Workban

Statistics · For management and team leads

What got done, by whom and how long it took.

The figures come along on their own: from the status changes agents and people make anyway, and from the sessions of the agents. Nobody runs a stopwatch, nobody keeps a spreadsheet. Every period stands next to the one before, for a single project if you like.

Wk 39Sample data
Done
47+24%
Wk 38: 38
AI share
81%
Wk 38: 74%
Lead time
1 d 3 h
Wk 38: 1 d 9 h
AI cost
$186.30
Wk 38: $141.60
  • 8 key figures
  • 11 charts
  • Compared with the period before
  • AI cost per agent and project

Example

One week, next to the week before.

This is how a weekly report reads. Hover over a row and the change against the week before appears. The figures are sample data.

Weekly report

Wk 39

All projects, compared with the week before

47done

+24%Wk 38: 38

Weekly report, this week next to the week before, with sample data
Wk 39Wk 38
Done per day
Mo97
Tu118
We69
Th86
Fr127
Sa11
Su00
Done by
Claude Code2921
Codex97
no agent910
AI share81%74%
Times
Work time, median38 min44 min
Lead time, median1 d 3 h1 d 9 h
Rework6%11%
Projects
Web-App2115
API1714
Mobile99
Tokens and cost
Tokens58.2M44.1M
Cost, estimated$186.30$141.60
Avg. per task$4.90$5.06
Sample data

Key figures

Eight key figures, each with its trend and comparison.

Every tile shows the value for the chosen period, the change against the comparison and the trend as a small curve. Above them, “At a glance” sums up the biggest changes in plain sentences.

Done
Tasks finished in the period.
Created
Tasks added in the period.
Working time
The time tasks spent in progress or in testing. Waiting on hold does not count.
AI share
The share of done tasks an agent worked on.
Working time per task
The median: half the tasks were done faster, the other half slower. A few outliers do not skew it.
Lead time
The median from created to done, waiting included. This is how long a piece of work is on its way in total.
Rework
Tasks that went back from testing into progress.
AI cost
Estimated cost from the measured tokens, as a total and as an average per task.

Periods

Every period, fairly compared.

You pick the period and what it is compared with. The choice sits in the address: a link shows your team or your management exactly the same view. A filter narrows everything down to one project.

Period

  • This or last week
  • This or last month
  • This or last quarter
  • This year
  • The last 7, 30 or 90 days
  • A range of your own, up to 366 days

Compared with

  • the period before
  • the same period last year
  • a range of your own
  • nothing
01No stopwatch.
The times come from the status changes the agent makes anyway. Waiting on hold does not count as work.
02A fair comparison.
While the week is still running, last week's figures only count up to the same point in time. Otherwise a Monday morning would compete with a whole week.

Charts

Eleven charts, each one also a table.

One view for every question. One click turns a chart into a table with every value. The app itself speaks German, the names below are translated.

Done per day, week or month
Finished tasks, section by section next to the comparison period.
Done, cumulative
When the line runs above the dashed one, the team is further along than in the comparison period.
Inflow and outflow
Created against done, with the balance of open tasks in the period.
Open tasks
Tasks not done at the end of each section. When the line rises, more comes in than goes out.
Who does the work
Done tasks by person and agent.
Projects compared
Done tasks or working time per project.
People compared
Done tasks or working time per person. Agents count for the person they work for.
Working time distribution
The median and the time 90 % of tasks stay under. The further left, the more predictable.
AI cost per day, week or month
Estimated at API list prices, with the tokens of the period.
Current state
All tasks by status, whatever the period.
Contributions per project
Who got how much done, as a grid of squares: the darker, the more.

AI cost

Tokens measured, cost estimated in the open.

When an agent closes its task, the CLI reads the log of its session. Only what falls into this task's working time counts. So every task carries the tokens it took.

  • For Claude Code its session file, for Codex its rollout log
  • Tokens split by model, cache writes by how long they last
  • Cost at the list prices of the API. Workban leaves out subscriptions and discounts, so the real bill can be lower
  • Totals per task, per agent and per period, and with the project filter per project
  • Other agents report their own figure when they close a task

#42

Fix the login redirect after the magic link

Project
Web-App
Agent
Claude Code · Mara

  1. 09:12opened

  2. 09:13In Progress

  3. 09:41Testing

  4. 09:52On Hold

    Pause 18 min

  5. 10:10In Progress

  6. 10:24Done

Working time (without pause)
53 min
Tokens
1,284,000
Cost, estimated
$4.10
Sample data

Origin

Where the figures come from.

No stopwatch and no estimates by hand. The times follow the status changes, with fixed rules for pauses and gaps.

Running time
The clock only runs while a task is in progress or in testing.
Pauses
While a task waits on hold, the clock stands still.
Gaps
A gap without any activity counts as 45 minutes of work at most.
Left alone
After four hours without activity, a task moves to On Hold by itself.
Correction
If a time is wrong, it can be corrected on the task or through the CLI. The history shows who changed it.
API
The API returns the same figures for your own reports, with a personal token.

Next

The same figures hang on the wall.

Get started

Try it with your next task.

Workban is free. Your workspace is ready in a minute, and the first task appears as soon as your agent gets going. The app itself speaks German for now.