StackLift
All playbooks

Playbook

Data Migration Plan

Most migration failures come from data nobody checked and a cutover nobody rehearsed. This plan covers the phases, the checks, and the rollback gate.

Best reader

Teams moving data between systems during a rebuild, replatform, or modernization

Outcome

A rehearsed migration with source-to-target mapping, reconciliation reports, and a tested way back.

Use this sequence

1

Inventory source systems, owners, and what must move.

2

Profile the data to find gaps, duplicates, and bad formats.

3

Write the source-to-target mapping with every transformation rule.

4

Build the migration scripts and run rehearsals on a copy.

5

Reconcile counts, totals, and samples after each run.

6

Freeze changes, cut over, and verify with business users.

7

Keep the old system available until the rollback window closes.

Profile and map before you move anything

Bad data moved faithfully is still bad data. Profile first, decide what to clean, archive, or leave behind, and record each rule in a mapping that lives in version control.

Inventory

Profiling results

Mapping document

Cleanup decisions

Rehearse and reconcile

Run the whole migration more than once on a copy. After each run compare record counts, key totals, and sampled records at table, record, and field level, and fix the differences before the real run.

Rehearsals

Reconciliation reports

Error handling

Timing of the cutover steps

Cutover and a rollback you have tried

Decide the conditions that stop the cutover and the steps that restore the old system. A rollback plan that has never been run is a hope, not a plan.

Go or no-go gate

Freeze window

Rollback steps

Post-cutover checks

Start in 60 seconds

Turn this checklist into a scoped plan.

Answer a few questions and get an AI-assisted, architect-reviewed scope, cost range, and timeline for your software project.

Frequently asked questions

Big-bang or phased migration?

A big-bang cutover moves everything at once and needs a strong rollback. A phased approach lowers the risk at each stage but takes more coordination and means two systems run for a while.

Who should sign off on migrated data?

The business owners of that data, not only the engineers. They know which records and totals must be right, and they should confirm it before the old system is retired.

The complete guide

Data Migration Plan: the complete guide

Updated November 18, 2026 ยท Published by , Founder, StackLift AI. Written with AI assistance; how we make content.

Short answer

A data migration plan moves data from an old system to a new one in controlled phases: inventory and profile the data, map every field and transformation rule, rehearse the load on a copy, reconcile counts and totals after each run, then cut over with a go or no-go gate and a rollback you have already tried.

What is a data migration plan?

A data migration plan is the document and schedule that govern moving data from a source system to a target system. It names what moves, how it is changed along the way, who owns it, how you will prove it arrived correctly, and how you will go back if it did not.

Migrations fail in two common ways. Teams move data nobody checked, so errors from the old system appear in the new one. Or they plan a cutover they never rehearsed, so a step fails at the worst moment with no fallback. The plan below addresses both. It fits a rebuild, a replatform, or a legacy modernization.

What are the phases of a data migration?

Phases overlap, and what you learn late feeds back into earlier ones. Profiling results change the mapping, and rehearsals change the runbook. Keep mappings, scripts, and the runbook under version control so each rehearsal uses the latest.

Migration phases and outputs
PhaseWhat happensOutputWho is involved
PlanSet goals, scope, owners, success measures, and the approachMigration plan and risk listProject lead, data owners
Inventory and profileFind sources, volumes, duplicates, gaps, and bad formatsProfiling report and cleanup decisionsEngineers, data owners
Design and mapDefine source-to-target fields and transformation rulesMapping documentEngineers, analysts
BuildWrite extract, transform, and load scripts, plus checksTested scripts and reconciliation queriesEngineers
RehearseRun the full migration on a copy and fix what failsRehearsal reports and a timed runbookEngineers, QA
ValidateBusiness users check the migrated dataSigned validation resultsBusiness users, QA
Cut overFreeze changes, load final data, switch usersLive system and a go or no-go recordEveryone
StabilizeMonitor, fix, keep rollback open, then retire the old systemClosure reportEngineers, support

How do you write the plan, step by step?

  1. List every source system, database, file store, and spreadsheet that holds data the new system needs. Output: an inventory with an owner for each.
  2. Decide what moves, what is archived, and what is left behind. Output: a scope list with reasons.
  3. Profile each source for duplicates, missing values, odd formats, and broken relationships. Output: a data quality report.
  4. Agree who cleans what and when. Output: cleanup tasks assigned before the build.
  5. Write the source-to-target mapping with each transformation rule. Output: a mapping document that engineers and business owners both review.
  6. Choose the approach: big bang, phased, or parallel run. Output: a recorded choice with its rollback.
  7. Build the scripts and the reconciliation checks together. Output: a migration you can rerun and measure.
  8. Rehearse at least twice on production-like data and time every step. Output: a runbook with real timings.
  9. Set the cutover gate and rollback conditions. Output: a written go or no-go checklist.
  10. Cut over, verify, and keep the old system until the rollback window ends. Output: a signed closure.

Start the inventory early, even before the new system is designed. What the old data looks like often changes what the new system needs to store.

What goes into a source-to-target mapping?

The mapping is the heart of the plan. For every field in the new system, it says where the value comes from and what is done to it. Business owners should read it too, because they can spot a rule that is technically correct and wrong for the business.

Mapping document columns (example rows)
Target fieldSource fieldRuleCheck
customer.full_nameCUST.FNAME + CUST.LNAMEJoin with a space, trim, fix casingNo empty names after load
customer.statusCUST.STAT codeMap old codes to new statuses with a lookup tableEvery old code has a mapped value
order.totalORD.AMTConvert to the standard currency unit; keep original in a note fieldTotals match by period
order.created_atORD.DT (text)Parse as a date; flag unparsable valuesCount of flagged rows reviewed
invoice.customer_idINV.CUST_NOLook up the new customer by legacy keyNo orphaned invoices

Keep a legacy key on every migrated record. It lets you trace any value back to its source, and it makes reconciliation and later questions much easier.

How do you handle bad or messy data?

Moving bad data faithfully gives you bad data in a new system. Profile first, then decide for each problem whether to fix it at the source, fix it in the transformation, load it with a flag, or leave it behind. Record each decision so the owners can see what was changed.

  • Duplicates: define the matching rule, merge or flag them, and agree who decides on conflicts.
  • Missing required values: supply a default, fetch from another source, or block the record for review.
  • Inconsistent formats: standardize dates, phone numbers, currencies, and addresses in the transformation.
  • Orphaned records: records pointing at things that no longer exist, to archive or repair.
  • Obsolete data: records past a retention need, which may be archived instead of moved.
  • Sensitive data: confirm what may be moved, masked, or removed, and who can see test copies.

Use masked or synthetic copies for development where your rules require it, and limit who can see real data during rehearsals.

How do you prove the migrated data is correct?

Reconciliation compares source and target at several levels and explains every difference. Automate it so it runs after each rehearsal and again just before cutover. A migration without reconciliation relies on people noticing problems later, usually in front of customers.

Reconciliation levels
LevelCheckExample
Table or objectRecord counts match, or differences are explainedCustomers in the source against customers in the target
TotalsKey sums and balances matchOrder totals by month; account balances
RecordSampled records match field by fieldA chosen set of customers checked end to end
RelationshipsLinks between records are intactEvery invoice points to a customer
BusinessUsers confirm that reports and screens look rightMonth-end report agrees with the old one
RejectsRows that failed are counted, listed, and resolvedA reject file with reasons

The business owners of the data should sign off on the results. Pair this with a user acceptance testing plan so real users try real workflows on migrated data.

How do you plan cutover and rollback?

A big-bang cutover moves everything in one window and needs a strong rollback. A phased cutover moves groups of users or data in stages, which lowers the risk at each step but means two systems run for a while and need to stay consistent. A parallel run keeps both live for a time and compares results. Choose based on how much downtime and risk the business can accept.

Cutover approaches compared
ApproachStrengthRisk
Big bangOne switch, no dual runningAll-or-nothing; needs a proven rollback
PhasedLower risk per stageMore coordination; data in two places
Parallel runCompare old and new on live workDouble effort and a longer period
  • Set the freeze: when users stop changing the old system, and how late changes are captured.
  • Write the go or no-go checklist with named owners and the conditions that stop the cutover.
  • Write the rollback steps, and rehearse them. A rollback no one has tried is a hope.
  • Keep the old system readable, and ideally restorable, until the rollback window closes.
  • Plan communication: who tells users, when, and how they report problems.

The controls overlap with the production readiness checklist, including backups, monitoring, and an incident owner.

What are the common migration risks and how do you reduce them?

Risks and responses
RiskResultResponse
Unprofiled source dataErrors surface after launchProfile before mapping
Mapping rules only engineers seeRules that are wrong for the businessBusiness owners review the mapping
No rehearsalSurprises during cutoverRehearse the full run with timings
Untested rollbackNo safe way backRehearse rollback on a copy
Changes during the freezeLost or inconsistent recordsFreeze rules and a way to capture late changes
Integrations forgottenOther systems break when data movesInventory every consumer of the old data
Sensitive data exposed in test copiesPrivacy and compliance troubleMask or limit access

Integrations are a common blind spot. Reports, spreadsheets, and other systems often read from the old database. Check the system integration glossary entry for the wider picture.

When should you bring in help?

Bring in help when the data is large or messy, when nobody on your team has done a cutover, when several systems feed into one, or when the data is regulated. A migration is easier to plan at the start of a rebuild than to rescue near the end.

StackLift plans data migration as part of modernization and rebuild work, with CI/CD, observability, security review, and rollback in the plan from sprint one. Read the legacy modernization roadmap for the wider plan, see the legacy modernization cost range, or scope your own project with the free estimate or by contacting us.

Questions people ask

What are the steps of a data migration?

Inventory the sources, profile the data, map every field and rule, build and rehearse the migration, reconcile the results, cut over with a go or no-go gate, then keep the old system until the rollback window ends.

How do you validate migrated data?

Compare source and target at table, totals, record, and relationship levels, review rejected rows, and have business owners check reports and sample records. Automate the checks and rerun them before cutover.

What is the difference between big-bang and phased migration?

A big-bang migration switches everything at once and needs a strong rollback. A phased migration moves groups in stages, which lowers risk per stage but needs coordination while two systems coexist.

How many rehearsals should you run?

At least two full rehearsals on production-like data, and more if the data or steps are complex. Each run should be timed and reconciled so the cutover runbook reflects real durations.

Who should own data migration?

A named project lead, with engineers building the migration and business owners accountable for the data and its sign-off. Shared ownership with no names leads to gaps.

Should you clean data before or during migration?

Decide per problem. Fix at the source when the old system will stay in use, fix in the transformation for consistent rules, and flag records that need human review. Record each decision.

How long does a data migration take?

It depends on the number of sources, data quality, and complexity of the mapping. Most first versions of a new system take 6 to 16 weeks depending on scope, and the migration is part of that plan. Get a scoped view through an estimate.

Sources

Sources and further reading

The standards, rules, research and official documentation this page draws on.