People represented

Positive users

Credits per person per week

Credits per 1,000 tokens

Consumption baselines by function

Baseline table

Population-level highlights

Consumption volume and consumption mix are separate. The highest-credit people are not automatically the most reasoning-intensive, and the standard Power User definition is based on sustained actions rather than credit cost.

The clearest Ways of Working difference is channel mix. Higher-credit users lean more towards chat and less towards email. Total collaboration volume is more sensitive to function and seniority than to consumption itself.

Reasoning-intensive use carries a weak workload signal in this simulation. Within the same broad credit band, reasoning-intensive users have slightly more after-hours collaboration and a longer collaboration span. This is an association to investigate rather than evidence of harm.

How to read credit intensity

Credit intensity is the number of credits consumed per 1,000 tokens. It is a cost-mix measure rather than a productivity or value measure. Lower intensity means the observed token mix consumes fewer credits. It does not establish that the work was more efficient, because the report does not observe output quality, task difficulty or business value.

Concentration of token consumption across users

Concentration bands

Shape of the token-volume distribution

Why a concentration curve and a log histogram?

Token and credit consumption in Copilot data is typically right-skewed: most people use a modest, fairly similar amount, while a small group of heavy users account for a disproportionate share of total volume. A boxplot or violin plot would compress this shape into a small number of summary points and understate how concentrated the top tail is.

The concentration curve above answers the practical question directly: what share of tokens does the top share of users account for? The histogram (log scale) shows the shape that produces that concentration, and the gap between the median and mean lines is itself evidence of the right skew.

This has a practical implication for interpreting averages elsewhere in this report. A mean such as “credits per person per week” can be pulled upward by a small number of heavy users, so organisational and segment comparisons should be read alongside this distribution rather than in isolation.

Ways of Working metrics by credit-consumption group

Credit groups

Collaboration mode mix by credit-consumption group

Workload and network

Consumption and collaboration volume by function

Reading the chart

The upper-right functions combine high token volume with high collaboration volume. That does not imply that one causes the other. Their role mix, seniority and working practices may raise both.

The colour adds a third dimension. A darker point consumes more credits for the same number of tokens. This can indicate a more expensive reasoning mix, but it may also reflect harder work. The appropriate follow-up is to examine task mix and value, rather than treating lower intensity as automatically better.

Workload and credit intensity by function

Function comparison matrix

Credit volume by credit band and reasoning-token profile

Profile definitions

The label reasoning-intensive describes the measurable consumption pattern without assuming recklessness. Business value and task difficulty are needed before judging whether the additional expense was warranted.

Workload by reasoning-token profile within credit bands

What work consumes the credits?

Credit-volume distribution within usage segments

Why keep both?

The standard usage ladder answers an adoption question: is activity frequent and sustained? Credit volume answers a resource-consumption question: how much credit is used? A Power User can have a standard-cost token mix and moderate credit volume, while someone with fewer actions can consume more credit through longer or more reasoning-intensive requests.

The report therefore keeps usage segments as a useful supporting view without letting them substitute for the unique information in the consumption query.

Segment baseline

Interpretation guardrail

Neither segment nor credit group is a performance category. They describe observed use. The table and chart align activity to the latest observation week, while the segment reflects sustained behaviour over its trailing 12-week window. Results should be reported for privacy-safe cohorts and interpreted alongside role, task mix, entitlement and organisational context.

Join and aggregation workflow

Core rules

Simulation design

This dashboard is a demonstration of analytical possibilities. Every person, token, credit and Ways of Working value is synthetic.

The simulation creates 1,200 people across eight functions and 26 weeks. Function and level affect both AI appetite and some Ways of Working measures so that raw associations contain realistic compositional confounding. Credit consumption is derived from input, output and reasoning tokens using synthetic relative weights. Reasoning tokens have the highest weight, which makes credit intensity responsive to token mix rather than merely to volume.

The following effects are deliberately modest:

• Higher consumption is associated with more chat and less email as a share of collaboration modes.
• Total collaboration volume is raised mainly by organisational composition.
• Internal network size has a small residual association after composition is considered.
• Reasoning-intensive use has a weak positive association with after-hours collaboration and collaboration span.
• External network size is a null.

Synthetic source assets

The uncompressed source-shaped CSVs sit under:

examples/utility-r/_data/consumption/
  consumption-query/
  person-query/
  reference/

They are intended for demonstrations, workshops and adaptation of the template. They contain no customer data or real identifiers.

Definitions and limitations

Credit intensity means credits consumed per 1,000 tokens. It is preferred over “token cost-efficiency” because the report observes resource consumption rather than the quality or value of the output.

Reasoning-intensive means the upper quartile of reasoning-token share among people with at least 10,000 observed tokens. It is preferred over “reckless” because expensive reasoning can be appropriate for a difficult task.

The example Consumption export that motivated this template did not expose input, output and reasoning-token fields. A real implementation can only reproduce the reasoning profile when those governed token fields and applicable credit weights are available. If they are absent, omit the profile rather than inferring it from total credits.

The dashboard shows associations. It does not establish that AI consumption caused a working pattern, and it does not observe output quality, task difficulty or business value.

Replacing the synthetic data

1. Replace the CSVs in _data/consumption with schema-aligned governed exports.
2. Validate the identity crosswalk and disclose the match rate.
3. Confirm whether missing consumption rows mean zero activity, suppression or ineligibility.
4. Recompute credit weights from the applicable governed rate card.
5. Keep person-level averaging so each person carries equal analytical weight.
6. Retain the privacy threshold and the organisational composition checks.