What I Learned Migrating 12 CRM Databases: A Data Analyst's Field Guide to B2B SaaS Migration

Day 17: The Ghost in the Lead Record

I still remember opening the first post-migration report for Acme Corp's Salesforce-to-HubSpot migration and seeing 3,412 leads marked 'Created Date: 1970-01-01'. Not an error---just the Unix epoch default when timestamp fields were unmapped. That single oversight erased four years of lead source attribution. We rebuilt it from web log backups---but only because we had archived raw nginx logs six months prior. Lesson one, carved in fire: never assume your target platform honors null dates the same way your source does.

Day 42: Why Your Clean Data Is Not Clean (and What to Do Instead)

We spent three weeks prepping data for a mid-market fintech's marketing automation migration. Their sales team swore their CRM was 'audit-ready'. Turns out, 68 percent of Company Name fields contained typos like 'Intuit Inc.' vs 'Intuit, Inc.' vs 'Intuit LLC'---all treated as distinct accounts. Worse: 14 percent of contacts had duplicate emails across multiple records, with mismatched job titles and last-touch dates.

So we paused. Built a lightweight deduping layer in dbt using fuzzy string matching on company domain + name + phone, then manually reviewed top 200 matches with the customer success lead. Result? We reduced account-level duplication by 91 percent---and recovered 11 months of previously orphaned engagement history by stitching contact records to their true parent accounts.

The 3-2-1 Rule We Now Enforce

- 3 sources of truth validation: CRM, billing system, and marketing automation logs

- 2 human sign-offs: one from sales ops (for ownership logic), one from marketing (for campaign lineage)

- 1 immutable archive: all raw CSVs, API response dumps, and transformation logs---stored in versioned S3 buckets with object lock enabled

Day 89: When ETL Becomes ELT---Then Back Again

For a global SaaS client moving from Pardot to HubSpot, we tried pure ELT---loading everything raw into Snowflake first, then transforming. But HubSpot's API rate limits (200 calls/minute, max 10K/day) meant our initial sync stalled at 18 percent after 36 hours. We pivoted: built a staging layer in Airbyte that batched and deduped *before* ingestion, added exponential backoff plus jitter, and prioritized records by last activity date---so hot leads synced within 90 seconds of trigger.

Final stats: 4.2 million contacts, 1.7 million companies, 22 years of historical deal stages---all migrated in 58 hours, with zero record loss and less than 0.03 percent field corruption (mostly due to legacy HTML formatting in notes fields, which we sanitized with a Python regex pass).

Day 131: The Forgotten Fields (and Why They Cost $217K)

One client lost $217K in pipeline visibility because we did not map Forecast Category from their old Salesforce org to HubSpot's Deal Stage Probability. Their sales team used Forecast Category to signal deal health---Best Case, Commit, At Risk---but HubSpot only accepted numeric probability values. We assumed Commit = 75% until week two of UAT, when their VP of Sales said, 'No---we treat Commit as 90% because it is legally binding'. That required rebuilding the entire stage mapping logic and retraining 37 reps.

Now, before any migration kicks off, I run a Field Intent Audit: interview 3 reps, 2 marketers, and 1 sales ops analyst about *how they use each field*, not just what it is named. We document intent, usage frequency, and business impact---not just data type.

Day 184: The Real Migration Starts After Go-Live

Migration is not done when the last record lands. It is done when the first rep confidently says, 'Yes, this deal shows the right 2022 email campaign in its timeline'. For that to happen, we build a 30-day data fidelity sprint: daily reconciliation reports (lead count, opportunity value delta, campaign member counts), Slack alerts for greater than 2 percent variance, and shadow-mode logging of all new entries for 14 days to validate write-path integrity.

Last month, that caught a silent bug where HubSpot's Contact Property Sync setting overwrote custom fields from our enrichment service---because the checkbox was accidentally toggled during sandbox testing. Fixed in 11 minutes. No data overwritten. Because we watched.

Data migration is not about moving bits. It is about preserving context, intent, and trust---one field, one timestamp, one human decision at a time.