Canonical page: https://tellenze.com/help/delivery-insights
Read your delivery signals
Interpret throughput, cycle time, forecasts, and activity without confusing estimates with promises.
Open the delivery view
Open a Project and choose Insights when delivery insights are enabled. Administrators can use Admin → Insights for a broader workspace view with additional filters. These pages describe observed delivery flow within accessible work.
Start with a specific question, such as “Is review accumulating work?” or “How much completed work is returning?” Use the numbers to identify items and process changes worth discussing with the team.
Choose a useful scope
Set From and To, then choose the relevant Work type. Where available, narrow by Project, Workflow, Milestone, or Assignee. Project member breakdowns and assignee filtering require the project creator or an administrator. Options also depend on available data and enabled workload features. Changing a filter reloads each chart independently; each chart shows its own loading state and can finish while others are still calculating. Refresh requests current calculations. If one chart fails, the others remain available and Refresh retries the failed chart.
Check the displayed date range and timezone before comparing periods. The initial range is approximately twelve weeks. A comparison is more meaningful when both periods use the same scope and have adequate recorded history.
Read the delivery measures
| Measure | What it tells you |
|---|---|
| First shipped | Work reaching completion for the first time in the selected range |
| Rework | Later completion cycles after work has been reopened |
| Scope entered | Work entering scope, alongside completed, removed, and reopened counts |
| Current WIP | Current non-archived open work, grouped by common workflow category |
| Cycle time | Elapsed time from creation to the first completion |
| Open work age | How long current open work has existed |
| Blocked duration | Observed time spent blocked |
| Time in stage | Observed time spent in each common workflow category |
Current WIP and open work age are current snapshots. Moving the historical date range does not turn them into past snapshots. This matters when comparing a recent delivery period with the work still open today.
For example, increasing review-stage time together with a large current review queue suggests checking review capacity or unclear acceptance criteria. Open the relevant work and inspect its blockers and discussion before drawing a conclusion.
Explore the charts
The charts resize with the page in both Project and Admin Insights. Hover over or tap a bar for exact values; keyboard users can focus a chart and use the arrow keys. Use the data table button on any chart for a readable table, including full labels and sample counts. The adjacent chart button returns to the visualization. Charts use the current appearance theme and do not animate.
- Delivery & scope switches between Completions (first shipped and recompletions) and Scope changes (entered, completed, removed and reopened). These are event counts for the selected period, not a historical trend or mutually exclusive portions of a total.
- Work by stage compares current open-work counts across workflow categories.
- Delivery timings and Time in stage compare p50 and p85 on a shared hours axis. Toggle the percentile buttons to focus on one series; at least one remains visible. The table always contains both percentiles and sample counts. Withheld values are omitted rather than plotted as zero.
- Where permitted, Workload distribution lets you switch between Work items, Complexity units, and Blocked items. Member order stays unchanged when switching measures; it is not a productivity ranking.
- AI token usage shows the known total attributed to visible work, with input, output, and Split unavailable portions. Its data table gives exact chart values.
All charts follow the page filters and Refresh control. Empty selections explain why there are no bars. A zero duration remains a valid observation, while Not enough samples indicates withheld values. The historical date range applies to delivery events and observed durations; current WIP, open work age and workload remain current snapshots.
Understand AI usage
AI usage uses the latest report timestamp of each agent run to place its current token allocation in the selected local date range. If a run is amended later, it may move to a later period. Project, workflow, work-type, milestone, and assignee filters select work by its current properties. Only work the viewer can access contributes; workspace-wide Insights remain administrator-only.
The total sums reported token totals allocated to matching work items. Shared runs are divided across implemented items, so their attributed values can be fractional and are labelled approximate. Input and output bars use allocations with a known total and both component counters. Split unavailable includes known totals whose input or output counter is missing. Allocations without a known total are counted as unavailable and are excluded from the total and bars. A blank total means no known total was reported for that selection; it does not mean zero tokens were used. These are client-reported usage figures, not a price or a provider invoice.
For example, one execution using 1 million tokens across two implemented tasks contributes 500,000 tokens to each task and 1 million overall. In Agent context → Agent usage → Run history, reported conversation and execution IDs help identify shared work. The same member cannot submit a second report for an already identified execution; its original report should include all worked tasks. Different executions in one conversation still count separately. Reports without execution identity cannot be automatically checked for overlap, and existing historical duplicates require a reviewed correction.
Interpret percentiles and coverage
Timing distributions show a sample count, p50, and p85. The p50 is the middle of the observed distribution; p85 describes the point reached by approximately 85% of observations. Charts and tables use hours consistently so durations can be compared on the same scale.
Not enough samples means percentile values are deliberately withheld. It does not mean the duration is zero. Broaden the date range or filters if that still answers your question, or wait for more observed delivery.
Expand Metric coverage to see when capture began and the basis used for each measure. Work created before event capture can have a current-state baseline without complete historical transitions. The application does not invent blocked or stage history before that boundary. Metric definitions provides the exact interpretation alongside the report.
Use workload counts carefully
When enabled, Workload distribution lists work count, complexity units, and blocked count by assignee, including unassigned work. These describe distribution; they do not measure individual productivity or rank colleagues. A large count may reflect coordination responsibilities, work size, or blocked dependencies.
If Insights is missing, check feature availability and your page permissions. If results look unexpectedly small, check filters, current source access, archived work, and the coverage dates. If a chart cannot load, use Refresh after resolving the displayed error. A milestone’s forecast is a separate estimate of future delivery and should be read alongside these observed measures.