Skip to content
What’s my data worth?

How AI uses business data

How AI teams use business records.

Your records can teach an AI system, test it, or show it how people corrected its work. Each use needs different details and permissions.

What’s my data worth?

Training

Help a model learn from approved examples.

Made-up example

Use resolved support cases to learn how people investigate a problem and explain a fix.

What you need

  • The task and useful examples
  • Permission for training
  • Clear fields and quality checks

Permission to test a model does not automatically include permission to train it.

Testing

Check whether an AI system can do a task correctly.

Made-up example

Give an AI system a shipment problem. Check its plan against the allowed routes and confirmed dates.

What you need

  • Inputs that do not include the expected answer
  • Checks that define a good result
  • Test cases kept separate from training examples

A test result covers the cases and system version tested. It does not prove performance on every task.

Human feedback

Use saved reviews to understand what people changed and why.

Made-up example

Compare an AI draft with the reply an employee approved, including the reason for the changes.

What you need

  • The original and reviewed answer
  • The review reason and rules, where recorded
  • Permission to use the feedback and its inputs

These are records a business already holds. Korra does not claim to supply a reviewer network.

AI task examples

Show a realistic task, its rules, and a successful result.

Made-up example

Create a scheduling case with a customer request, available times, allowed actions, and a confirmed appointment.

What you need

  • The starting request and available actions
  • The rules and details needed to do the work
  • Checks for completion and exceptions

The records describe a task. Building an interactive test system is separate work.

See a record structure.

Three invented records show a support case, a workflow, and an AI correction. The pack contains no real customer data.

Download example pack Read the format guide

JSONL records, field definitions, and a format guide. These examples show a structure, not available inventory or measured results.

Example record, shortened
{
  "synthetic": true,
  "record_type": "operational-workflows",
  "request": "Reschedule a shipment after a route closes.",
  "work": [
    "Check approved alternatives.",
    "Record the required approval."
  ],
  "result": {
    "status": "completed-for-example",
    "outcome": "A feasible route is approved."
  },
  "evaluation": {
    "criteria": [
      "Uses a permitted route.",
      "Records the approval."
    ],
    "acceptable_alternatives": [
      "A plan requiring approval before release."
    ]
  }
}
Download JSONL only

Have records
like these?

Tell us what your team records and how far back it goes. We check which uses fit, and you approve every term in writing.

What’s my data worth?

Before choosing records

  • Choose training, testing, or saved feedback
  • List the details each example needs
  • Agree the checks before preparing records
  • Check permission for the use
Review dataset quality