A database change is a business change as well as a technical one. The important question is not simply whether a script runs. It is whether the people using the system can still find, trust and act on the right records afterwards.
Start with the records and decisions
Write down what the system stores and what people do with it. An order, a customer, a project and an invoice may look like rows, but each has a different owner, lifecycle and legal significance. Find out which system is authoritative for each fact. A new reporting database should not quietly become an alternative accounting ledger.
Describe the problem in observable terms: repeated entry, missing validation, slow reports, conflicting copies or access that is too broad. That gives a useful acceptance test. “Move everything into a new database” does not explain what should improve.
Find the dependencies before changing a field
A column can feed a report, an integration, a scheduled export or an older desktop application. List the tools that read and write the data, the people responsible for them and the important operating times. Identify whether more than one application may run during a transition.
Check who owns the system and who can authorise access. Do not send a production database, customer records or passwords with an initial enquiry. A schema description or a synthetic example is often enough to begin discussing the work.
Treat recovery as a demonstrated capability
Having a backup file is different from knowing it can be restored. Agree what must be recoverable, where the copy is kept, who can decrypt or access it, and how long restoration can reasonably take. A useful rehearsal restores a copy and checks meaningful records and relationships, not merely that a command reports success.
Some changes are easy to reverse before new work arrives and difficult to reverse afterwards. The plan should say how changes made during a transition will be handled. A rollback that discards the morning’s orders is not a safe business recovery plan just because the old software starts.
Rehearse with the least sensitive data possible
Use synthetic records where practical. Include awkward cases: empty values, duplicate identifiers, unusual characters, retired records and linked transactions. If a rehearsal genuinely requires personal or confidential data, agree the authority, minimisation, access, storage location and disposal first.
Check counts, totals where meaningful, unique keys, required fields, permissions and the relationships between records. Test that the wrong user cannot read another customer’s data. If old and new application versions may overlap, test the supported compatibility window explicitly.
Make the decision before the change window
The written plan should identify the exact change, responsible operator, approved source, expected checks and stop conditions. Agree when users must pause work, how they will be told the result and who can accept the outcome. Engineering verification, business delivery acceptance and customer contractual acceptance answer different questions.
If a check fails, stop and investigate rather than improvising a wider change. Record what happened and keep the original recovery evidence. The safest next step may be to restore, defer or narrow the work.
Leave a maintainable result
Ask for the schema and migration source, configuration and licence information, a concise operating note, recovery instructions and an export route. Clarify which monitoring or support is included and who owns the ongoing system.
The goal is not a more complicated database. It is a dependable record of the business, with a change process another competent person can understand.
Sources and review
Prepared by Yellow Brick Tech. General guidance; the source and test conditions for your own situation still matter.
Next review: 30 April 2027