The pattern: download, edit offline, upload, verify result file, fix errors and re-upload. The cycle is fast for small updates (under 100 rows) and manageable for bulk audit work (thousands of rows). Always test in sandbox before applying to production.
- Download current classification state via Data Classification Download
The CSV is the starting point. Save with a timestamped filename so the audit cycle is traceable.
- Edit the CSV offline
Add missing classifications, update stale ones, coordinate with field owners on accurate values. Save the corrected CSV.
- Validate the CSV schema before upload
Confirm column headers match the expected schema, Object API Name and Field API Name spelling is correct, classification values match the org's value sets.
- Upload in sandbox first
Test on a sandbox with similar field schema. Verify the result file shows expected successes and errors.
- Upload in production
Setup, Data Classification Settings, Upload. Pick the CSV. Pick the Upload mode (Update for additive, Replace for authoritative). Submit.
- Review the result file
The Upload produces a per-row success/failure CSV. Save it as evidence of the change. Identify failed rows for correction.
- Fix errors and re-upload the corrected subset
Build a corrected CSV with only the failed rows. Re-upload. Iterate until all rows succeed or the failures are documented as intentional non-updates.
Update (leave empty cells untouched) or Replace (clear classifications for empty cells). Pick deliberately.
Object API Name, Field API Name, plus columns per enabled classification dimension. Match the Download schema.
Errors on invalid values, missing required columns, unmatched fields. Some orgs configure stricter validation than defaults.
Per-Upload row count. Smaller batches make error recovery faster; larger batches reduce ceremony.
Large uploads benefit from off-hours timing to avoid competing with user-facing requests.
- Misspelled API names produce errors. Object and field identifiers are case-sensitive; pull the correct spelling from the Download CSV.
- Replace mode clears classifications for empty CSV cells. Use Update mode unless the CSV is authoritative; otherwise the upload destroys existing classifications.
- Invalid classification values (a Sensitivity Level not in the org's value set) produce errors. Validate values against the org's policy before upload.
- Large uploads compete with user-facing requests. Schedule for off-hours on big orgs.
- Sandbox first is non-optional for production-affecting bulk updates. The sandbox dress rehearsal catches schema issues at the right time.