Run a CRM Cutover Rehearsal That Produces a Go or No-Go Decision
Prepared with AI assistance by Alactic Editorial. Examples identified as hypothetical are illustrative, not customer results.
A successful import does not prove that your CRM migration fits its cutover window. Extraction may take longer after writes stop, transformation may depend on a missing reference file, and validation may discover a defect after the time reserved for recovery has already disappeared.
Run the rehearsal as a timed release decision, not a collection of technical tests. Start with the same write restrictions, dependency order, validation gates, and recovery deadline planned for production. Record actual timings and evidence as the sequence runs. Finish with an authorised go or no-go assessment, even if every import job reports success.
The runbook below is a fictional worked example. All organisations, counts, timings, thresholds, defects, and evidence references are hypothetical. It demonstrates a recommended operating design, not a claim about controls available in any particular CRM.
1. Define what the rehearsal must prove
The concrete question is: can the team reach a validated, releasable CRM state within the available window while preserving enough time to restore service if a gate fails?
Answering that question requires more than measuring data transfer. Include the time needed to stop or queue writes, drain in-flight work, extract a consistent dataset, transform it, import objects in dependency order, validate the result, make a decision, and recover when necessary.
Before scheduling the run, fix these boundaries:
- Dataset: the objects, relationships, attachments, and history included or explicitly excluded.
- Write paths: user edits, integrations, scheduled jobs, and inbound processes that could change the source or target.
- Environment: where the rehearsal will run and how it differs from production.
- Release authority: who can approve opening the target and who can order recovery.
- Evidence: which saved reports, logs, and test results establish a pass.
Use a production-representative dataset under appropriate access controls. A smaller sample may expose mapping errors, but it cannot establish full-volume timing. If the environment or volume is not representative, label the timing result provisional rather than extrapolating it into a production commitment.
2. Budget backward from the recovery deadline
Do not give the import process the whole window and hope recovery remains possible. Reserve recovery time first, including restoration checks and the decision to reopen the original service.
In this hypothetical rehearsal, the operating window is 18:00–22:00. The team reserves 45 minutes for recovery and 15 minutes for uncertainty. That leaves 180 minutes to reach a release decision:
240-minute window − 45-minute recovery allowance − 15-minute reserve = 180-minute decision budget
The latest decision is therefore 21:00. If the target cannot be released by then, the default action is recovery. Finishing an import at 20:58 is not a pass if validation still needs 35 minutes.
Each task needs a named owner and a start condition. A task cannot start merely because its planned clock time has arrived. For example, importing opportunities must wait until account imports and their source-to-target identifier mappings have been validated.
Parallel work can shorten the sequence, but it also creates dependencies that are harder to observe. Validate completed batches early if useful; retain a final check against the complete target state. Do not treat an early partial check as proof that later imports left the data intact.
3. Set release gates before anyone sees the results
Agree pass conditions before the rehearsal. Otherwise, a late finish can encourage the team to redefine defects as acceptable exceptions.
The hypothetical team uses these gates:
- G0 — Safe start: recovery materials are available, write restrictions are verified, and outbound side effects are isolated.
- G1 — Complete extraction: the extract manifest matches the frozen source scope, with a recorded extraction boundary and file checksums.
- G2 — Accounted transformation: every extracted record is represented in a load file or an explicitly documented exception. Unexpected rejects block progression.
- G3 — Successful load: target counts, rejected rows, and identifier mappings reconcile with the approved load files.
- G4 — Validated operation: required relationships, ownership, access restrictions, and agreed business workflows pass; no release-blocking defects remain.
- G5 — Timely authorisation: the release authority receives the evidence and records a decision by 21:00.
For this example, missing in-scope records, incorrect opportunity ownership, unintended outbound activity, and access failures are release blockers. Cosmetic display differences may be accepted only with a named owner, recorded business impact, and a resolution plan.
Validation should distinguish completeness from correctness. Matching counts will not reveal records assigned to the wrong owner. Likewise, a few successful user journeys cannot prove that all expected relationships survived.
4. Use a filled rehearsal runbook, not an empty checklist
Fictional worked example: the migration includes 12,000 accounts, 48,000 contacts, and 16,000 opportunities. Historical attachments are excluded by an agreed scope decision. The rehearsal uses an isolated target; outbound actions remain disabled. Names below identify hypothetical task owners.
Each entry records its start condition, actual interval, observed duration, evidence, and result. Evidence labels refer to hypothetical files the team would retain, not existing documents.
Task A — Establish the cutover boundary
Owner: Maya, cutover lead.
Start condition: all owners present; recovery package available; target isolated.
Observed: 18:00–18:10, 10 minutes.
Action: restrict source writes, pause integration writers, confirm in-flight work is settled, and verify outbound processes are disabled.
Evidence: boundary log A01 records the final source change marker, paused writers, and isolation checks.
Result: G0 passed. No uncontrolled write path was observed.
Task B — Extract the frozen source
Owner: Ravi, source administrator.
Start condition: Task A signed off.
Observed: 18:10–18:35, 25 minutes.
Action: extract in-scope objects and relationship keys using the agreed boundary.
Evidence: manifest B01 lists 12,000 accounts, 48,000 contacts, and 16,000 opportunities, plus file checksums and extraction timestamps.
Result: G1 passed. Counts match the frozen-source reconciliation report.
Task C — Transform and check load files
Owner: Elena, migration engineer.
Start condition: manifest and checksums verified; approved owner-mapping file available.
Observed: 18:35–19:05, 30 minutes.
Action: normalise fields, apply mappings, preserve source identifiers, and check required references.
Evidence: transformation report C01 accounts for all 76,000 records, with no reported rejects; configuration version is recorded.
Result: G2 passed under the existing checks. Later validation reveals that those checks were insufficient to detect an incorrect owner mapping.
Task D — Import accounts and establish identifier mappings
Owner: Elena, migration engineer.
Start condition: G2 passed; target is in the agreed clean state.
Observed: 19:05–19:30, 25 minutes.
Action: import accounts, then reconcile source identifiers to target identifiers.
Evidence: import log D01 and mapping report D02 show 12,000 accounts loaded, no rejects, and no duplicate source identifiers.
Result: account checkpoint passed; dependent imports authorised.
Task E — Import contacts and opportunities
Owner: Elena, migration engineer.
Start condition: account mapping checkpoint passed.
Observed: 19:30–20:10, 40 minutes.
Action: load dependent records and retain per-record outcomes.
Evidence: E01 reports 47,988 contacts loaded and 12 rejected; E02 reports 16,000 opportunities loaded. Total target count is 75,988.
Result: G3 failed. Defect D-01 opened for 12 missing contacts. No release progression authorised; diagnostic validation continues within the decision budget.
Task F — Validate the complete attempted load
Owner: Noor, data quality lead, with Sam, business approver.
Start condition: import processing stopped and outcomes captured; cutover lead authorises diagnostic checks without waiving G3.
Observed: 20:10–20:45, 35 minutes.
Action: reconcile counts, compare ownership and relationship keys, test restricted-user access, and exercise agreed business workflows.
Evidence: F01 confirms the 12-contact shortfall; F02 identifies 240 opportunities assigned to the wrong owner; F03 records passing relationship checks for loaded records, access tests, and scripted workflow checks.
Result: G4 failed. Defect D-02 opened for incorrect ownership.
Task G — Record the release decision
Owner: Maya, with Sam as release authority.
Start condition: evidence packet and defect register available.
Observed: 20:45–21:00, 15 minutes.
Action: review failed gates, remaining time, and recovery readiness.
Evidence: decision record G01 states no-go, signed at 21:00.
Result: elapsed decision time is 180 minutes. Recovery begins; target access remains closed.
Task H — Recover and verify service
Owner: Ravi, supported by Maya.
Start condition: no-go recorded; no business writes accepted in the target.
Observed: 21:00–21:45, 45 minutes.
Action: preserve evidence, quarantine the attempted target load, verify the unchanged source, and restore source access and paused processes in controlled order.
Evidence: H01 records source availability checks, writer restart checks, and a representative transaction through the restored source path.
Result: recovery check passed; 15 minutes of reserve remained.
5. Record defects as changes to the next rehearsal
A defect register should explain what failed, why the earlier checks missed it, and what must change before another run.
- D-01: 12 rejected contacts. The target rejected required-field values that transformation checks had accepted. Elena owns the mapping correction. Add a target-compatible required-field check before import, then verify the corrected records load without silently dropping data.
- D-02: 240 opportunities with incorrect owners. An outdated owner crosswalk contained valid target identifiers, so structural checks passed. Noor owns a comparison against the approved business ownership map. Record the approved crosswalk version and compare expected ownership across the full opportunity dataset.
- D-03: no demonstrated timing margin. The decision used the entire 180-minute budget without attempting repairs. Maya owns a revised sequence or an explicitly approved window change. The next run must measure the revised plan rather than assume defect fixes are time-neutral.
Do not quietly repair these issues during the run and report only the successful final state. Record diagnosis, correction, retry, and revalidation time separately. If a repair changes previously loaded records, repeat the checks it could invalidate.
A local fix may prove that a mapping correction works. It does not prove that a clean end-to-end cutover now fits.
6. Test the recovery boundary, not just a restore command
This example has a deliberately simple recovery model: the source remains authoritative, and the target never accepts business writes. Reopening the source therefore does not require merging new target activity.
That assumption stops being valid once users or integrations write to the target. A later return to the source may require capturing, reconciling, and replaying those changes. Do not describe that as a simple rollback unless the procedure has been designed and rehearsed.
Separate the pre-release recovery plan from the post-release incident plan. Mark the precise action that crosses the boundary, such as enabling target business writes. Require explicit authorisation for that action.
Recovery evidence should cover service restoration, not merely infrastructure completion. Confirm that users can reach the intended system, paused writers restart against the correct destination, queued work is accounted for, and outbound actions have not been duplicated.
If recovery exceeds its allowance during rehearsal, the original decision deadline was too late. Recalculate it from the observed recovery duration plus an agreed reserve before attempting another full run.
7. Make the assessment explicit and bounded
The hypothetical decision record reads:
Assessment: NO-GO. The team reached the decision deadline in 180 minutes, but completeness and ownership gates failed. Twelve contacts were not loaded, and 240 opportunities had incorrect ownership. No repair was authorised within the remaining release budget. Recovery completed in 45 minutes, leaving the reserved 15 minutes unused. Production release is not authorised on this evidence.
The result separates three conclusions: the sequence was executable, recovery worked under the tested pre-release conditions, and the migrated target was not releasable. Calling the whole rehearsal successful would conceal the failed data gates; calling it useless would discard valuable recovery evidence.
A future go decision should name the dataset, configuration version, environment differences, passed gates, accepted exceptions, and authorised release boundary. It is evidence for that tested plan—not a guarantee that every later run will finish on time.
8. Turn the findings into the next measured run
Before the next rehearsal, assign an owner and completion date to every blocking change. Require evidence that the two mapping defects are corrected, add the missing pre-import checks, and review whether the decision budget has enough margin for ordinary variation.
Then rerun from the cutover boundary, not from the failed import step. Reuse the same evidence structure so the team can compare extraction, transformation, import, validation, decision, and recovery durations separately.
Your next measurable action is to schedule one full-window rehearsal with named owners and a fixed decision deadline. By its end, require a completed runbook, reconciled record outcomes, a defect register, recovery evidence, and a signed go or no-go assessment. If any of those are missing, the team has performed migration activity—but has not yet proved a releasable cutover.
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