How to migrate data to a new system with checks and recovery

Migration transfers information and the meaning that makes it usable. Importing a file without errors does not demonstrate that customers, requests, statuses and values are correct. Prepare a rehearsed migration with reconciliation criteria, exception handling and a clear decision on when to begin working in the new system.

By Lyntev. Updated on .
About 5 minutes to read.

A request that stays connectedThe same reference follows the request and prevents duplicates.

Translation published on . Original Portuguese guide; source revision .

By the end, you can:

  • Define scope and mapping
  • Check totals and relationships
  • Prepare transition and recovery

Decide what needs to move across

Inventory systems, spreadsheets, attachments and owners. Separate configuration data, master records and ongoing operations. Not all history needs to enter the new application: some may remain in an accessible archive with defined retention. The decision must consider actual use and applicable obligations in context.

Microsoft’s migration guidance distinguishes these datasets and recommends planning dependencies, transformation and validation. In a fictional example, Oficina Norte migrates active customers and open requests; closed documents remain in a searchable archive. The team checks how to find that archive and avoids calling migration complete while elements needed for work remain missing.

Decide what needs to move across
DatasetFictional decisionCheck
Active customersImport into the new systemIdentity and contact details
Open requestsImport with relationships and statusOwner and next action
Necessary attachmentsTransfer and link to the requestOpening and correspondence
Closed historyKeep the defined archiveSearch and authorised access

Write transformations before executing them

For each field, state source, destination, type and transformation. Record what happens to unknown values, old codes and empty fields. Do not guess when converting an ambiguous date: 03/04 may mean different days depending on format. Confirm the source convention and use an unambiguous representation in the working file.

PostgreSQL’s COPY reference makes format, delimiter, encoding and null-value options explicit. Confirm these decisions even with another importer. A code with leading zeros must remain text where those zeros are part of its identity. Decimal values, accents and line breaks need examples in the rehearsal.

Fictional status mappingtext
Source «Novo»       = destination «recebido»
Source «Em curso»   = destination «em_execucao»
Source «Aguardar»   = review: the reason still needs identifying
Empty source       = exception; do not assume «concluido»

cliente_antigo_id = retain as a source reference
cliente_novo_id   = identifier assigned at the destination

Preserve identifiers and respect dependencies

Load elements needed by subsequent ones first. Customers before requests, requests before activities, according to the chosen model. Keep a mapping of old to new identifiers to link each record to the correct destination. A name or spreadsheet position is not a safe substitute for this mapping.

Foreign keys documented by PostgreSQL are an example of a mechanism maintaining valid relationships. Request an exception list for unmatched records instead of approximate associations. Then follow a request end to end: customer, owner, status, history and attachments must still describe the same case.

Run a rehearsal repeatable without duplication

Use a separate environment and a sample including common and difficult cases: similar names, empty fields, attachments, old dates and open statuses. Record file and rule versions. A second run on the same sample must have a defined result: update, ignore or replace the test dataset under control.

Do not use an experimental run to send emails, generate charges or change real integrations. Identify effects an import may trigger and disable them for the rehearsal through appropriate configuration. Also measure duration at a representative volume; importing ten rows cannot confidently estimate a much larger migration.

Reconcile totals, statuses and relationships

Compare the source-selected quantity with imported records, approved exclusions and pending exceptions. Explain every difference. In a fictional example, 500 selected customers produce 480 imports and 20 held for review. The total reconciles, but does not yet prove those 480 identities are correct.

Add group checks, relevant values and individual cases. Count requests by status and owner; confirm attachments open; compare transformed fields against the approved rule. An overall sum may hide errors cancelling each other out. Ask operational users to find and continue some cases at the destination.

Reconcile totals, statuses and relationships
CheckWhat it may reveal
Count by entityOmissions or duplicates.
Requests by statusIncorrect code mapping.
Unmatched relationshipsMissing customer or owner.
Sample of dates and valuesIncorrect format, time zone or unit.
Opening attachmentsIncomplete transfer or broken link.

Define the switch point and what to do if it fails

Choose a transition window and define how changes after the first extraction are handled. There may be a write freeze or controlled final synchronisation. The team needs to know from when work is recorded at the destination. Two editable systems without reconciliation rules produce divergent versions.

Prepare go/no-go criteria, responsibilities and communication. Keep a verified source copy and a recovery procedure. If new work has already reached the destination, returning to the source requires preserving and reconciling those changes; restoring an old copy without this analysis may lose work performed after the switch.

  • Confirm access and configuration before the final import.
  • Record the cut-off point and last included change.
  • Check critical journeys after transition.
  • Keep support and responsible people available during launch.

Close migration with evidence and responsibilities

Approval must state what was migrated, what remains archived and which exceptions were accepted. Keep reconciliation results and the location of protected copies. Disable the source only after confirming use, recovery and retention conditions. Define who handles problems found during the first days.

Worksheet for your project

Migration worksheet
Source, destination and owners:
Included and excluded entities and archive:
Extraction version and cut-off point:
Mapping of fields, statuses and identifiers:
Import order and dependencies:
Handling blanks, formats and duplicates:
Sample, environment and rehearsal duration:
Source, imported and exception counts:
Checks of relationships, values and attachments:
Handling changes after extraction:
Criteria to proceed, postpone or recover:
How to preserve new work during recovery:
Approval, launch support and date:

Sources and criteria

Content prepared with AI assistance and research using technical sources, with checks of the relevant examples and calculations. The references support the guide’s decisions; examples identified as fictitious let you try out the method.

Wellinton Oliveira

Written by

CEO of Lyntev

Technology with context, original analysis and sources you can consult.