A retail company is migrating its warehouse to a new platform. What makes these migrations fail, and what sequence reduces the risk?

What makes them fail

• Undocumented consumers. A warehouse accumulates dashboards, scheduled exports, spreadsheets pulling directly, and jobs written by people who have left. The consumer list is always longer than anyone believes, and the ones discovered after cutover are the expensive ones.
• Semantic differences between platforms — type coercion, null handling, date arithmetic, string collation, rounding — that produce subtly different numbers with no error. Nothing fails; the report is simply different, and the discrepancy is attributed to the migration only if someone was comparing.
• Query dialect differences, which are tedious and mechanical and are the part everyone estimates.
• Performance characteristics that differ, so a query tuned for one platform's layout is slow on another's.
• Cost model differences, where a workload that was affordable becomes expensive because the new platform charges differently.

The sequence that reduces risk

1. Inventory consumers from query logs, not from a survey — the logs contain the jobs nobody remembers.
2. Dual-run: both platforms fed, both producing outputs, with automated comparison of results. The exit criterion is every difference explained, not zero differences, since some will be the new platform being correct.
3. Migrate consumers progressively, starting with the least critical, rather than cutting over simultaneously.
4. Keep the old platform available for a defined period after cutover, with a date — because the consumers discovered late need somewhere to go.
5. Decommission deliberately, with a date and an owner assigned before the migration starts, since retrofitting a decommissioning project after the fact rarely gets funded — and until the old platform is off, the organisation pays for two and has reduced neither cost nor risk.

The scoping decision that matters most

Does all the history need to move? Migrating ten years into an expensive new platform when queries rarely reach beyond eighteen months is a common avoidable cost.

Migrate the active window and archive the rest with a documented retrieval path, which reduces the migration's size, duration and risk simultaneously — and is frequently the single largest simplification available.