Homeware Retailer: Server Room to Azure

Most migration surprises come from things nobody wrote down: a scheduled job that copies a file at 2am, a report that connects to a server by name, a supplier feed hard-coded to an IP address. Before a single workload moved, we mapped every integration, scheduled task and hard-coded server name across the estate, so problems were found on paper rather than on cutover night.

That map became the wave plan. Low-risk foundations and file shares went first, then reporting, then the ERP database and finally the web store, so each wave proved the next one was safe.

• Integrations, scheduled jobs and server names inventoried
• Workloads grouped into waves by risk and dependency
• A landing zone built in Terraform so environments are repeatable

The ERP database was the heart of the move. Using Azure Database Migration Service, we replicated the on-premises SQL Server to Azure SQL continuously in the weeks before the switch. By cutover weekend the copy in Azure was already current, so the final synchronisation took minutes rather than the hours a backup-and-restore would have needed.

Moving to a managed database also removed a list of chores from the team: patching, backups and high availability are now handled by the platform rather than by whoever is on call.

We ran two full dry runs on real data before the live switch, timing every step and writing down what to do if any of them failed. On the night, the runbook was followed line by line on a Saturday evening: freeze, final sync, repoint the web store, switch DNS through Cloudflare, smoke-test and reopen. A rollback path was ready at every step and never needed.

Because the web store and the shops kept trading, the migration was invisible to customers. On Monday morning, staff simply carried on working.

A cloud bill can drift quickly if nobody owns it. We set up reserved capacity for the steady workloads, chose storage tiers to match how often files are actually opened, and configured budget alerts so the team hears about unexpected spend before the invoice arrives. The result is a monthly infrastructure cost 31% lower than the forecast for replacing the hardware.

With automated backups and geo-redundancy in place for the first time, the retailer is also better protected than it ever was in its own server room, which was emptied before the lease ran out.