Forecastwell documentation

What Forecastwell does

Forecastwell turns your Jira history into flow metrics and forecasts: a cycle time scatterplot with percentiles, work item age, throughput, a cumulative flow diagram, and Monte Carlo answers to "when will these items be done?" and "how many will be done by this date?". It works for any JQL, on a page (Apps > Forecastwell, or Forecastwell in a project's sidebar) and as two dashboard widgets.

Set up in three minutes

Open Apps > Forecastwell and click New view. From a project's sidebar, Forecastwell starts with a draft for that project.
Enter the JQL for the work you want to measure (one project, a board's filter, several teams).
Click Copy stages from a board and pick the team's board. Its columns become stages, with "Work starts" on the first column with an in-progress status and "Done" on the first column where every status is done. Adjust if needed.
Pick how many days of finished work to learn from (90 is a good start), tick Count working days only if your team works a fixed week, and click Create view.
Optional: add the Flow metrics (Forecastwell) or Monte Carlo forecast (Forecastwell) widget to a Jira dashboard and pick the view.

Views and stages

A view is a JQL query plus your workflow as ordered stages, like board columns. Each stage holds one or more statuses; a status can be in one stage only.

Work starts is the commitment point: work counts as started the first time it reaches this stage or any stage after it.
Done is the delivery point: work counts as finished when it reaches this stage (or a later one) and stays there. Reopened work goes back into progress.
Work in progress is exactly the stages from "work starts" up to (not including) "done".
Statuses in no stage are never counted as in progress. Work sitting in them is listed in a note above the charts so nothing disappears silently.
Leave out (JQL), for example resolution = Duplicate, removes matching work from every chart. Forecastwell finds those items with a separate search and subtracts them by id.
Personal or shared: personal views are yours alone. Shared views are visible to everyone (created by Jira admins, or by anyone if an admin allows it in settings). Everyone's numbers still follow their own Jira permissions.

How the numbers are counted

Days: whole days in the settings time zone, both ends counted. Work started and finished on the same day took 1 day. With Count working days only, only workweek days that are not holidays count (never less than 1).
Cycle time: from the start to the finish of finished work.
Work item age: from the start until today for work in progress.
Work that skips the start stage (straight from To Do to Done) starts when it finishes: 1 day.
Blocked days: days on which Jira's Flagged field was set at any moment, between start and finish (or today). Tick Leave days with a Jira flag out of cycle time and age to subtract them.
Percentiles use the nearest-rank method: the 85th percentile is a cycle time that at least 85% of finished items met, and it is always a real observed value.
Throughput: items that reached the done stage per day, zero days included.
History with unreadable dates (some imports) is skipped and counted in a note.

The charts

Cycle time scatterplot: one dot per finished item, by finish date and cycle time. Items with the same finish day and cycle time share one bigger dot with a count; click to open them. Dashed lines are the 50, 70, 85 and 95% percentiles.
Work item age: one column per in-progress stage, one dot per item, coloured by how its age compares with your percentiles. Red rings mean the item is flagged now. The table lists the ten oldest.
Throughput: finished items per week or per day, with the average.
Cumulative flow: items in each stage at the end of every day. Work finished before the range is left out, so the done band starts at zero. Stages marked * are work in progress.

Forecasts

Forecastwell replays your daily throughput thousands of times (10,000 by default): each simulated day draws one real day from the window at random, quiet days included.

When will it be done? Enter the number of items left, or JQL that finds them (counted when you run the forecast). You get the date by which 50, 70, 85 and 95% of simulated futures finished. With working days on, dates skip weekends and holidays.
How many by a date? Pick a date; you get the number of items finished in at least 50, 70, 85 and 95% of simulated futures.
The same data and question always give the same answer (the simulation is seeded from them), so a dashboard doesn't change its forecast on every reload.
Forecasts assume the future looks like the window. After a team change, shorten the window.

Dashboard widgets

Flow metrics (Forecastwell): pick a view and one chart (scatterplot, work item age, throughput, cumulative flow).
Monte Carlo forecast (Forecastwell): pick a view and a question; for "when", give JQL for the work left (counted each time the dashboard loads) or a number.
Each viewer sees the chart computed from their own Jira permissions.

Freshness and big views

The first time a view loads, Forecastwell reads the history of every work item it covers; big views continue in the background and show a progress bar. After that, a view refreshes at most every 15 minutes on its own and only re-reads items that changed. Refresh updates now. A view covers open work plus everything updated within its window, up to the item limit in settings (5,000 by default; most recently updated first).

Settings (Jira admins)

Jira settings > Apps > Forecastwell: time zone for day boundaries, working days, holidays, the item limit per view, and who may create shared views.

Moving from Data Center

Data Center teams often measured flow in an app or with exported spreadsheets. In Cloud, create one view per team: use the board's filter as the JQL, copy the stages from the board, and set the window to match the period you used to report. Compare the 85% cycle time with your last Data Center report for a week before you share forecasts.

Questions

Does Forecastwell change anything in Jira? No. It only reads.
Hours per status per person? That is Dwellwise, our time-in-status app. Forecastwell counts whole days for flow metrics and forecasts.
Where is my data? In your site's Forge storage, hosted by Atlassian. Nothing leaves Atlassian.
Story points? Forecastwell forecasts in items, from throughput.