Back to all articlesCRM migration

Choose Between a CRM Write Freeze and a Final Delta Import

Alactic Editorial10 min read

Prepared with AI assistance by Alactic Editorial. Examples identified as hypothetical are illustrative, not customer results.

Your initial CRM export is already being loaded, but salespeople are still editing opportunities, new leads are arriving, and someone has deleted a duplicate contact. Unless those changes have an agreed route into the destination, a successful import can still leave the new CRM wrong.

Choose a controlled write freeze when you can tolerate the interruption and cannot reliably identify every relevant change. Choose a final delta import when you can capture additions, edits, and deletions, apply them repeatably, and validate them within a shorter closing window. A delta usually reduces the interruption; it does not remove the need for a final boundary after which the old CRM stops accepting authoritative changes.

The decision is not simply whether users can keep working. It is whether your team can demonstrate what changed, where those changes went, and when the destination became the system of record.

1. Define the two cutover options precisely

With a controlled write freeze, the team stops relevant changes in the source, waits for in-flight work to finish, takes a stable final extract, imports it, validates it, and opens the destination for work. The final extract may replace a rehearsal load or use another tested loading method. Do not assume it is safe to append a second full export to existing destination records.

With a final delta approach, the team loads a baseline while the source remains active. It then identifies and applies the changes since that baseline. At the end, a shorter closing pause or another proven handoff mechanism establishes the final source boundary. This article uses a closing pause because it does not assume reliable live change tracking exists.

The difference is what must fit inside the interruption:

  • Full freeze: final extraction, loading, and validation of the agreed scope.
  • Final delta: closing change capture, application, and validation, after the baseline is already loaded.

Neither option is complete if people, imports, integrations, or background jobs can keep changing the source unnoticed. Define the freeze by the writes it stops, not just by the announcement sent to users.

2. Establish what your extracts can actually prove

Before estimating downtime, test the evidence available from the source. A visible “last modified” field is not, by itself, a complete change-capture method.

  • Additions: Can you identify new records with stable source IDs, including records created during an export?
  • Edits: Does the change indicator cover every migrated field and relationship? Check owner changes, associations, cleared values, and relevant custom fields.
  • Deletions: Can you obtain deleted IDs, or compare complete ID inventories taken at known boundaries? A missing row in a filtered export is not sufficient evidence of deletion.
  • Extraction consistency: Does an export represent a fixed snapshot, or can different pages reflect different moments?
  • Repeatability: Can the destination recognize an already-applied source record or operation without creating another copy?

Test these questions with deliberately created, edited, and deleted test records in an approved environment. Include a record changed while extraction is running. Record the result rather than relying on an export label.

If the source cannot expose deletions, a stable final inventory comparison may be an alternative. If neither is available, the delta has an unresolved gap. Choose a stable full-state reconciliation under a freeze, reduce the migration scope with explicit approval, or postpone the cutover. Do not silently treat “not detected” as “did not happen.”

3. Use a decision table that includes validation time

Fictional worked example: A team has already loaded a baseline into a destination that is not yet open for normal work. All quantities, times, and test outcomes below are hypothetical, not product capabilities or customer results.

The team can tolerate a 30-minute interruption. Its rehearsal suggests a stable full extract, load, and validation would take 90 minutes. A final delta could fit within 25 minutes, but only if its deletion evidence and retry behavior pass a further test.

Hypothetical filled decision table
Factor                 Evidence in this scenario             Decision effect
Change volume          240 additions, 610 existing-record    Delta appears manageable;
                       edits, 18 deletions since baseline     volume alone is not approval
Extract reliability    Stable baseline boundary verified;    Delta remains conditional on
                       final reread possible under pause      deletion and coverage tests
Interruption tolerance 30 minutes approved; full route        Full freeze exceeds the
                       rehearsed at 90 minutes                current allowance
Validation capacity    Data lead plus business reviewer;      Delta can fit only with
                       5 minutes reserved for closing checks prepared reconciliation
Retry safety           Source-ID mapping tested; one          Resolve failed-operation
                       interrupted batch test outstanding     retry behavior before go
Deletion handling      Complete ID comparison available       Verify comparison uses the
                                                             same scope and permissions
Decision               Conditional final delta with a         If tests fail, renegotiate
                       25-minute closing pause                interruption or defer

The 25-minute plan allocates 12 minutes to extraction and comparison, eight to application, and five to closing validation. Those are planning assumptions to verify in rehearsal, not a performance promise. If the team cannot complete meaningful validation in five minutes, the plan must change.

Change volume also includes complexity. A small number of company merges or relationship changes may require more work than many simple field edits. Estimate the delta by operation type and dependency, not just row count.

4. Fill the change-capture worksheet before cutover

The worksheet below continues the same hypothetical scenario. The counts are deliberately separated: the 610 edits concern existing baseline records, not the 240 additions. The 18 deletions concern other baseline records. Real batches need equivalent classification so summary totals are not misleading.

Hypothetical change-capture worksheet
Scope: contacts, companies, opportunities, and agreed relationships
Authority: source until release; destination after signed handoff
Destination: normal user writes disabled until release
Identity: source object type + source ID mapped to destination ID
Baseline T0: 09:00 UTC, verified fixed snapshot boundary
Closing target: 17:00 UTC; actual boundary recorded after writes drain
Capture rule: changes after T0 through actual closing boundary
Timestamp precision: seconds; UTC normalization checked in rehearsal
Boundary safeguard: reread from T0 inclusively; deduplicate by identity

Operation       Evidence and treatment                 Expected reconciliation
Additions       240 new source IDs; load their          Each expected ID has exactly
                final state at closing boundary         one destination mapping
Edits           610 baseline IDs; apply final           Compare migrated fields,
                state, including explicit clears        owners, and relationships
Deletions       18 baseline IDs absent from a           Apply approved deletion or
                complete stable final inventory         exclusion treatment; verify
Timestamps      Preserve source time in audit file;     No timezone conversion or
                record extraction and apply times       precision gaps at boundaries
Cutoff          Stop writers at 17:00; confirm          Replace target cutoff with
                in-flight writes have completed         actual settled boundary
During extract  Contact C-184 changed at 16:46          Final stable reread includes
                during a 16:40–16:50 staging export      its final state, not stale row
Retry control   Batch ID, source ID, operation,          Replaying completed work
                outcome, and error recorded              creates no extra records
Exceptions      Owner missing, broken relationship,     Named reviewer resolves or
                conflicting state, failed operation     blocks release

The staging export is not the closing delta. Because it may contain mixed-time records, the team reruns the relevant capture against the stable source after the pause. The final comparison establishes the expected state; it does not assume the earlier export saw every late edit.

If your evidence is an operation log rather than state comparison, document its ordering, completeness, and deletion coverage separately. A record created and then deleted between snapshots may never belong in the destination’s final state, but any requirement to retain its history needs its own verified capture method.

5. Close the timestamp boundary without leaving a gap

Write down what the baseline represents. “Export started at 09:00” is not equivalent to “all records reflect their state at 09:00.” If the export is not a fixed snapshot, document its coverage and use an overlap or stable reread capable of recovering changes made during extraction.

For a timestamp-based delta, specify the timezone, precision, comparison operator, and upper boundary. Records sharing the boundary timestamp must not fall between batches. An inclusive overlap can help, provided duplicate application is safe; it cannot repair a field that fails to update its timestamp.

A practical closing sequence is:

  1. Stop or redirect every identified source writer, including scheduled imports and integrations.
  2. Confirm in-flight work has drained. Record the actual settled boundary rather than assuming the announced time was exact.
  3. Take the final stable extract, change set, or inventory comparison using the tested method.
  4. Apply changes using stable identities and the recorded operation outcomes.
  5. Validate before authorizing destination writes and redirecting approved workflows.

If someone changes the source after the recorded boundary but before release, keep the destination closed. Investigate the writer, restore the pause, and rerun the closing capture and affected checks. Do not carry an unexplained late change forward as a manual reminder.

6. Make application safe for retries and dependencies

A delta import is not just a smaller insert file. New records, changed records, removed relationships, cleared fields, and deletions require explicit treatment.

Match by stable source identity rather than assuming an email address or company name is unique and unchanged. Maintain a source-to-destination mapping for relationships. Load prerequisites before dependent records, and validate associations after loading. An empty value must have a defined meaning: leave unchanged, clear the field, or reject the row.

Decide how deletions will appear in the destination before running them. Depending on the approved migration design, the treatment might be removal, archival, or exclusion from the final migrated scope. Do not turn every missing export row into a delete instruction.

Test an interrupted batch. Resume it, then check whether completed operations are skipped or safely reapplied. The observable outcome should be no extra records, no missing updates, and an operation ledger that distinguishes success, failure, and work not attempted. This is a workflow requirement to verify in your tools, not an assumption that an import screen provides it.

7. Set release checks and a recovery decision

Counts are useful but insufficient. One duplicate and one missing record can cancel each other out in a total. Validate identity sets as well as quantities.

  • Additions: Every expected source ID has one intended destination record.
  • Edits: Compare the migrated fields for changed IDs, including cleared values and relationship changes.
  • Deletions: Confirm each approved treatment and inspect unexpected inventory differences.
  • Boundary cases: Check records changed during extraction and records sharing cutoff timestamps.
  • Business use: Have a designated reviewer inspect agreed critical records and workflows before release.

Define blockers in advance: unexplained ID differences, unresolved critical fields, failed dependencies, or evidence that the source was still changing. Assign a person authorized to stop the release.

If checks fail before destination users begin work, keep it closed and retry from the recorded batch or restore the tested pre-cutover state. If the interruption allowance expires, use the agreed abort procedure to resume the source; a later attempt needs a new closing capture.

Once destination writes begin, reopening the old source creates two competing histories. Recovery then requires identifying and preserving destination-side changes before any reversal. Avoid treating a simple switch back as a complete rollback plan.

8. Make the decision with one timed rehearsal

For the fictional team, the final delta remains the preferred route only if the deletion comparison, interrupted-batch retry, and complete closing validation pass within the approved window. Otherwise, it needs a longer freeze or a later cutover—not a less rigorous sign-off.

Your next action is to assign a migration owner and fill the worksheet using one representative rehearsal. Record four results: change volume by operation, unexplained identity differences, end-to-end closing time, and unresolved validation exceptions. Then have the operational owner approve either the measured interruption or the decision to defer. That produces a cutover choice grounded in evidence rather than an optimistic calendar slot.

CRM migrationData migrationCutover planningData validation

Put the next conversation to work.

Explore Alactic Communications for WhatsApp conversations, campaigns and AI-assisted workflows. Choose the product and plan that fit your team.

Explore Communications