Skip to content
custom software

Data migration strategy: a plan with owners, tests and a way back

A data migration strategy for mid-size companies: big bang or phased, the plan by phase, the checks that must match, cutover with a rollback point.

Two frosted glass database cylinders joined by a brass pipe, one full of light and one half full, with the headline Data migration strategy: steps, tests and cutover

A data migration strategy is the written plan for moving records from an old system or spreadsheets into new software without losing, doubling or changing any: what moves, in what order, who signs off each step, which tests must pass, and how you go back if cutover fails.

A migration is one stage of a larger build; see our guide to the custom software development process for the rest.

What should a data migration strategy decide?

Before anyone writes a load script, agree seven answers:

  1. Scope: which records move, which are archived and which stay behind.
  2. Approach: all at once, or in phases.
  3. Owners: a migration lead, and an owner for each kind of record.
  4. Tests: what must match before anyone says go.
  5. Downtime: how long the business can work without either system.
  6. The way back: what triggers a rollback, and the last moment it is possible.
  7. The old system: how long it stays readable, and how it is archived and wiped.

Whatever data migration methodology you follow, put these in writing first.

Should you move everything at once, or in phases?

The answer turns on downtime and on how tangled the records are.

  • Big bang. Everything moves in one cutover window, often a weekend. You rehearse one load and never keep two systems in step, but all the risk lands on one date, and the load, the checks and any rollback must fit the window.
  • Phased. Records move in waves, by entity, site or module. Each wave is smaller and teaches the next, but for a while work runs in both systems, and records that cross the line (a customer served by two sites) need a rule for which system owns them.

Agree the downtime first. The National Institute of Standards and Technology (NIST), in its contingency planning guide, defines maximum tolerable downtime as the total time the system owner "is willing to accept" for an outage. If a rehearsed load and its checks fit inside that with room to roll back, one cutover is usually simpler. If not, phase it.

If the old system itself is being replaced piece by piece, see our guide to legacy system modernization.

What goes in a data migration plan?

The plan turns the strategy into a data migration process: phases, each with a sign-off and a test to pass before the next.

Phase What happens Who signs off The test to move on
Inventory Every source listed, with record counts Migration lead Each source has an owner and a decision: move, archive or leave
Mapping Each old field gets a new field, a rule and a default Record owners No field left out without a written reason
Cleansing Duplicates merged, gaps filled, each fix written as a rule Record owners Remaining exceptions accepted in writing
Trial loads Full loads into a test copy, same scripts, timed Technical lead One full rehearsal runs clean inside the window
Reconciliation Counts, totals and samples compared, old against new Controller and record owners Every difference matched, or explained and signed
Cutover Freeze, backup, final load, checks, go or no-go Sponsor Go criteria met before the rollback point
Read-only period Old system readable, with no edits Migration lead An agreed date, such as the first month-end close
Retirement Records archived, access removed, media wiped Records owner and IT The archive opens; the wipe is recorded

Record owners, as an example: the controller for balances, the payroll lead for employees, sales operations for customers.

How do you map and clean the data?

Write the mapping as a table owners can read: old field, new field, rule, default and who decided. Three things catch teams out:

  • Code lists. Statuses, payment terms, units and cost centers rarely match one for one. Agree each translation, including codes nobody uses anymore.
  • History. Moving open items and opening balances is simpler; moving years of transactions keeps reports continuous but multiplies the checks.
  • Links between records. An invoice points to a customer, a job to a site. Load the parents first, and test that nothing points at a missing record.

Clean in the old system while it is in use, or in the load rules. The record owner decides each fix, and each becomes a rule that every rehearsal repeats. Moving off spreadsheets? Our guide to custom database development covers finding every copy first.

How do you run trial loads?

Load the full data set into a test copy of the new system, with the same scripts and order you will use on the day, and time it. Each run lists the records that failed and why; fix the rules, not the records by hand, and rerun until the list is empty or accepted.

Then hold a dress rehearsal: the whole cutover, checks and sign-offs included, against the clock. If it doesn't fit the agreed downtime, change the plan now.

Test copies hold real data, so protect them like the live system. NIST's catalog of security and privacy controls calls for minimizing personal information in testing, with "techniques such as de-identification or synthetic data."

What has to match before you cut over?

Agree the checks before the first rehearsal, and run the same ones every time:

  • Counts of each kind of record, by status: active customers, open orders, employees on payroll.
  • Control totals by entity: open receivables, open payables, the trial balance by account, stock value by location.
  • A sum that means nothing, such as the total of all invoice numbers. If it differs, a row was dropped or doubled.
  • Sample records the owners know well, compared field by field, plus awkward ones such as a credit balance.
  • The last transaction. NIST's contingency guide, as an example of checking data after a recovery, suggests a test to "see if the last known complete transaction was updated in the database." It works after a migration too.

A difference nobody can explain is a failure. Each one gets a reason and an owner's signature.

How do you cut over with a way back?

Write a runbook: every step in order, with a name and a time.

  1. Freeze the old system at an agreed time.
  2. Back it up in full and test the restore. The NIST catalog asks that backups be tested and a sample restored.
  3. Run the final load and the reconciliation checks.
  4. Hold the go or no-go call. Each record owner confirms their checks; the sponsor decides.
  5. Open the new system to users.
  6. Back up the new system in full once it is live, as NIST's contingency guide advises after a recovery.

The rollback point is the moment after which going back would mean retyping work done in the new system. Before it, a failed check means reopening the old system. Decide in advance which failures trigger a rollback.

What happens to the old system after cutover?

Keep it read-only for a set time: people will look up old notes, and auditors may ask how balances were carried over. NIST's contingency guide says to keep the old system's contingency plan ready until the new system is "operational and fully tested," and notes that a replaced system can be a fallback "if a loss or failure of the new information system should occur." Put the end date in the plan, such as after the first quarter-end close.

Then retire it in order:

  1. Archive what you must keep, in a format you can read without the old software. The NIST catalog notes that retention requirements can extend "beyond system disposal." The Internal Revenue Service, for example, says to keep employment tax records "for at least 4 years after the date that the tax becomes due or is paid, whichever is later," and to check other obligations before discarding records.
  2. Remove access: user accounts, connections and scheduled exports.
  3. Wipe the media. NIST's guidelines for media sanitization describe three methods (clear, purge and destroy) and a certificate recording what was sanitized, how and by whom. For cloud storage, the guide says cryptographic erase may be the only workable way to purge. Check retention first, so the wipe doesn't destroy records that must be kept.
  4. Delete the migration's own copies: extract files, staging databases and test copies. The NIST catalog says disposal applies "to originals as well as copies and archived records."

How does Raileon handle data migration?

We build custom software around how your company already runs, then run it for you. That includes moving a spreadsheet-and-email process, or an aging tool, onto something your team can rely on. If an existing product fits what you need, we'll tell you on the first call.

It starts with a free 20-minute call with the founder. A written proposal sets out the scope, the approval rules and the exact price before anything starts. You review working screens as they come together, then we train your people.

After launch, we host, monitor, update and back up what we built, and a failed step or an expired connection alerts your Raileon team. The build is priced by scope. Monthly care starts from $1,999 a month. Cancel any time. Your plan runs to the end of the month you've paid for. No exit fee.

Every client's data is kept separate, encrypted in transit and at rest, and backed up daily. Our terms are plain: Raileon owns the code it builds, and your data stays yours.

FAQ

What is data migration?

Data migration is moving records from one system or format to another, such as from spreadsheets into new software, and proving they arrived complete and correct. It ends when the old system can be retired, not when the load finishes.

What is the difference between a data migration strategy and a data migration plan?

The strategy sets the decisions: scope, approach, owners, tests, downtime and the way back. The plan turns them into dated phases, each with a sign-off and a test that must pass.

How long does a data migration take?

It depends on the number of sources, the volume and how clean the records are. Trial loads show how long the final load takes; plan time for mapping, cleansing and rehearsals too.

Should we move all of our historical data?

Not always. Open records and opening balances must move; closed history can often be archived in a readable form instead, kept for as long as your retention rules require.

Which system are you moving off?

Bring the list of systems and spreadsheets your records live in today. On a free 20-minute call, we'll tell you what we'd build first.

Book a free 20-minute call

About the author

Konstantin, Founder & CEO of Raileon

Konstantin

Founder & CEO, Raileon

Konstantin has spent more than 15 years in systems, processes, automation and development. He has built businesses from scratch, from early crypto to tech ventures that led to acquisitions, and founded Raileon to build the software and automation he wished he'd had.

  • 15+ years in IT, automation and development
  • Founder & CEO of Raileon
  • Based in North Miami, Florida

Let's put your operations on rails

Book a 20-minute call with our team in Miami, Florida, and we'll show you exactly what Raileon would automate for your business.

Book a 20-minute call

No commitment. No credit card. Just a conversation.