation, shipping and return emails need their content and sender address checked. The number of stores that launch on default templates is surprising.
Payments. Place a small real order and confirm the money arrives. A gateway that works in test mode can behave differently live.
Pre-launch checks that catch most of it
Before removing the password, run these end to end: add a product to cart and complete payment; create a return request; check that an out-of-stock product displays correctly; run the whole flow on mobile; and apply a discount code.
Those five move most of the first-week problems to before launch, where they are cheap.
Then leave everything else alone for a month. Migrations that also redesign and rewrite at the same time make it impossible to attribute any change in performance to any cause.