Guide

Usage and outcome reports

Where to see the tokens your coworkers used and what they cost, how people tag a conversation's outcome, and how cost per completed task is worked out.

4 min read

Who sees which reportCopy link to this section

  • My usage, under Settings, is for everyone: the runs you started this month on every Bot, what they used and what they cost, against any cap set for you. Nobody else's spending is shown there.
  • Usage, in the admin console's Usage & billing group, shows the whole workspace. Administrators, and people whose role includes View usage, can read it; only administrators can change caps, prices and the exchange rate.
  • Reports and Insights, in the same group, are for administrators.

Note

Months and days are UTC calendar months and days everywhere in these reports.

Tokens and cost for a monthCopy link to this section

On Usage, choose a Month: the last 12 months are offered. The totals at the top give Cost, Input tokens, Cached input tokens, Output tokens and Runs. Below them, By Bot lists each coworker's spending against its own Monthly cap, and By model shows the same month by the model that served it, whichever Bot asked.

Tokens by day stacks every Bot together for each day of the month, split into input that was not cached, cached input, output and reasoning. Show as a table gives the same numbers as rows. Previous months lists the last six months.

Each Bot with a cap gets an Estimate: the month-to-date spend divided by the days elapsed, times the days in the month, and how many days remain until its cap at the last 7 days' average. At 80% of a cap administrators are warned; at 100% the Bot is paused until the next month or until the cap is raised.

The cost trendCopy link to this section

Cost by month draws the last 12 months for the five costliest Bots, with the rest together as Other.

Warning

Every month is priced with the price table as it is today. A month from before a price change shows the new price, not what was billed.

How cost is worked outCopy link to this section

Input, cached input and output tokens are each multiplied by their price per million tokens from Prices. A blank cached input price uses the input price, and reasoning tokens add nothing. The prices shipped with the workspace are estimates: replace them with what you actually pay. Changing a price restates every month and is recorded in the audit trail. A model with no price shows Not priced and is left out of the cost.

Yen rate is optional: with a rate you enter, costs are also shown in yen. Nothing looks a rate up for you.

Failed or stopped runs count nothing, and the workspace's own background model calls are not charged to any Bot.

Tagging a conversation's outcomeCopy link to this section

In a conversation's header, Outcome sets one of three tags: Done, Partly done or Not done. Anyone who can open the conversation can set or change it, and Clear outcome removes it. A conversation holds one tag at a time, it counts toward the conversation's first Bot, and setting or clearing it is recorded in the audit trail.

Note

Outcomes are tagged by people, never by the coworker itself, so the reports only know what someone marked.

Cost per completed taskCopy link to this section

On Reports, Outcome rate per Bot gives, for the month chosen, how many conversations were Tagged for each Bot and the Done rate: the share marked Done. A conversation counts in the month its tag was last set.

Cost per completed task divides each Bot's Estimated spend for the month, priced as on Usage, by the conversations tagged Done that month. A Bot with no Done tags shows No Done tags instead of a number.

On Usage, Tokens per completed task and Cost per conversation outcome give the same view in tokens, and the average cost of conversations tagged Done, Partly done and Not done.

Downloading usageCopy link to this section

Download usage, on Usage, writes a CSV file of usage and cost by day, Bot, person and model for the period between From and To, at most a year apart. Download with Japanese headings gives the same rows with Japanese column names. To share the file more widely, tick the box that names people Person 1, Person 2 and so on, with no addresses. Every download is recorded in the audit trail.

For caps and what happens when one is reached, see Budgets and cost control.

Last updated