How to Plan Data Migration When Replacing Business Software
Replacing business software needs a data migration checklist built around the records and relationships the replacement must preserve. Decide what to move, map and clean it, rehearse the import, then reconcile the results before switching daily work. Include attachments, historical records and changes made during the move. Define rollback around the point when the replacement starts creating business activity that the old system cannot safely recover.
What must the replacement system preserve?
Start with business work, then identify the records behind it. Ask what someone needs to fulfil an open order, resolve a complaint or explain an earlier transaction.
For each record type, document its owner, unique identifier, related records, attachments and required history. Classify it as active, historical, duplicate or excluded.
| Record type | What to preserve | Acceptance check |
|---|---|---|
| Customer | Identity, addresses and account status | Open work belongs to the correct customer |
| Order | Items, quantities, status and customer link | A user can continue processing it |
| Product | Identifier, variants and units | Order items resolve to the intended product |
| Document | File, description and parent record | An authorised user can open it |
| Activity history | Event, sequence and original actor where available | A user can understand what happened |
Choose representative business journeys as acceptance criteria. “Continue an open order with its supporting documents” is more useful than “import the order table”.
How should you map fields and relationships?
Create a mapping sheet before writing import code. For each source field, record the destination, conversion rule, treatment of empty values and validation rule.
Include fields with no destination. Decide whether to add support, archive the value or deliberately exclude it. Do not silently discard information because it does not fit.
Illustrative mapping example:
| Source | Destination | Rule |
|---|---|---|
| Customer reference | External reference | Preserve exactly, including leading zeros |
| Customer reference on an order | Customer link | Resolve through the reference mapping |
| Status: “Awaiting dispatch” | Status: “Ready to ship” | Confirm both mean the same business state |
| Blank account owner | Unassigned owner | Keep unassigned unless an assignment rule is approved |
| Document reference | Attachment link | Resolve to the imported file and its parent |
Keep a lookup between old and new identifiers. Reject unresolved relationships for review rather than guessing from names.
Review status meanings carefully. Similar labels do not establish that two systems allow the same next actions. Resolve those differences with the people who use the workflow.
What should you clean before importing?
Agree cleanup rules with a business owner and preserve an untouched export before applying them.
Look for duplicate customers, missing required values, inconsistent units, invalid references and conflicting statuses. Give every exception a decision: correct it, merge it, exclude it with a reason or block the affected import.
Do not merge records solely because their names match. Check stable identifiers and related activity. If records are merged, record which identifier survives and how their relationships will be reassigned.
Separate formatting changes from changes in meaning. Removing unwanted spaces is different from deciding that two accounts represent the same customer.
Apply agreed transformations through a repeatable process. Keep corrections traceable so a later import does not depend on someone remembering which spreadsheet cells they edited.
How should you handle attachments and historical records?
Treat attachments as migration items with their own inventory. Record each file’s parent record, name, type, access requirements and import result.
Test whether an authorised user can open the imported file from the correct record. Also test that a user without permission cannot open it. Review failed transfers and files without a parent separately. Plan access to migration copies alongside access in the replacement; data security in modern applications provides related context.
For history, choose among three explicit treatments:
- Move into the replacement: history needed for daily work.
- Keep in a searchable archive: history needed occasionally.
- Exclude under an agreed policy: information no longer required.
Define who can search the archive, what they can retrieve and who maintains access. Test retrieval before ending access to the old software.
Where supported, preserve original event times and actors separately from import details. Avoid making an old event appear to have occurred when it was imported.
What should trial imports prove?
Use a small, varied trial first: an open order, a closed order, a duplicate account, an attachment and an incomplete record. Follow it with a full rehearsal using the intended migration process.
A trial should test both the data and the business journeys. It should also measure the elapsed time needed for export, transfer, import, validation and exception handling.
Include these checks:
- Required fields and relationships survive.
- Rejected records have understandable reasons.
- Users can complete agreed workflows.
- Access permissions behave as intended.
- A failed batch can be resumed or safely repeated.
- Repeating an import does not create unintended duplicates.
- Notifications and connected actions stay controlled during testing.
Fix failures and repeat the affected checks. Keep the final rehearsal close to the planned cutover procedure. The broader approach belongs in your software testing and quality assurance plan.
How do you reconcile the imported data?
Compare results against the same source snapshot. Record later changes separately so they do not become unexplained differences.
Use several checks together. Counts test completeness, totals test selected values, relationship checks test connections and workflow checks test usability.
Illustrative reconciliation example: an export contains 800 orders. If 790 import and 10 are rejected, account for all 800 before approval. Review each rejection. Also compare order values by status and inspect customer links; matching the overall count alone does not establish that the orders are correct.
For each record type, record:
- Source, imported, rejected and deliberately excluded counts.
- Relevant totals, grouped by meaningful business categories.
- Missing relationships and attachment failures.
- Results of representative workflow checks.
- Unresolved differences, their impact and the approval decision.
Define acceptable differences before cutover. Require an explanation for each accepted difference rather than allowing a general tolerance to hide unknown errors.
How should cutover and rollback work?
Write a timed cutover sequence with a named owner for each action. Include the final export, handling of changes since rehearsal, import, reconciliation, approval and opening the replacement to users.
Define when the old system stops accepting updates. Include background activity: integrations, scheduled jobs and incoming requests may also need to pause or queue. Assign responsibility for restoring each connection.
Set go or no-go conditions in advance. Unresolved record links, unexplained totals or inaccessible essential documents should have an explicit decision rule.
Define the rollback boundary by events, not just a clock time. Before new business activity begins, the plan may allow the team to discard the replacement’s imported state and resume the old system.
Once the replacement accepts orders, edits, payments or external messages, returning needs a way to recover that activity without loss or duplication. A backup restores an earlier state; it does not account for later work.
Identify the earliest event that makes return difficult, who authorises crossing it and how subsequent changes would be recovered. Rehearse the return procedure before relying on it.
What common mistakes should you avoid?
Treating a successful upload as acceptance. Require reconciliation and business workflow checks before approval.
Moving everything without a purpose. Give historical information a deliberate destination and retrieval plan.
Leaving exceptions until cutover. Resolve known mapping gaps and rejected records during rehearsals.
Keeping both systems writable without ownership rules. Specify which system owns each record and how competing changes are resolved.
Promising rollback without a recovery procedure. Show how new activity would return to the old system.
A simple contact-list move may need only a short mapping sheet and import check. A replacement involving linked transactions, documents and connected services needs the fuller plan.
What questions come up most often?
Should all historical data move?
Base that decision on operational needs, retention requirements and retrieval needs. Test any archive before relying on it.
Who should approve the migration?
Name technical and business approvers. Ask them to review reconciliation results, workflow checks and unresolved exceptions against the agreed acceptance criteria.
When is the migration complete?
Define completion as reconciled data, working business journeys, restored connections and assigned ownership of remaining exceptions. Retiring the old system needs its own access and archive checks.
What should a software partner provide?
Ask for the mapping sheet, exception list, rehearsal results, reconciliation report and cutover procedure. Each should show what was tested and what still needs a decision.
Once the replacement is live, see how to safely decommission the old software.
*Planning to replace business software? Discuss your migration requirements →*