CRM Data Migration: 2026 Best Practices
A CRM migration is not a file copy. It is a data quality project that happens to change platforms. The teams that treat it that way open the new system to clean, complete records on day one. The teams that treat it as an export-import job spend the next quarter chasing duplicates and orphaned deals. This guide walks the sequence that keeps the cutover boring: audit, dedupe, map, validate, then go live.
Treat migration as a data quality project, not a file copy
Most teams underestimate what a CRM migration involves. The mental model is export from system A, import into system B, done. In reality the catastrophic failures are structural, not technical: orphaned deals with no valid account, broken attribution, duplicate contacts, lost consent records, corrupted forecasting. A flat CSV export silently destroys the relational links that make your data useful, so the records land but the pipeline does not. Industry surveys consistently put migration trouble rates high, with a large share of projects running over schedule or missing their objectives, and the root cause is usually people and process rather than a failed import job. The right frame is simple. You are not moving data, you are improving it while you move it. Done well, the migration becomes a forcing function to purge dead leads, retire custom fields nobody has touched since 2019, and merge accounts that should have been one record all along.
Audit and profile the source before you touch anything
Before cleaning a single record, understand what you actually have. Count contacts, companies, deals, and activities. Measure how many records are missing an email or phone, how many deals have no owner, how many accounts have no associated deals, and how many records have had no activity in 18 months or more. That last group is usually an archival candidate, not a migration candidate. Also profile your duplicate rate. Most B2B CRMs carry somewhere in the 10-30% duplicate range once you run a proper check, and a rate above 10% is actively degrading sales operations regardless of which platform you are on. This audit is what sets the real scope of the project and tells you how much cleanup work is coming. Skipping it is how migrations get quoted as a two-week job and finish in two months.
Deduplicate with fuzzy matching, not a spreadsheet sort
Exact-match deduplication is not enough, because real-world duplicates are fuzzy. John Smith, J. Smith, and Jon Smith are the same person, and a spreadsheet sort will never catch it. Use fuzzy matching across name, email, company, and phone, and merge with explicit survivorship rules so you know which record wins and which fields carry forward. Choose match fields by business type: email, phone, name, and company for B2B, email and lifecycle stage for B2C. Automated tools handle scale, but they miss nuanced cases like the same person under two email domains, so combine automation with a round of human review on the ambiguous matches. Dedupe at the source, before migration, not at the destination after import. A realistic target is under 2% duplicates before any import runs. Treat deduplication as an ongoing discipline too, otherwise the clean new system slowly re-accumulates the mess you just cleared.
Map every field, and document every transformation
Field mapping is where data silently disappears. A source field with no destination, or a custom field that was never built in the new CRM, has nowhere to land, so it vanishes without an error. Custom fields are the hardest part of any migration because every business has built up dozens over the years, and some are mission-critical while others are dead weight. Map the ones that matter and deliberately drop the rest. It is common to eliminate 30-40% of columns in a long-running system, and that is healthy pruning, not data loss. Build a single mapping document that captures source field, target field, data type, and transformation logic for each entry, including how null and empty values are handled. Picklist fields like industry, country, and lead source need value-by-value mapping when the options differ, with no Other dumping ground that quietly destroys your segmentation. Keep the document in version control and triple-check it, because one field pointed at the wrong destination corrupts reporting for everyone downstream.
Load in dependency order and validate relationships, not row counts
Relationships only survive if records load in strict hierarchical order: users and owners first, then accounts and companies, then contacts, then deals, then activities and notes, then attachments, and custom objects last. Each step depends on capturing the new target-system IDs from the previous step. To make re-runs safe, store each source-system ID as an immutable legacy key on the target object so you can upsert instead of creating accidental duplicates on a second pass. Then validate the right thing. Matching record counts between source and destination tells you almost nothing, because a deal can import with the count intact yet have lost its account link, making it invisible to the pipeline. Run a full dry-run load into a sandbox, then validate relationships: every deal has a valid account and contact, owners survived, notes attached to the correct record, picklist values mapped cleanly. Run two or three test cycles to surface edge cases, and put actual sales users on the result for acceptance testing. Fixing a broken mapping on 200 sandbox records is trivial. Fixing it on 50,000 production records is a rollback.
Mind the platform API limits when you move at scale
On large migrations the platform itself becomes the constraint, and scripts that ignore rate limits fail silently halfway through. On the Salesforce extraction side, the Bulk API allows large batch submissions per day with up to 10,000 records per batch, and PK chunking keeps queries on big objects from timing out. On the HubSpot write side, the native bulk import endpoint is capped at roughly 500 imports per 24 hours per portal regardless of plan, so for relational data you use the batch CRM object APIs at up to 100 records per call, which also gives you per-record control over associations and timestamps. Watch the search API if your script matches records for deduplication, since that endpoint throttles hardest and becomes the real bottleneck. Build exponential backoff and partial-failure handling into every script, check job status instead of blindly retrying, and verify your exact quotas, which vary by edition and subscription tier, before you plan throughput. A clean migration of 50,000 to 200,000 records typically runs a few days of execution time on top of two to four weeks of audit, mapping, and testing.
Plan a calm cutover: backup, freeze, delta, rollback
Pick the cutover pattern that fits the data. Big bang works for small clean datasets, phased is the safer default, and a parallel run for two to four weeks suits long deal cycles where a hard switch would disrupt tenured reps mid-deal. Whichever you choose, four things are non-negotiable. Back up the source first and prove the backup by restoring a small slice before cutover day, so it is verified rather than assumed. Freeze data entry in the source during the move, or define a soft freeze where reps keep working and every change is tracked. Plan delta processing, the step teams most often skip: records created or updated during the migration window must be captured and synced, or you lose the deal that closed Friday afternoon. And have a documented, tested rollback so you can revert without data loss if something critical surfaces. After go-live, keep the old CRM read-only for several months, re-run the validation checklist on production, and budget a week or two to reconnect and test every integration, because email, calendar, and accounting links rarely come back automatically.
Handle consent and retention while the data is in motion
A migration of personal data is a compliance event, and under GDPR Article 35 it is one of the named cases where a Data Protection Impact Assessment may be required, so run that during planning with your DPO involved. Verify that marketing consent, lawful basis, and opt-out flags exist as structured fields rather than buried in free-text notes, because legacy systems are notorious for importing unclear consent straight into the new platform. Use the move to enforce retention too: indefinite retention is not defensible, and a typical window for inactive data is 12-24 months unless a sector rule like SEC 17a-4 or HIPAA says otherwise. With twenty US states now operating comprehensive privacy laws, the pragmatic approach is to meet the strictest standard everywhere, since satisfying GDPR and California requirements generally covers the rest. Archive what you must keep but will not actively use, keep that archive inside your data map so erasure requests can reach it, and document what moved, why, when, and what you deliberately left behind.
Get operations insights weekly
Join service-business operators who get one actionable insight on ops, automation, and growth every week.
No spam. Unsubscribe anytime.
You’re subscribed! Look for our next issue in your inbox.
That did not go through. Please try again, or email [email protected].
Frequently asked questions
Should I clean my CRM data before or after migration?
Before, without exception. Cleaning during or after the move is several times more expensive because you have lost the context of the source system, users are entering new records in real time, and every duplicate or broken link multiplies once replicated. Deduplicate and standardize first, then move clean data into a clean target. The migration window is the wrong place to discover your data was a mess.
How long does a CRM migration take?
For a mid-size organization, plan on two to four weeks of audit, mapping, and testing, followed by a few days of actual execution for a typical 50,000 to 200,000 record dataset, with a few weeks of monitoring after go-live. The execution is the fast part. The auditing, deduplication, and field mapping are where the real time goes, and front-loading that work is exactly what keeps the cutover uneventful.
Why do records disappear during a CRM migration?
Usually broken field mapping. A source field with no destination, or a custom field that was never created in the new CRM, has nowhere to land, so the data is dropped without an error. The other common cause is flattening relationships into a CSV, which strips the links between deals, contacts, and accounts. Map every field deliberately and load records in dependency order to avoid both.
What does it mean to validate relationships instead of row counts?
Matching the number of records between source and destination only proves quantity, not integrity. A deal can import with the count intact yet have lost its account association, which makes it invisible in the pipeline and breaks forecasting. Validating relationships means checking that every deal has a valid account and contact, owners survived, notes attached to the right record, and picklist values mapped cleanly, before you trust the migration.
What is delta processing and why does it matter at cutover?
Delta processing captures the records created or updated in the old CRM during the migration window and syncs them to the new system. It is the step teams most often forget. Without it, any deal that closes or any contact that changes while the migration runs is simply lost. Either freeze data entry during the cutover or track every change for reconciliation through a delta sync afterward.
How do I make sure I do not lose data during the move?
Take a complete backup of the source and prove it by restoring a small slice before cutover day, so it is verified rather than assumed. Run the full process in a sandbox first, store source IDs as immutable legacy keys so re-runs upsert safely instead of duplicating, validate relationships after each batch, and keep the old CRM read-only for several months. A documented and tested rollback plan is your final safety net.
Ready to ship?
Tell us where you are stuck. We will review it and send a focused recommendation within one business day.