Site Migration Case Study

Migration case studies usually come in two flavours — the triumphant "we lost nothing!" and the cautionary catastrophe. This one is more useful: a real mid-size migration (platform change plus URL restructure, ~1,400 URLs, a revenue-dependent content site) that went mostly right, dipped exactly where the checklist predicts, and leaked in two specific places we caught and fixed — making it a map of where migrations actually bleed even when professionally run.

The setup and the discipline

Scope frozen deliberately: platform + URL structure only, templates visually near-identical, content untouched (the checklist's don't-stack-migrations rule, honoured for once). The prep: full crawl inventory with traffic/rankings/links per URL, an exact-counterpart redirect map (1,400 rows, the ~200 with external links triple-checked), staging locked and pre-verified (titles, canonicals, schema and internal links all pointing at new URLs), benchmarks exported. Launch day ran the mechanical hour: deploy, redirect map live, gate off, indexability asserted, new sitemap submitted with the old one kept temporarily reachable.

What the monitoring caught

1. The expected dip: organic traffic −12% across weeks 1–3 as the recrawl processed the moves — important pages re-evaluated within days, the long tail over weeks, matching the equity-test timeline exactly. No action; that's the settling.
2. Leak one, week 2 — the parameter ghosts: the 404 report filled with old-URL variants the map never contained — tracking-parameter and pagination permutations of old URLs, arriving from years-old external links and bookmarks. The map had covered canonical forms; the wild carries variants. Fix: pattern-based redirect rules (regex over enumerated rows) catching old-structure-plus-anything — 404s with referring links dropped to near zero. Lesson: redirect maps need pattern rules, not just row lists.
3. Leak two, week 4 — the template's quiet chains: the new platform's category pagination linked through a redirecting URL form (trailing-slash mismatch), creating one-hop internal chains sitewide — invisible to users, visible to the crawler, precisely the friction the hygiene rules ban. Fix: template-level URL form corrected; the crawl-and-diff habit caught it, nothing else would have.
4. Recovery: week 6 traffic at 97% of baseline, week 10 at 104% — the restructure's cleaner architecture (shallower money pages, better inlink distribution) plausibly responsible for finishing above par, though the honest attribution is "clean migration plus mild architecture dividend."

The transferable findings

Migrations bleed at the edges of the map, not its centre — the enumerated rows worked perfectly; the variants and template-generated forms leaked. Budget the post-launch weeks for edge-hunting (daily 404-with-referrers review, weekly crawl diffs) rather than map-admiring. The dip is informational, not actionable — teams that "fix" mid-settle create the second incident; the pre-agreed no-panic window (we used four weeks, checkpoints only) is a governance tool as much as a technical one. And the benchmark export is what made "97%" a knowable number — per-page baselines turn recovery from vibes into arithmetic, per the measurement discipline.

Frequently asked questions

Would you ever migrate domain and structure together?

When forced (rebrand + platform end-of-life), with doubled settling expectations and the same edge-hunting doubled — domain moves add trust-transfer lag on top of URL churn, per the checklist's deeper-and-slower warning. Sequenced migrations remain the recommendation whenever the calendar permits.

How long did you keep the old sitemap and redirects?

Old sitemap: until its URLs drained from the index (~two months). Redirects: forever, in version control — old-URL hits continue at low volume indefinitely (we still log them), and every one is a linked or bookmarked visitor the map keeps monetised.

What single practice most protected this migration?

The crawl-diff habit — pre-launch staging crawl, launch-day spot crawl, weekly diffs after: it caught leak two, verified leak one's fix, and confirmed settling wasn't drift. Second place: scope freeze. Everything else was diligence on the standard list — and the equity being protected was, as always, the years of earned links that made the site worth migrating carefully (we help build that kind).