Preparing Business Data for AI Research
2 min read
Start with purpose and permission
Business systems contain useful examples of collaboration, but not every record belongs in a research dataset. Before preparing material, define the intended use and establish who can authorize sharing. A contributor’s access to a file does not automatically give them the right to license it.
Remove what the task does not need
Names, email addresses, account identifiers, and confidential details can appear across documents and conversations. Preparation should consider both obvious identifiers and indirect clues. Removing a name alone does not guarantee anonymity, especially when several records can be linked together.
Preserve the useful structure
Privacy work should be paired with an understanding of the research task. Consistent replacements can retain the relationship between a request, a decision, and its outcome without retaining the original identity. The right approach depends on the source material and the risks involved.
Review before release
Quality checks should examine privacy, completeness, and whether the prepared examples still make sense. Arcana’s focus on preparing and licensing practical knowledge starts with these questions. Contributors should discuss sensitive material and restrictions before sharing source records.