E-commerce migration: a real case study (without breaking the store)

Guilherme Rodrigues
Guilherme Rodrigues
November 19, 2026
E-commerce migration: a real case study (without breaking the store)

A migration breaks a store in one of two places: the URLs search engines already rank, and the templates that render your PLP and PDP. Get those two right and a replatform is boring, which is exactly what you want. This is how a migration actually runs, with two real ones (New Balance and Ri Happy), and the safe-change loop that keeps the storefront live while you move it.

What is an e-commerce migration (and what actually breaks)

An e-commerce migration, or replatforming, means moving your storefront from one platform to another: VTEX IO to FastStore, a monolith to headless, Strapi or Magento to something you can edit without a dev ticket. The catalog, the content, and the traffic have to arrive on the other side intact.

Two things break most often, and neither is exotic. Missing or wrong redirects drop the URLs you already rank for: Google's own guidance is blunt that each old URL should point to the closest new page, and that blanket-redirecting everything to the home page gets treated as a soft 404 (Google Search Central). Broken templates are the other half: a product page that renders on three SKUs and collapses on the fourth, a category filter that returns an empty shelf. The demo looks fine; the long tail does not.

The reassuring part: a migration is not fated to cost you traffic. Most replatforms go smoothly, and some practitioners report migrations that raised SEO traffic because the old site was slow and poorly structured (Practical Ecommerce). The ones that go wrong almost always skipped the two checks above.

Two real cases: New Balance and Ri Happy, from VTEX IO to FastStore

New Balance Brasil moved its whole store to FastStore in three days. It ran on a VTEX IO storefront with years of customization that had passed through many hands, and its mobile PageSpeed sat at 37. The home, product and listing pages were rebuilt on the FastStore design system over a few weeks and tested before any traffic arrived. Then traffic moved in stages, from 25% to 50% to 75% to 100% between October 5 and 7, each step running on real shoppers before the next one opened. Mobile PageSpeed went from 37 to 92, JavaScript errors fell 81%, and conversion on direct and organic traffic rose 9.3% over the first 21 days against the 21 days before. Sessions dipped the following month for seasonal reasons, which is why any before-and-after comparison should say what else changed.

Ri Happy, Brazil's largest toy retailer, went live on FastStore in 14 working days, from kickoff on September 17 to go-live on October 5, ahead of Children's Day. A replatform that size usually takes about a year. A parity agent rebuilt the pages one at a time and checked each one against the original (about 200 pages in a single 12-hour run), so the people on the project spent their time on business rules and the edge cases the agent flagged. Checkout and order management stayed on VTEX, untouched, and the cutover rolled out by percentage of shoppers, with rollback measured in minutes.

What made both moves boring instead of terrifying was not a bigger plan. It was running the migration as a loop of small, reversible changes instead of one big-bang cutover, and proving each step before it reached the whole store. The agents did the repetitive rebuilding and comparison; the team decided the scope, what to keep, and when each traffic step opened.

How to migrate without breaking the store

The safe-change loop is the same one deco runs on every storefront change, and a migration is just a dense sequence of them: make a change, test it, preview it, publish it, watch it. Four practices do the heavy lifting.

  • Map every redirect before cutover. Use permanent (301/308) redirects, point each old URL at its closest match, and send genuinely dead pages to a 404/410 rather than a fake destination. Google says permanent redirects do not cost you PageRank, so a complete map preserves the rankings you already earned (Google Search Central).
  • Preview and stage, don't deploy on faith. Every change renders on a preview URL that mirrors production before it goes live. The storefront a shopper sees only changes when someone approves what they are looking at.
  • Put the store under version control. With Git behind the storefront, every change is a branch you can discard and every publish is a commit you can roll back. "It can look ugly, but it cannot break" is the whole point: a bad change is reverted in seconds, not firefought for a day.
  • Change one thing at a time, and let it settle. Google recommends moving in sections and changing only one variable at once, and notes it can take a few weeks for a medium-sized site's new URLs to fully replace the old ones in search (Google Search Central). A sectioned rollout also means a problem shows up on one slice, not the whole catalog.

VTEX makes the same point from the platform side: replatforming can hit performance, so continuous A/B testing is the best ally during the move, not an afterthought once it is done (VTEX). Measure the new storefront against the old on real traffic before you flip everyone over.

This is where deco fits: it is the AI platform and team that operates the storefront's whole lifecycle, and a migration is the nascer / migrar phase of that lifecycle. The agentic part, an AI that makes the change, a divergent QA agent that checks it, a human who approves it, is the "how" underneath the loop, not a reason to hand over the keys. You see what changed, and you can revert it.

How long does an e-commerce migration take?

Honest answer: it depends on catalog size and how many custom templates you carry, and sectioned migrations run longer than they look because each slice needs to settle in search. Plan for weeks, not a weekend, and treat the first few weeks after cutover as a monitoring window, not a finish line. Google explicitly tells you to keep redirects live for at least a year and to expect gradual, not instant, reindexing (Google Search Central).

FAQ

Will I lose SEO traffic when I replatform?

Only if you lose URLs. Permanent redirects preserve ranking signals; the traffic drops come from missing redirects, lost metadata, and orphaned pages, not from the move itself (Google Search Central).

Big-bang cutover or gradual?

Gradual, in sections, changing one thing at a time. It is easier to monitor and far easier to roll back.

How do I test the new store without risking the live one?

Preview and staging for every change, A/B against the current store on real traffic, and Git so any publish can be reverted.

While you are rebuilding, it is worth making the new store answer two questions the old one often could not: can AI engines cite it, and can it describe its own catalog at scale. See GEO: how to make AI cite your store for the first, and Gerador de descrição de produto com IA for product content that fills a fresh catalog without a content backlog.

check_circle
Free migration-risk assessment

Most broken migrations were predictable. Before you move, run a free migration-risk assessment: a fast slice that scores where your replatform is most likely to break, from an incomplete redirect map to orphaned URLs to templates that only render on the happy path. It takes minutes, it needs only your store's domain, and it tells you what to fix before cutover instead of after. Request the assessment.