Every data migration requires some degree of reconciliation, to verify its success against pre-defined criteria. The types of reconciliation to be carried out may include any of the following checks:

• matching the numbers of records of a certain type between the source
  and target systems (assuming a 1:1 mapping between them)
• matching hash totals of chosen data items between source and target systems
• comparing financial balances between the source and target systems
• comparing financial balances between the target system(s) and the
  general ledger.

Reconciliation checks on particular data items are defined in the data dictionary.
Every such definition should include the tolerances within which any reconciliation differences are considered acceptable. These tolerances may be expressed as absolute values or percentages. For some reconciliations, such as comparisons of record counts, there should be "zero tolerance", since the loss of even one record during a migration may indicate a serious failure of the process. In other cases it may be necessary to set higher tolerances, to recognise that certain values may be
derived differently in the target system than in the source system.

Reconciliation checks and associated tolerances need to be defined as part of the planning process for any data migration, for two main purposes:

• so that precise criteria are available at every stage during the actual
  execution of the migration in order to decide whether to proceed to the
  next stage – in other words, to facilitate "Go / No Go" decisions
• to facilitate any subsequent audit of the migration.