Metrics Catalog

A reference of your organization's metrics with the ability to create, edit, and import.

Why a Catalog is Needed

The metrics catalog is a centralized repository of information about all indicators used by the team. It serves as a mini-wiki for documenting:

• Which metrics are used in experiments
• What type each metric is (conversion, numeric, ratio)
• How composite metrics are calculated (formulas)
• Brief description of the metric's purpose

Important: The catalog is a reference system. Experiments are analyzed based on uploaded data, independently of the catalog.

Types of Metrics

Understanding metric types is critically important for correct statistical analysis. When the system analyzes uploaded data, it applies different methods depending on the metric type.

1. Conversion (Conversion Metrics)

Binary metrics that measure the fact of performing or not performing an action.

Characteristics:

• Take values of 0 or 1 for each user
• Measure the proportion of users who performed the target action
• Result is expressed as a percentage (from 0% to 100%)

Examples:

• registration_rate — proportion of users who completed registration
• purchase_conversion — proportion of users who made a purchase
• button_click — proportion of users who clicked the button

Data format:

user_id,group,registration_completed
user_1,control,0
user_2,control,1
user_3,treatment,1

Interpretation: change in percentage points (10% → 12% = +2 pp)

2. Numeric (Numerical Metrics)

Continuous metrics that measure quantitative indicators per randomization unit.

Characteristics:

• Can take any numeric values
• Measure the average value of the indicator per randomization unit (usually per user)
• Allow tracking the intensity of actions

Examples:

• revenue_per_user — average revenue per user
• session_duration — average session duration
• page_views_per_user — average number of page views per user

Data format:

user_id,group,revenue,session_duration
user_1,control,0,120.5
user_2,control,49.99,340.2
user_3,treatment,129.99,450.8

Interpretation: change in average value per randomization unit

3. Ratio (Ratio Metrics)

Metrics that are calculated not per randomization unit.

Characteristics:

• Calculated as a ratio of two numeric metrics
• Aggregated not by randomization unit
• Require special statistical processing

Examples:

• average_order_value = revenue / orders — average order value (per order, not per user)
• ctr = clicks / impressions — click-through rate (per impression, not per user)
• revenue_per_session = revenue / sessions — revenue per session (per session, not per user)

Creating ratio metrics:

Ratio metrics do not need to be supplied in the data in advance — they are created on the fly in the experiment analysis interface based on existing numeric metrics:

1. Open the experiment analysis section
2. Select "Add ratio metric"
3. Specify the formula: numerator / denominator
4. The system will calculate the metric with correct statistics

Interpretation: change in average value per another aggregation unit

Why Metric Types Matter

The correct metric type is critically important for correct statistical analysis. The system applies different methods depending on the metric type.

Sample size:

• Different metric types require different sample sizes
• Ratio usually requires a larger sample size due to additional variability

Correctness of conclusions:

• Wrong type will lead to incorrect p-values and confidence intervals
• The metrics catalog helps document the correct types for use in future experiments

Viewing Metrics

Open Metrics Catalog via the sidebar menu. You will see a table with columns:

• Type — conversion, numeric, or ratio
• Name — metric identifier
• Description — explanation of purpose (if filled in)
• Formula — how the metric is calculated (if specified)

The table is automatically sorted by type, then by name. You can change sorting with buttons in the header.

Creating a Metric

Click "Add Metric". In the modal window, fill in:

Name (required, up to 50 characters):

• Use snake_case: conversion_rate, revenue_per_user
• Name must be unique
• Recommended to match column names in data

Type (required):

• conversion — binary events (0/1): registration, purchase, click
• numeric — numerical indicators per randomization unit: revenue per user, session time
• ratio — ratios aggregated not by randomization unit: average order value, CTR

Description (optional, up to 300 characters):

• Explain what the metric measures
• Indicate the context of use
• Add notes for the team

Formula (optional, up to 100 characters):

• For conversion: fixation condition (user_registered = 1)
• For numeric: calculation method (sum(revenue) per user)
• For ratio: mathematical expression (revenue / orders)

After filling in, click "Save". The metric will appear in the table.

Editing a Metric

Hover over the metric row — a menu button [...] will appear on the right. Click it and select "Edit".

In the modal window, you can change all fields, including the metric type. Changes apply only to the reference — uploaded experiment data is not affected.

Click "Save" to apply changes.

Deleting a Metric

Only organization administrators can delete metrics.

Hover over the metric row, open the menu [...] and select "Delete". Confirm the action in the dialog box.

Deletion is irreversible, but does not affect historical experiment data where this metric was used.

Importing Metrics

When uploading experiment data, the system automatically analyzes column headers and suggests importing new metrics into the catalog.

Import process:

1. The system compares data columns with metrics in the catalog
2. Finds new metrics (not in the catalog)
3. Automatically determines the type based on data:
   • If all values are 0 or 1 → conversion
   • Otherwise → numeric
4. Shows a modal window with a list of found metrics
5. You choose which metrics to add to the catalog

Important points:

• Import to catalog is optional — data will be analyzed anyway
• After import, it's recommended to check types and add descriptions

The system does not create duplicates — if a metric with that name already exists, it is skipped.

Best Practices

Name standardization: Agree on a unified naming format in the team. For example, use prefixes for grouping: payment_conversion, payment_revenue, payment_refunds.

Documentation: Fill in descriptions for all key metrics. This helps new team members quickly understand what is being measured.

Updating: Periodically audit the catalog — remove unused metrics, update descriptions of outdated ones.