A CRM migration is three jobs, and firms usually budget for one. Moving the data is the job everyone plans. Rebuilding what the data depended on, meaning automations, reports and permissions, is the job that expands. Proving the target matches the source, record by record, is the job that gets skipped, and it is the one that decides whether anyone trusts the new system in month two.
The odds, before you start
Sources: Lowcode Agency on CRM implementation failure, and SyncMatters on data migration outcomes.
Those numbers are not an argument against migrating. They are an argument for spending the first two weeks on the source data rather than on the target configuration.
What it costs
| Scope | Typical range | What drives it |
|---|---|---|
| Data audit and cleanup | $3,000 to $25,000 | Years of legacy records, duplicate households or matters, inconsistent picklists |
| Data migration alone | $5,000 to $50,000 | Record volume, object count, attachment size, how much mapping is bespoke |
| Full mid-market migration | $30,000 to $150,000 | Rebuilding automations and reports, retraining, parallel running |
| Enterprise Salesforce org | $250,000+ | Apex, custom objects, ten or more live integrations |
What never transfers
Vendors quote the data. The data is the easy part. Five categories behave differently, and the last three are where projects slip.
- Records transfer. Contacts, households, matters, accounts, opportunities. Mostly mechanical.
- The record model does not. A contact-centric CRM into a household-centric one, or matters into projects, is a re-modelling exercise, not a copy. Someone has to decide what a household means in the new system before a single row moves.
- Activity history transfers unevenly. Notes, tasks, calls and emails are stored differently in every product. Recurring tasks are the classic casualty: Wealthbox documents that future occurrences cannot be imported from Redtail, and that Redtail's Status, Servicing Advisor and Writing Advisor fields are remapped rather than copied.
- Automations, reports and permissions never transfer. Flows, validation rules, dashboards, sharing rules and role hierarchies are rebuilt from scratch. Budget this as build work, not migration work.
- Integrations do not follow. Every connected system, the DMS, the custodian feed, the billing platform, the e-signature tool, needs to be repointed and retested. This is the line item most often missing from the original quote.
Direction changes the work
| Moving onto Salesforce | Moving off Salesforce | |
|---|---|---|
| Hardest part | Modelling your firm into objects designed for sales pipelines | Extracting logic that lives in Apex and Flows, which no export contains |
| Data mechanics | Data Loader or the Import Wizard, external IDs, upsert order across objects | Reports and the Bulk API, one extract per object, then reassembly |
| Hidden cost | A named administrator you did not have before | Reproducing reports partners have been reading for years |
| Common trigger | Outgrowing an advisor or legal CRM, or buying Litify, Practifi or FSC | Admin burden and license cost outrunning the value the org delivers |
| What to test first | One household or one matter end to end, including documents | Whether the target can express your permission model at all |
The five phases that work
- Audit the source. Count records per object. Find duplicates, orphans, dead records and picklist drift. Decide what does not come with you, and get that decision signed off before mapping.
- Map the fields against real records. Not against the schema. Pull fifty real households or matters, including the ugly ones, and walk each field. Every remap is a decision with a compliance consequence, so write down who made it.
- Dry run into a sandbox. Load the full set, not a sample. Sample loads hide the failures that only appear at volume: attachment limits, API throughput, required-field collisions.
- Cut over on a quiet window. Freeze writes in the source, load, repoint integrations, unfreeze. Keep the source read-only for at least a quarter.
- Reconcile. Compare target to source record by record and field by field. Produce a variance report. This is the phase that separates a migration people trust from one they work around.
Reconciliation is the step nobody budgets
Most migrations end at load. Someone spot-checks a few hundred records, the counts look close, and the project closes. Two months later a partner notices a matter with no history, or an advisor finds a household whose fee schedule came across wrong, and confidence in the new system quietly collapses.
Full reconciliation is unglamorous and mechanical: pull both sides, compare every field on every record, classify each difference as expected, remapped or wrong, and hand back a list. It is exactly the kind of work firms either skip because it is expensive in human hours, or pay a contractor to do in Excel over three weeks.
The compliance clause
Retention obligations survive the migration. Registered advisers are subject to the Advisers Act books-and-records rule (17 CFR 275.204-2), and law firms carry state retention rules plus duties over the client file. Practically, that means two things: a gap in history is a finding, not a cosmetic issue, and the mapping decisions you made in phase two are part of the record. Keep the field map, the variance report and the sign-offs where an examiner or a general counsel can find them.
The honest summary
Where Caddi fits
Caddi does the mechanical half of a migration and then keeps doing it. Show it the task once and it reads the source, applies your field map, loads the target, then compares both sides record by record and hands back a variance report you can act on. Because it runs over APIs across your stack, the same agent handles the parts a migration tool ignores: attachments filed to the right matter, documents renamed to the target's convention, and the post-cutover hygiene that otherwise becomes someone's Monday.
Migrating a specific pair of systems? See the migration guides by system. Also useful: before you hire a Salesforce developer, best CRM for law firms, and best CRM for financial advisors.
Migration without the manual pass
See a migration reconciled record by record
Caddi reads the source system, applies the transform, loads the target, then checks every record and reports the variances.
Frequently asked questions
How much does a CRM migration cost?
The data work alone commonly runs $5,000 to $50,000. A full mid-market migration lands between $30,000 and $150,000 over several months, and Salesforce orgs carrying Apex and ten or more integrations regularly pass $250,000. Data cleanup and customization discovered mid-project is what pushes first-year cost 30 to 50 percent above the original estimate.
How long does a CRM migration take?
Plan on six to twelve weeks for a clean mid-size migration and three to six months where a vertical platform is being implemented at the same time. Firms with years of legacy data should budget several weeks for cleanup before any data moves.
What does not transfer in a CRM migration?
Automations, reports, dashboards, validation rules and permissions never transfer; they are rebuilt. Activity history, attachments and recurring tasks transfer partially and vary by vendor. Wealthbox, for example, documents that recurring-task future occurrences cannot be imported from Redtail, and that Redtail's Status, Servicing Advisor and Writing Advisor fields get remapped rather than copied.
Why do CRM migrations fail?
Around 30 percent of CRM implementations fail outright and roughly 55 percent never deliver the planned benefit, while more than 60 percent of data migration projects finish partial at best. The usual causes are unaudited source data, a field mapping nobody validated against real records, and no reconciliation step after cutover to prove what actually landed.
Should we clean the data before or after migrating?
Before, for anything structural: duplicates, entity relationships, dead records and inconsistent picklists. Migrating dirty data doubles the problem because you then own it in two systems. Cosmetic cleanup can wait; anything that affects how a record maps cannot.