Step 1: Upload your file
Navigate to Data → Import and drag your file into the upload zone, or click to browse. The wizard accepts CSV (UTF-8, comma or semicolon delimited) and XLSX files up to 50 MB.
As soon as the file uploads, Arvexi parses the headers and previews the first 20 rows. If the file uses an unusual encoding or delimiter, the wizard prompts you to confirm before proceeding.
Step 2: AI column matching
Arvexi Cortex analyzes your column headers and sample data to automatically map each column to the corresponding Arvexi field. Cortex recognizes over 150 header variations across six languages, including common abbreviations and ERP-specific naming conventions.
For example, all of these map to the “Account Number” field:
Account,Acct,GL AccountAccount Number,Account No.,Acct #HKONT(SAP),ACCOUNT_CODE(Oracle)Kontonummer,Numéro de compte
Each mapping shows a confidence score. Mappings above 90% confidence are applied automatically. Mappings between 70% and 90% are shown as suggestions you confirm with one click. Anything below 70% requires manual selection from a dropdown.
Step 3: Review mappings
The mapping review screen shows every column side by side: your file’s header on the left, the matched Arvexi field on the right, and the confidence score. Adjust any mapping by clicking the Arvexi field dropdown.
Unmapped columns are highlighted in amber. You can:
- Map manually: Select the target field from the dropdown.
- Skip: Mark the column as ignored. It will not be imported.
- Create custom field: If the column contains data that does not fit any standard field, create a custom field on the fly.
Required fields that have no mapping are flagged with an error badge. You cannot proceed until all required fields are mapped.
Step 4: Staging rows
After confirming mappings, Arvexi parses every row and loads them into a staging table. The staging view shows:
- Valid rows: Rows that pass all validation rules (data types, required fields, referential integrity).
- Warning rows: Rows that parsed successfully but have potential issues, like a date in an unexpected format or a negative amount.
- Error rows: Rows that failed validation. Each error row shows the specific field and validation rule that failed.
You can edit individual cells directly in the staging table to fix errors. Changes are validated in real time. Error rows that you fix move to the valid bucket automatically.
Step 5: Approve and commit
When you are satisfied with the staging data, click Approve & Commit. Arvexi writes valid and warning rows to the database in a single transaction. Error rows are excluded but remain in staging so you can fix and re-commit them later.
The commit summary shows:
- Total rows processed
- Rows committed successfully
- Rows skipped due to errors
- New records created vs. existing records updated
- Duplicate rows detected and skipped
Every commit is a versioned event. If you need to roll back an import, go to Data → Import History, find the commit, and click Revert. This deletes all records created by that import and restores any records that were updated.
Step 6: File fingerprinting
After a successful import, Arvexi generates a fingerprint for the file based on its structure: column names, column order, data types, and row count range. This fingerprint is stored in your Import Profiles.
The next time you upload a file with a matching fingerprint, Arvexi skips the column-matching step entirely and applies the saved mappings automatically. This means recurring imports, like a monthly trial balance export from your ERP, go from six steps to two: upload and approve.
If the file structure changes (new columns added, columns renamed), Cortex detects the drift and prompts you to update the mapping. The old profile is preserved as a version so you can compare what changed.
File fingerprinting also powers scheduled imports. When Arvexi picks up a file from SFTP, it matches the fingerprint to an import profile and processes the file without human intervention.