Migration guide
Migrating from Salesforce to a custom CRM: the inventory before the export
Map the records, files, permissions and connected work before leaving Salesforce. Use the invented account example to plan a rehearsal and record what still needs testing.
Why the export is the easy part
An export is one input to a migration. First establish which objects, fields, history and file contents your chosen export method can retrieve with your account's permissions. Save its settings and record any exclusions.
Then map the work around those rows: relationships, access rules, automation and connected systems. Data files alone do not specify how a different application should behave. Salesforce's object metadata documentation, checked October 7, 2026, describes a separate resource for retrieving field and child-relationship information.
This guide covers leaving Salesforce for a system you own. It is an inventory. It sets no timetable. We do not give migration timelines before the inventory exists, and you should be careful with anyone who does.
The seven inventories
Make each one a spreadsheet tab. One row per item. For every row, record an owner and one of three decisions: carry over, rebuild differently, or leave behind.
1. Relationships
List every object you use, standard and custom, and how they connect. Accounts to contacts is obvious. The ones that hurt are less visible: a custom object that links an account to a territory, a junction object between products and price books, a lookup someone added three years ago that a report depends on.
- Every object with a record count
- Every lookup and master-detail relationship
- Record types, and what each one changes
- Fields that are required, unique or used as external IDs
2. History
Decide what history has to come with you and in what form.
- Activities: tasks, events, logged calls and emails
- Field history tracking, where it is switched on
- Stage history on opportunities
- Notes, Chatter posts and case comments
Some history needs to be live and searchable on the account. Some only needs to exist in an archive you can open during a dispute. Those are different amounts of work, so decide per item.
3. Files
List file volume, versions, record links, ownership and who may download each file. Salesforce's Data Loader file-export guidance, published July 7, 2026 and checked October 7, distinguishes CSV metadata from the physical file contents. Confirm that your chosen method retrieves both. Locate the original of any signed agreement and record its retention decision with the responsible owner.
4. Permissions
Write down who can see and change what, in plain language, before you look at how Salesforce implements it. Profiles, permission sets, roles, sharing rules and territory rules are the mechanism. What you need to carry over is the intent.
- Which roles exist in the business
- What each role may see, edit, approve and export
- Records that are restricted for a reason: pricing, personal data, legal holds
Record uncertain rules explicitly and name the person who can decide them.
5. Automation
Everything that happens without a person doing it.
- Flows, and any remaining Process Builder or workflow rules
- Apex triggers and scheduled jobs
- Validation rules and assignment rules
- Approval processes
- Email alerts and templates
For each one, write its trigger, business rule, side effect and owner. Decide which rules to retain before designing their replacement.
6. Integrations
Every system that writes to Salesforce or reads from it.
- Connected apps and installed packages
- Middleware and scheduled syncs
- The ERP, accounting, marketing, support and data warehouse connections
- Spreadsheets that someone uploads by hand every month
For each, note the direction, frequency, owner on the other side and consequence of a missed run. Use those dependencies to plan cutover order.
7. Reports
Ask owners which reports and dashboards support a decision. Use available usage records to check the answers. Record filters, calculations and field dependencies, including reports used only at quarter-end.
A worked dependency inventory
Synthetic demonstration: invented records. This planning example uses illustrative ID labels. No migration was performed in a Salesforce org for it. No real client or measured customer result is shown, and no customer team has reviewed this example.
The pilot workflow is a rep opening an account, reading a contact's last call and downloading an agreement. A rep assigned to another account must not reach that agreement. The destination names describe a proposed model.
| Source item | Proposed destination and dependency | Decision and owner |
|---|---|---|
| Account A-101, Cedar Supply | Account C-501; retain A-101 in an ID map | Carry over; operations owner |
| Contact P-201 | Contact C-601 under C-501; resolve account before loading | Carry over; sales owner |
| Call T-301 linked to A-101 and P-201 | Activity C-701 with both relationships and original time zone recorded | Carry over; sales owner |
| Agreement F-401 linked to A-101 | File C-801 plus bytes, version and account link; compare a content checksum | Carry over; document owner |
| Rep access rule for A-101 | Assigned rep can read account/activity/file; other rep cannot | Rebuild and test; access owner |
Copy these columns for each dependency: source type and ID, target type and ID, parent IDs, owner, decision, transformation, expected check, observed result and unresolved issue. Keep the original ID even when the new system assigns another. Never use a matching company name alone to merge records; have the data owner approve ambiguous matches and preserve the mapping.
Rehearse this account before the full load
- Save a dated source snapshot and the export settings. Record counts and exclusions for the chosen scope. Keep a separate original copy.
- Load the account, then dependent contacts, activity and file links using the ID map. Treat unmatched parents as exceptions for review.
- Inject an invented second contact with the same email. Expect an exception awaiting an owner's decision. Record whether the loader actually does that.
- Check that the call resolves to the intended account and contact. Open the file and compare its checksum with the source copy; matching record counts alone cannot detect a wrong link or corrupted file.
- Sign in as the assigned rep, another account's rep and the administrator. Try the file's direct URL as well as the account screen. Record allowed and denied results.
- Change the contact after the snapshot and repeat the import. Expect one updated contact, no extra activity, and no unapproved outbound email. Test deletion and failed-import recovery separately.
Observed result: not run. These are expected checks for an invented dataset. Complete an observed-result column in your test environment; none of these checks has been executed against a live CRM for this article.
Reconciliation: how you know it worked
Agree on the acceptance checks before the first test load.
| Check | What to compare |
|---|---|
| Record counts and source-ID coverage per object | Missing, duplicate and unexpected IDs; explain approved exclusions |
| Totals on money fields, by period and by owner | Differences after type, rounding and currency transformations |
| A sample of accounts reviewed by the rep who owns them | The record looks right to the person who knows it |
| Each applicable integration direction, including failure and retry | Expected records, side effects and recovery; one successful run is only one test |
| One full workflow performed from start to finish by a real user | The tested path matches the agreed task; log exceptions and untested branches |
| Allowed and denied actions for each role and account boundary | Tested access matches the written rule, including direct file access |
Write down who signs each check. A migration without named sign-off is a migration nobody accepted.
Run both systems on purpose
Plan a period where Salesforce and the new system run side by side for one team. Decide in advance which one is the system of record during that time, how changes flow between them, and what ends the period. Parallel running that nobody designed is how records drift apart.
Record a cutoff time and how to capture creates, updates and deletions after the snapshot. Reconcile that final change set before transferring writes. Pause jobs that would send duplicate emails or create duplicate downstream records.
Keep a recovery worksheet: trigger for stopping, decision owner, last accepted snapshot, changes made since that snapshot, method to replay or reconcile those changes, and owner of reopened work. Rehearse it and record elapsed time. Returning to an old snapshot without accounting for new work can lose changes; do not promise a rollback time before testing the plan.
What to leave behind
Mark unused fields, retired objects and obsolete automation for review. Have the business and retention owners approve each exclusion. Keep an accessible archive where required before deleting source records.
Common mistakes
- Treating the export as the plan. Rows without relationships, permissions and automation are not a working system.
- Migrating everything without a decision. Carry-over scope needs an owner and a reason.
- Forgetting the quiet integrations. The nightly job nobody remembers is the one that stops invoicing.
- Setting a date before the inventory. The inventory sets the date.
- No named sign-off. If nobody accepts the migration, everybody disputes it.
Using this with us, or without us
Use this inventory alongside the three-year CRM cost worksheet: include test loads, file handling, parallel operation and support in the migration line. The custom CRM development overview explains how one workflow becomes a scoped pilot.
Bring the system names and one dependency you cannot map yet. Use invented or sanitized records; do not send credentials or customer exports. We can work through the inventory after a discovery call, and you keep it whatever you decide.