AI Readiness (AIR) Score
What the AI Readiness Score measures, how to read and act on it in the Metadata Viewer, and how AIR affects AskEdi and Auto Generate Charts.
Clean data is necessary for reliable AI analysis. But it is not sufficient on its own. An AI model working with a column named amt or flg_1 has no way to know what those values represent in your business. It is guessing. The AI Readiness (AIR) Score measures how much of that guesswork has been replaced with verified, human-confirmed context.
Every integration in Edilitics gets an AIR Score alongside its DQ Score. The two scores work together: DQ tells you how clean the data is, AIR tells you how well the AI understands it.
How AIR Is Built
The AIR Score combines two things equally.
The structural half (50%) comes directly from the DQ Score. Clean, complete, well-formatted data is the foundation. Without it, no amount of documentation helps. This half is automatic.
The documentation half (50%) comes from how well each column in your schema is described and whether a human has confirmed those descriptions are accurate. This half is in your control.
Each column contributes to the documentation half based on its current status:
| Status | Contribution |
|---|---|
| No description | 0 points |
| AI-generated description | 0.2 points |
| Validated by a human | 1.0 points |
The validation percentage is the average of all per-column point scores across the integration. A schema where every column has an AI-generated description scores 20% on the documentation half. A schema where every column has been validated by a human scores 100%.
This creates a ceiling effect that matters in practice:
- Perfect DQ, no descriptions: 50% AIR, Grade D. The AI has clean data but no context.
- Perfect DQ, full AI-generated descriptions: 60% AIR, Grade C. The AI has descriptions, but they are unverified.
- Perfect DQ, every column validated by a human: 100% AIR, Grade A. The AI has clean data and verified context for every column.
Grade Scale
The AIR Score maps to the same letter grade and colour badge system as DQ:
| Grade | Score | What It Signals |
|---|---|---|
| A | 90 – 100% | Schema is fully documented and human-verified. AI analysis will be accurate and business-specific. |
| B | 75 – 89.9% | Most columns are verified. AI analysis will be reliable for the majority of queries. |
| C | 60 – 74.9% | Descriptions exist but are largely unverified. AI may misinterpret ambiguous columns. |
| D | 45 – 59.9% | Limited documentation. AI is largely working from column names alone. |
| F | 0 – 44.9% | No documentation or very low DQ. AI output will be unreliable. |
Before every AskEdi session and Auto Generate Charts run, an AI Advisory is shown. It reflects the current DQ and AIR grades for that table and adjusts its tone accordingly: a positive confirmation for Grade A or B, an informational note for Grade C, and a clear warning for Grade D or F. The advisory never blocks access to either feature.
If the AIR Score shows as N/A, AI Column Insights have not been generated yet for this integration. Use Generate AI Insights from the integration hover menu to start the process.
The Metadata Viewer
The Metadata Viewer is where you read, assess, and improve your AIR Score. Open it via View AI Insights in the integration hover menu.
Before the viewer opens, Edilitics checks whether every table in the integration has AI Column Insights generated. If any tables are missing insights, a modal titled "AI Column Insights Not Generated" appears first. It lists the specific tables that are pending and gives you two options:
- Generate Insights: starts generation for all pending tables using your existing privacy mode and LLM provider settings, then opens the Metadata Viewer.
- Continue Anyway: skips generation and opens the Metadata Viewer immediately. Only tables with generated insights are listed. Tables without insights are not shown.
Once inside the viewer, select a table from the left panel to see its columns laid out in a table with four columns:
| Column | What It Shows |
|---|---|
| Column Name | The name of the column as it appears in your source. |
| Data Type | The declared data type (text, integer, date, etc.). |
| Data Quality | The per-column DQ score. Hover the info icon to see null count, distinct count, noncompliant count, average, and standard deviation for that column. |
| Column Metadata | The description, its status, the date it was last updated, and the action icons. |
Each description cell shows one of two states:
AI-generated: A sparkle icon and "Auto Generated" label with the date. This description was written by the AI based on the column name, data type, and DQ statistics. It is a starting point, not a verified record.
Validated: A person icon and "Validated by [Your Name]" label with the date and a shield badge. This description has been confirmed or written by a human. It is what the AI will use as authoritative context.
Approve and Edit: What Each Action Does
Two actions are available on every description cell: Approve and Edit.
Click Approve when you have read the AI-generated description and it accurately describes what the column means in your business. No editing required. The status updates immediately to "Validated by [Your Name]".
Approval is permanent. Once a column is approved, it cannot be rolled back. If the description later turns out to be wrong, use Edit to correct it. Saving any edit also counts as validation.
Click Edit to open the Column Metadata modal. This shows you:
- The column name and data type
- The column's DQ score, null count, and distinct count at a glance
- A full DQ breakdown: noncompliant count, average value, standard deviation, min value, max value, and the most frequently occurring values
Use this information to write a description that is specific to your data. The description field accepts between 200 and 300 characters. The character counter turns red when the description is outside the valid range and green when it is within it. The Update button activates only when the description is within range and has been changed.
Saving any edit automatically marks the column as validated by you. You do not need to click Approve separately.
Edit can be done as many times as needed. Each save updates the "Validated by" label with your name and the current date.
What a Good Description Contains
This is the most important part of improving your AIR Score. A weak description wastes the opportunity. A strong description gives the AI precise, verified context it cannot derive from the column name or data type alone.
A good description answers four questions:
- What does this column represent? Not the data type. The business meaning.
- What unit or scale? Currency, percentage, days, kilograms. The AI needs to know.
- What values are possible? For categorical columns, list the values and what they mean.
- How is it used? Is it a reporting metric, a filter dimension, a join key?
Examples:
| Column | AI-Generated (weak) | Human-Written (strong) |
|---|---|---|
net_revenue | "Numeric column containing revenue data." | "Net transaction revenue in GBP after refunds and discounts, excluding VAT. Core metric in monthly P&L reporting. Range: 0 to 500,000." |
status | "String column with status values." | "Order fulfilment status updated by the warehouse system. Values: pending, processing, shipped, delivered, cancelled." |
customer_id | "Unique identifier for customers." | "Primary customer identifier. Joins to the customers table. Null if the order was placed as a guest checkout." |
created_at | "Timestamp column for record creation." | "UTC timestamp when the order was placed. Used as the primary date dimension in sales reporting and trend analysis." |
Notice what the human-written versions contain that the AI versions do not: units, join relationships, the specific values a categorical column can take, how the column is actually used in analysis. That context is what separates a reliable AskEdi session from a generic one.
Prioritise columns that appear most frequently in your AskEdi sessions and reports. Validating ten high-traffic columns delivers more improvement to analysis quality than validating fifty rarely-used ones.
How AIR Affects AskEdi and Auto Generate Charts
Every time AskEdi answers a question or Auto Generate Charts suggests a visualisation, it reads the column descriptions stored in your Metadata Viewer as context. The quality of that context directly determines the quality of the output.
With a low AIR Score:
AskEdi is working from column names and data types alone. A column called amt could be revenue, a discount, a quantity, or a tax amount. Without a description, the AI picks the most likely interpretation and may be wrong. Results will be generic, may use incorrect units, and may aggregate columns that should not be aggregated together.
With a high AIR Score:
AskEdi knows that amt is "Net transaction revenue in GBP after refunds and discounts, excluding VAT." It knows that status can be "pending, processing, shipped, delivered, or cancelled." It knows that created_at is "the UTC timestamp used as the primary date dimension in sales reporting." Every query benefits from this verified context. Calculations are more accurate, chart axes are correctly labelled, and explanations are grounded in your actual business logic rather than generic inference.
Auto Generate Charts uses descriptions to select the most relevant chart type and to label axes in business terms. A column described as an order fulfilment status with five specific values will be suggested as a bar chart showing status distribution, not a line chart or a scatter plot. The descriptions tell the AI what the data means so it can recommend the visualisation that actually answers your business question.
Where AIR Scores Appear
| Location | What You See |
|---|---|
| Integration card | AIR score pill alongside the DQ pill. Click to open the Score Breakdown Drawer. |
| Score Breakdown Drawer | AIR breakdown: DQ Baseline progress bar and Human Validation progress bar, each contributing 50% to the final score. |
| Metadata Viewer | Per-column validation status, description content, and last updated date for every column in every table. |
| AskEdi session start | AI Advisory shown before every session. Tone adjusts by grade: positive for A/B, informational for C, warning for D/F. Never blocks the session. |
| Auto Generate Charts | AI Advisory shown before every generation run. Same grade-based tone as AskEdi. Never blocks chart generation. |
Frequently Asked Questions
Next Steps
AI Column Insights
How AI-generated descriptions are created, privacy modes, and how to enable and validate insights.
Data Quality Profiling (DQ)
How the structural half of AIR is calculated and how to improve a low DQ Score.
DQ Refresh and Schema Drift
How manual and scheduled profile refreshes work and how schema changes are detected.
Need help? Email support@edilitics.com with your workspace, job ID, and context. We reply within one business day.
Last updated on