Skip to main content

Overview

Column metadata configuration allows you to provide business context for your data columns, significantly improving the Query Builder’s ability to understand and answer questions about your data.

Auto-Detection

Automatic data type and format detection on import

Business Context

Add descriptions and categories for better AI understanding

AI Suggestions

Get AI-powered description suggestions

Templates

Apply pre-configured metadata from templates

Why Configure Column Metadata?

When you import a CSV file, the Query Builder automatically detects column names and basic data types. However, it doesn’t know the business meaning of your data.

The Problem

Consider a column named Inspector_Detail_Date_stop_date. Without context, the AI might not understand:
  • What this date represents
  • How it relates to other columns
  • What valid date ranges are expected

The Solution

By adding metadata, you help the AI understand:
  • This is a date field representing when a condition report was completed
  • Valid values range from January 2024 to present
  • It relates to the start_date column (same inspection session)
  • It belongs to the “CR Data” category
Well-configured column metadata can dramatically improve query accuracy, reducing the need for query refinement.

Auto-Detected Metadata

When you import a dataset, the system automatically analyzes each column and detects:
DetectionDescriptionExample
Data TypeString, number, date, boolean, timestampdate
Date FormatSupported date patternsYYYY-MM-DD
Date RangeMin/max dates in the column2024-01-01 to 2024-12-31
Numeric RangeMin/max values for numbers0 to 1000
CategoricalColumns with limited unique values≤50 unique values
Sample ValuesRepresentative values from the data"Lease", "Consignment"
Primary IdentifierColumns likely to be identifiersvin, stock_number

Data Type Detection

The system infers data types by analyzing actual values:
TypeDetection Logic
dateValues match common date patterns (YYYY-MM-DD, MM/DD/YYYY, etc.)
timestampDate patterns with time component
numberAll non-null values are numeric
booleanValues are true/false, yes/no, 1/0
stringDefault for unrecognized patterns

Categorical Detection

A column is marked as categorical when:
  • It has a limited number of unique values (≤50 by default)
  • Sample values are captured for query suggestions
  • The AI uses these values for accurate filtering

Configuring Column Metadata

Accessing the Editor

1

Navigate to Datasets

Go to AnalyticsDatasets in the sidebar
2

Select Your Dataset

Click on the dataset you want to configure
3

Open Column Editor

Click the Edit Columns button or Columns tab

Editable Fields

For each column, you can configure:
FieldPurposeExample
Display NameHuman-readable name shown in UI”Check-In Date”
DescriptionBusiness context for the AI”Date the vehicle was received at the auction”
CategoryGroup columns by domain”Check-In Data”, “CR Data”, “Vehicle Info”

Column Editor Features

Click on any field to edit it directly in the table. Changes are saved automatically.
  • Click display name → Edit name
  • Click description → Edit description
  • Click category dropdown → Select or create category
Click the Suggest button next to any column to have AI generate a description based on:
  • Column name analysis
  • Sample data values
  • Data type and format
  • Common patterns in automotive data
The AI considers context like “CR” (Condition Report), “VIN”, “stock”, etc.
Select multiple columns and assign them to a category at once:
  1. Check the boxes for columns to update
  2. Click Bulk ActionsAssign Category
  3. Select or create a category
  4. Click Apply
Export and import metadata as JSON for:
  • Backing up configurations
  • Sharing metadata between environments
  • Version controlling metadata changes
Export: Click Export → Download JSON file Import: Click Import → Upload JSON file
If you have a similar dataset already configured:
  1. Click Copy from Dataset
  2. Select the source dataset
  3. Matching columns are automatically configured
  4. Review and adjust as needed

Category Organization

Categories group related columns together, making it easier for:
  • Users to navigate large datasets
  • AI to understand column relationships
  • UI to display organized column lists
CategoryUse For
Vehicle IdentificationVIN, stock number, run number
Check-In DataArrival dates, check-in user, receiving info
CR DataCondition report dates, writer, completion info
Sale DataSale type, seller, buyer, auction lane
LocationPhysical location, lot, bay information
TimestampsCreated/updated timestamps, processing times
CalculatedDerived fields and computations

Creating Custom Categories

1

Select Category Dropdown

Click the category field for any column
2

Type New Category

Type a new category name that doesn’t exist
3

Create

Press Enter or click “Create” to add the new category

Best Practices

Writing Effective Descriptions

Do

  • Be specific about business meaning
  • Include valid value ranges
  • Note relationships to other columns
  • Mention data quality considerations

Don't

  • Use technical jargon only
  • Leave descriptions empty
  • Duplicate the column name
  • Include sensitive information

Description Examples

ColumnPoor DescriptionGood Description
check_in_date”A date""Date vehicle arrived at auction. Used to calculate days-on-lot and time-to-CR metrics.”
sale_type”Type""Vehicle sale category: ‘Lease’ (off-lease returns) or ‘Consignment’ (dealer/fleet consignments). Affects processing requirements.”
cr_writer_id”Writer ID""User ID of the condition report writer. References users table. Used for writer productivity analysis.”

Maintaining Metadata Quality

  1. Review Auto-Detected Values: Verify the system correctly identified data types
  2. Add Context After Import: Configure descriptions before heavy Query Builder use
  3. Update When Schema Changes: Re-verify metadata after CSV format changes
  4. Use Templates: Apply templates for consistent metadata across similar datasets

Viewing Metadata Impact

In Query Builder

Configured metadata appears in the Query Builder’s column reference panel:
  • Display names replace raw column names
  • Descriptions appear in tooltips
  • Categories group columns in the panel
  • Sample values inform filter suggestions

In Smart Suggestions

The Query Builder uses metadata to generate contextual suggestions:
  • Time-based queries for date columns
  • Categorical breakdowns for enum columns
  • Aggregations for numeric columns
  • Comparisons using related columns

Troubleshooting

Cause: Auto-detection may misidentify types for mixed dataSolution:
  1. Open Column Metadata Editor
  2. Find the column
  3. Manually set the correct data type
  4. Re-trigger analysis if needed
Cause: Column name doesn’t provide enough contextSolution:
  1. Add a descriptive display name
  2. Provide sample context in description
  3. Re-request AI suggestion
Cause: Category assignment not savedSolution:
  1. Check that changes were saved (look for confirmation)
  2. Refresh the Query Builder page
  3. Ensure dataset is selected in the selector