Glossary

What is Data Usability?

A dataset can be accurate and complete and still be unusable for a task. The team may lack permission to use it for that purpose, a contract with the data provider may exclude the use, or the records may hold personal data that cannot leave a region.

Usability is checked against four kinds of evidence:

  • access rights for the people and systems involved
  • the types of sensitive data in the set and the protection they require
  • contracts and policies that limit use or movement
  • the path the data takes to the system that will use it

Making data usable can mean protecting part of it, such as replacing direct identifiers, while keeping the relationships the task depends on. Protection that removes those relationships, as heavy data redaction can, leaves data that can be shared but no longer serves the task.

How is data usability different from data accessibility?

  • Accessibility asks whether someone can reach the data: a connection, a credential, a query that returns rows.
  • Usability asks whether the data may be used for this purpose, by these systems, along this path.
  • Access without usability looks like this in an AI project: the team can query the table but cannot send it to the model it planned to use.

Where does data usability break in AI work?

A support team wants an LLM to summarize customer tickets. The tickets are reachable and well structured, but they contain names, account numbers, and health details, and company policy does not allow that text to be sent to an external model. The project stalls until the sensitive values are handled in a way that keeps the summaries useful.

In Syntitan, usability is one of the six axes Ready for Raw Data uses to diagnose a project’s copy of the data, with access rights, sensitive data types, contracts and policies, and usage paths as the evidence. When the issue is sending sensitive text to an external model, LLM Capsule replaces sensitive or controlled values with context-preserving substitutes and restores the original values in the response. Data quality for AI lists the six axes and how they pair with a before-and-after test on the task.

Frequently asked questions

What makes data unusable for AI?

Missing permission for the intended use, sensitive data that cannot be moved or shared as it is, contract or policy terms that exclude the use, or no approved path to the system that will use it.

Is data usability the same as data quality?

Not in the traditional sense. Traditional data quality checks ask whether the values are correct and complete. Usability asks whether the data may be used for this task, under the access, contract, and movement limits that apply. Data can pass quality checks and still be unusable.

How do you make sensitive data usable for AI?

Protect the sensitive values in a way that keeps the relationships and meaning the task needs, and confirm that the use stays within the access, contract, and movement limits that apply.