Results
What does the score say?
- Total score
- --
- Items scored
- --
- Gate items at zero
- --
- Result
- --
Score every item to see the result.
Calculator Library / AI Operations
AI Data Readiness Check
Score the data for an AI use on ten questions with three gates, check the accuracy of a sample, and see what to fix first. Nothing is sent anywhere.
Tool
Score the data
Go = 16 or more, no gate at 0
Score each of ten questions from 0 (not in place) through 1 (partly) to 2 (in place and working), ideally with the people who create and use the data and with a real sample in front of you. Three questions are gates: a zero on any of them stops a go decision. Load the example to see work-order data scored.
| Question area | What good looks like | Gate | Score |
|---|
Optional: check accuracy against the source
Take a random sample of records, compare each with the source or an expert’s answer, and enter the counts.
Enter the counts to see the accuracy with a 95% interval.
Everything runs in your browser. Nothing is sent to a server, and entries are not saved between visits. The example is made-up data. The scores are judgments; check them against evidence.
Picture
Scores by item
What This AI Data Readiness Check Does
The check scores your data on ten questions: availability, accuracy, completeness, consistency, timeliness, ground truth, permission, sensitive content, ownership and lineage, and representativeness. Each is scored 0, 1, or 2. Three are gates (accuracy, ground truth, and permission): a zero on any of them stops a go decision. Otherwise a total of 16 or more is a go, 10 to 15 means prepare first, and below 10 means stop.
It also gives an exact-style 95% Wilson interval for the accuracy of a sample you checked against the source. It runs in your browser and does not send your data anywhere. It supports the Data Readiness for AI in Operations Guide.
How the Result Is Decided
| Condition | Result |
|---|---|
| Any gate item scored 0 | Stop: fix this first |
| Total 16 or more of 20, no gate at 0 | Go |
| Total 10 to 15 | Prepare first: close the gaps, with owners and dates |
| Total below 10 | Stop: not ready |
Worked Example
The pre-loaded scores are for maintenance work-order data. The total is 13 of 20 and no gate is at zero, so the result is Prepare first. The accuracy sample is 78 matches in 100 records, which gives an interval of 69% to 85%. The data are made up.
Limits and Good Practice
- Score from evidence: look at a sample of the data. Opinions about data quality are often too kind.
- The thresholds are a reasonable starting rule, not a standard. Adjust them to the risk of the use.
- A go on data readiness is not a go on the use. You still need validation, governance, and monitoring.
- Check permission with privacy and legal leads (see the Data Protection and Privacy Guide).
- Data quality drifts. Score again at each review.
AI Data Readiness Check Frequently Asked Questions
What is the AI Data Readiness Check?
A ten-question scoring tool that tells you whether the data for an AI use is ready: go, prepare first, or stop. Three questions are gates, so a zero on accuracy, ground truth, or permission stops the go decision.
What is ground truth?
A set of cases with correct answers agreed by experts, used to test the AI and monitor it. Without ground truth you cannot tell whether the tool is right.
How big a sample do I need to check accuracy?
Fifty to a hundred random records is a practical start. The tool shows the 95% interval, which narrows as the sample grows.
Are the thresholds a standard?
No. They are a reasonable starting rule, 16 or more for a go and 10 to 15 for prepare first. Adjust them for the risk of the use.
Does a go mean I can launch?
No. It means the data is not the blocker. You still need validation on a golden test set, governance, security, and monitoring.