Data migration
What is data migration?
Data migration is the move of existing records — products, customers, orders, content — from an old system into a new one, with every field mapped and the result checked. It is not a copy: old data is cleaned, duplicates are merged and formats are brought into the new model. It is planned as its own stage, with a rehearsal run, because mistakes made here surface months later.
Example
A shop is changing platform. There are 3,200 products, 11,000 customer accounts and 28,000 orders going back seven years. The rehearsal run shows that 140 products have no category, 300 customers appear twice with the email written differently, and units of measure have been typed four different ways.
The choice is simple: those 440 records are fixed now, in an hour or two of rules, or they are fixed later, one at a time, by whoever answers the phone. If only 2% of 11,000 accounts break and each costs ten minutes on a call, that is more than 36 hours of support spread over the following year. The figures are illustrative; the rule is not. Cleaning before the move always costs less than cleaning after it.
Why it matters for a business
Data outlives systems. An order from 2021 is both an accounting record and the evidence behind a warranty claim; the customer list is what marketing stands on; product descriptions are years of work. So a migration is judged not by “we moved everything” but by numbers anyone can check: how many records went in, how many came out, how many were rejected and why.
Personal data brings duties with it. Only what is still needed is carried across, test copies are deleted afterwards, and anyone handling them works under a processing agreement as required by Article 28 of the GDPR. Where the old system is a CMS, the content moves too — together with the links and the files it depends on.
We plan and run the migration as part of building the web software, not as the last task before launch.
What to ask
- Is there a written field mapping, and who signs it off?
- How many rehearsal runs are planned, and who checks what comes out?
- How is success counted: by record numbers, by control totals, by sampling?
- What is the plan for going back if the launch fails?
- How long does the old system stay available, read-only, after the switch?