Data migration checklist: moving off spreadsheets or old systems
A practical data migration checklist for moving from spreadsheets or a legacy system: inventory, mapping, cleaning, test loads, cut-over and a rollback plan.
By AM-IT Solution · Published · 5 min read

In short
Migrate data in stages: inventory what you hold, map every field to its new home, clean the data before moving it, run test loads that you check against the source, then cut over on an agreed date with a rollback plan. Keep the old system read-only until the new one has been trusted in daily use.
The new system is the visible part of a project, but the data inside it is what the organisation actually relies on. A good migration is quiet: nothing is lost, nobody has to ask where a record went, and the team can see why the new system is better. The checklist below shows how to get there.
What should you inventory before moving anything?
List every place the data lives today. Spreadsheets are the obvious source, but records are also hidden in email attachments, shared drives, a person’s own files and the old system’s attachments.
- Every source: files, databases, exports and who maintains each one.
- Volume: how many records, rows and files, and how far back they go.
- Owners: who understands the meaning of each column and status.
- Sensitive data: personal, financial or confidential information that needs care in transit and in storage.
- Retention: what you must keep, what you may delete and who decides.
Not everything needs to move. Deciding that closed records from many years ago stay in an archive, searchable but outside the new system, can remove a large share of the work. Under the UK GDPR, personal data should also be accurate, and the ICO expects reasonable steps to check data you hold. A migration is a natural point to do that.
How do you map fields from the old system to the new one?
A mapping document is the centre of the project. For every source field, record where it goes, how it is transformed and what happens when it is empty or wrong.
| Source field | New field | Rule | Question to settle |
|---|---|---|---|
| Name (one column) | First name, last name | Split on the last space; review any record with more than two words. | Who checks the exceptions? |
| Status (free text) | Status (fixed list) | Map each distinct value to one allowed status. | What do the unclear values mean? |
| Date (mixed formats) | Date | Convert to one format; flag anything that cannot be read. | Day-month or month-day for ambiguous dates? |
| Notes (long text) | Notes | Copy as written, with the source date. | Is any of it better as a structured field? |
Spreadsheets cause most of the surprises. Free-text columns, merged cells, one row holding two people and statuses spelled five ways are all normal. The mapping is where you decide how to handle each one before it reaches the new system.
How do you clean the data?
- Remove obvious duplicates and agree which record wins when two disagree.
- Standardise values: dates, phone numbers, countries, status names, currencies.
- Fill gaps that the new system requires, or decide what a safe default is.
- Fix broken links between records, such as a payment with no matching customer.
- Keep a log of every change, so a surprise later can be traced back to its cause.
Clean in the source copy or in a repeatable script, not by hand in the new system. A script can be run again on a fresh export when the cut-over date comes, and a manual fix cannot.
Why run test loads, and what should you check?
Load a copy of the data into a test version of the new system at least twice before the real move. Each load finds a new class of problem, and the second load shows whether your fixes worked.
- Counts: records in the source against records loaded, by type.
- Totals: any figure that adds up, such as balances or quantities, matches the source.
- Samples: a set of real records chosen by the people who know them, checked field by field.
- Relationships: each record is still linked to the right people, projects or accounts.
- Rejections: every record that failed to load is listed with the reason.
- Files: attachments open, and are attached to the right records.
How do you plan the cut-over and a rollback?
The cut-over is the moment the new system becomes the system of record. Choose a quiet time, tell everyone the date, and agree a freeze: after the final export, nobody edits the old system.
- Freeze the old system and take a final export and a backup.
- Run the load using the same script and mapping as the last test.
- Run the checks and have the owners sign off the results.
- Open the new system to users and make the old one read-only.
- Keep the old system and its backup available for an agreed period.
Decide the rollback before you need it. If the checks fail, what is the trigger for stopping, who decides, and how does the team carry on in the old system? A rollback plan you have never read through is not a plan.
What happens after the move?
Watch the first weeks closely. Expect questions about records that look different, and fix the pattern behind them rather than each case alone. Retire the old system only once nobody has needed it for a while and the retention decisions from your inventory have been carried out.
Our data migration service follows this sequence, starting with a paid discovery from £750 that produces a written scope for the migration. The field mapping and the load itself belong to the migration work. It can follow a system takeover, and a good example is replacing a case-tracking spreadsheet. When the old system came from another developer, the takeover checklist covers access and ownership first.
Common questions
Should we migrate everything?
Usually not. Move live and recent records, and archive the rest in a form you can still search. Less data moved means fewer errors and a shorter cut-over.
Can we keep working during the migration?
Yes until the freeze. Plan for one short period when nobody edits the old system, and tell people the dates in advance.
What if we find errors after cut-over?
Keep the old system read-only and its backup intact. Corrections are then made in the new system from a known source, and logged.