Skip to main content

AI Usage > Breakdown

Written by Tom Williams

Dataset: AI Usage

Entity: AI Usage Measurement

Field ID: breakdown

Type: Select list

Description: The cut of the provider report the record comes from. Each value names the dimensions the record is broken down by, so it also tells you which dimensions carry data and which are empty. Possible values are:

  • TOTAL the subject's day total. Surface type, model, client and language are empty.

  • SURFACE split per surface. Model and language are empty.

  • SURFACE_MODEL split per surface and model. Language is empty.

  • CLIENT split per client. Surface type, model and language are empty.

  • LANGUAGE_SURFACE split per language and surface. Model is empty.

  • LANGUAGE_MODEL split per language and model. Surface type is empty.

Note: A single provider report contains several overlapping cuts of the same activity. Always filter on a single breakdown: summing across breakdowns double counts.

Source: Calculated

Transformation logic: Assigned from the part of the provider report a record comes from.

App Mapping

GitHub Copilot

all six values

Anthropic Claude (coming soon)

SURFACE and SURFACE_MODEL

Cursor (coming soon)

TOTAL, SURFACE and SURFACE_MODEL

OpenAI Platform (coming soon)

SURFACE_MODEL only

OpenAI Codex (coming soon)

TOTAL and SURFACE

Reporting Use Cases

The Breakdown field is the second field that must appear in almost every AI usage filter. It says which slice of a provider's daily report a record came from.

  • Always filter on one value: A user's day total and their per-surface records describe the same work. Summing across breakdowns counts it several times.

  • Headline volume: Filter to TOTAL for the cleanest per-subject day figure, where the provider publishes one.

  • Where AI is being used: Filter to SURFACE and group by surface_ref to see the split between completions, chat, agents, the command line and code review.

  • Model and token reporting: Filter to SURFACE_MODEL. This is the only cut carrying model attribution, and for most providers the only one carrying tokens.

  • Language reporting: Filter to LANGUAGE_SURFACE or LANGUAGE_MODEL to see which parts of a codebase AI tooling is used on. GitHub Copilot is the only app publishing these, and they carry no prompt counts.

  • Tool and version reporting: Filter to CLIENT to split usage by IDE or command line, and to see plugin and client versions.

  • Empty dimensions are expected: A record's breakdown tells you which dimensions it carries. An empty model on a SURFACE record is correct, not missing data.

Related Articles
Did this answer your question?