Data migration consultant for companies in Germany

Why data migrations go wrong

A data migration is the move of records from an old system into a new one, with their history and their links. It goes wrong where the old system holds meaning that nobody wrote down, for example:

• a status code “9” that staff read as a blocked customer,
• a free-text field that sales use for the delivery note,
• dates in two formats, because an update once changed the input form.

Errors from a migration seldom show on the day of the switch. They show weeks later, when an invoice goes to an old address or a report no longer adds up. So I prove a migration with counts, and spot checks come on top.

How I move your data

The work runs in rehearsals. Each run goes from an export of the old system into a copy of the new one and ends with the same set of counts, so two runs can be compared.

Before the first run, a mapping sheet notes for each field of the old system where it goes and how its values change. A value that has no place in the new system is a decision for your staff, and the sheet records who made it.

Imports run under an account of their own and not under a personal login. A faulty batch can then be found and taken back by account and time range, as described on How other systems couple to the store. The rollback retires the rows the batch wrote and deletes nothing.

Before a table or a field goes out of use, a check proves that nothing reads or writes it anymore. The old system stays readable after the switch, and its data is deleted after a date you set.

Moving servers is a different job with risks of its own. I did one at the Frankfurt-based company Blocksize Capital, from Azure to another cloud, with a second site that takes over when the first fails. Details: Node operations and a cloud migration with failover.

The counts that prove a migration

After each run I compare the counts. The old system and the new one give the same number, or the difference has an explanation that your staff accept.

Count	Old system	New system	A difference means
Records per type, such as customers, products and orders	Count per table	Count per table	Records were lost or created twice
Records per status, such as active and archived	Count per status	Count per status	A status code was mapped wrong
Sum of open invoice amounts	Sum in euros	Sum in euros	Amounts, currencies or decimal places changed
Orders per customer	Count per customer number	Count per customer number	Links between records broke
Attachments	Number of files and total size	Number of files and total size	Files did not move or were cut off
Required fields left empty	Count per field	Count per field	The mapping drops a value
Merged duplicates	Reported on their own	Reported on their own	A merge rule went beyond what was agreed

Who I am

I am Michael Wutzke, based in Frankfurt, with more than twenty years in IT and media. I have set up servers and backup systems in data centers in Germany and the USA, and I ran a community platform on load-balanced servers whose databases I tuned. Today I help a client break up a monolithic application with AI coding agents. Details: Career stages.