Risk radar
Route: /org/tasks?panel=risk · Permission: View risk radar (Manage risk radar to recompute and change settings)
Module: Task Management · Entry points: Tasks toolbar → Risk · task badges and Why? · Roadmap milestone chips · Project settings → Risk Radar
What this is. A daily picture of which tasks, assignees, sprints and milestones are about to slip — and, on request, AI-ranked interventions. What it is for. Catching a slip while there is still time to reassign, split, descope or extend. The problem it solves. Overdue is too late. The radar flags schedule pressure, velocity gap, stale, blocked and milestone burn before the date passes.
1. What the Risk radar is
The radar is computed from your project's own data — due dates, remaining hours, time logged, activity, dependencies, milestones and who is assigned to what. It costs no AI credits: the signals are arithmetic, recomputed every morning by a platform job and whenever you press Recompute. AI is used only when you ask for interventions.
| Signal | Fires when | Severity |
|---|---|---|
| Overdue | The due date has passed | High (critical if also blocked) |
| Schedule pressure | Remaining hours exceed the working hours left before the due date | Low → high by ratio |
| Velocity gap | At the recent pace (hours logged over the look-back), the task lands after its due date | Low → high by overshoot |
| Stale | In progress with no activity for the configured number of days | Medium (high if due soon) |
| Blocked | A predecessor is still open | Medium (high if the predecessor is late) |
| Unassigned near due | Nobody is assigned and the due date is within the due soon window | Medium |
| Assignee overloaded | A member's remaining hours over the horizon exceed their capacity × the overload ratio | Low → high by load |
| Milestone burn | The pace needed to finish the milestone's open tasks exceeds the recent pace | Low → high by ratio |
| Milestone overdue | The end date has passed with open tasks | High |
Every task, assignee and milestone gets a score and a severity (none, low, medium, high, critical); the project's health is the worst milestone.
The radar reads the currently selected project. With no project selected, the Tasks page shows no Risk button.
2. How the radar fits together
3. Using the radar — step by step
Step 1 · Open the radar
Tasks → Risk. The button carries a count of medium-and-above items, coloured by the project's health. The panel has three sections: Milestone health (cards with tasks and hours remaining, days left and burn), At-risk tasks (severity pill and signal chips — hover a chip for the value behind it and the formula), and Team load (a bar per member over the coming two weeks).
Click a task to open it. Recompute today's snapshot now refreshes the picture without sending any notification (Manage risk radar).
Step 2 · Read the badges
At-risk tasks show a coloured badge in the list, on kanban cards, in the table's Risk column and in the task header. Why? lists the signals that fired. On the Roadmap each milestone shows a health chip; hover it for the burn figures.
Step 3 · Ask AI for interventions
Type optional guidance (protect the release sprint, Priya is on leave next week) and press Ask AI for interventions. You get a headline, an overall health, a short narrative, a ranked list of interventions — each naming a real task or milestone, an action (reassign, split, descope, extend, unblock, escalate), the rationale, expected impact and confidence — and a watchlist.
The advice is cached until the risk picture changes: asking again on the same day costs nothing.
Step 4 · Apply an intervention
Every Apply goes through a confirmation and then the normal task action, with your normal permissions:
| Action | What Apply does |
|---|---|
| Reassign | Changes the assignee (needs Assign or Change Assignee) |
| Extend | Sets the new due date (needs Edit or Change Due Date) |
| Split | Opens Suggest sub-tasks for that task |
| Descope · Unblock · Escalate | Adds a comment on the task with the rationale (needs Edit) |
A greyed Apply carries a tooltip naming the missing permission.
Step 5 · Tune the settings
Project → Settings → Risk Radar (Manage risk radar).
4. Field reference
| Setting | Default | Meaning |
|---|---|---|
| Daily capacity | 8 h | Working hours per person per day |
| Stale after | 5 days | Idle days before an in-progress task is stale |
| Velocity lookback | 14 days | Window used to measure pace |
| Due soon | 3 days | Window for unassigned near due |
| Load horizon | 14 days | Window for assignee load |
| Overload ratio | 1.2 | Load above which a member is overloaded |
| Severity thresholds | burn 1.25 / 1.75 / 2.5 · pressure 0.8 / 1.0 / 1.5 | Low / medium / high cut-offs (must ascend) |
| Notify project managers | On | Master switch for the daily alert |
| In-app bell · Email digest | On | Delivery channels |
| Minimum severity to notify | Medium | Items below it never alert |
| Auto-run AI advisor | Off | Let the daily job also ask AI (spends credits) |
5. The admin contract
- Permissions: View risk radar (panel, badges, chips, AI advice); Manage risk radar (recompute, settings). Applying interventions uses the task permissions above.
- Daily alerts are sent by the platform job Task Risk Radar Daily, which an operator enables per environment. Recipients are the project's managers and leads. An item alerts once when it appears and again only if it gets worse; an item that recovers stops alerting.
- Credits: about 600 per interventions request (ceiling 2,400); nothing for the radar itself.
5b. Why is this at risk? — the explanation
Every at-risk task, milestone and person now explains itself in plain English, right where the severity badge is:
- a summary naming what is wrong and the number behind it
- the other contributing signals, one line each
- what to do next — a fixed, sensible step per signal
- a 30-day score trend beside it, so you can see whether it is worsening or recovering
This is computed, not generated. It costs nothing, works when AI credits have run out, and cannot invent a figure — every number in it came from the engine.
Explain in detail is the optional paid second click: a longer narrative on that single item, grounded in the same facts plus its score history. It never proposes actions — ranked interventions remain the advisor's job. The free explanation stays on screen above it either way.
5c. Preview an intervention before applying it
Beside Apply on any suggested intervention is Preview impact. It recomputes the entire risk picture with that change applied and shows you:
- the project's severity and score, before → after
- how many tasks are at risk, before → after
- how each person's load moves
- which individual tasks change verdict
Nothing is written. The engine's inputs are copied, the change is applied to the copy, and the result is recomputed. A split is labelled approximate, because dividing the remaining hours evenly is a guess about work nobody has broken down yet.
Reassign, extend, descope and split can be previewed. Unblock and escalate cannot — they post a comment, which changes nothing the engine measures.
5d. Applying an intervention is recorded
Applying a suggestion goes through a server-side path that performs the same edit, re-checks the same permission a manual edit needs, and writes a line to the AI change ledger with the values before and after. So "an AI suggested this" is never lost, and the task's history is not just "a person edited a field".
6. Downstream
- Delivery forecast adds a P50/P80 finish date per milestone and the Forecast miss signal.
- Portfolio risk rolls every project you are on into one page.
- AI governance records every applied intervention.
- Progress reports use the milestone health from the latest snapshot.
- Applied interventions are ordinary task changes — they appear in the task's activity and notify as usual.
7. Don't confuse this with…
- Task analytics — describes what happened; the radar predicts what is about to.
- Service Desk SLAs — measure response and resolution times on tickets, not delivery risk on tasks.
- Plan with AI — creates work; the radar watches it.
8. Troubleshooting
| Symptom | Cause | Fix |
|---|---|---|
| The explanation is missing on a row | The stored snapshot predates Wave 3 | Press Recompute — explanations are computed on read from the stored signals |
| "Preview impact" is absent | That verb changes nothing the engine measures (unblock, escalate) | Expected |
| No Risk button | No project selected, or you lack View risk radar | Pick a project; ask for the permission |
| Radar says "computed at" yesterday | The daily job ran; nothing recomputed since | Press Recompute (Manage risk radar) |
| A task is stale but should not be | Stale after is too short for your cadence | Raise it in Project settings → Risk Radar |
| Everyone is "overloaded" | Daily capacity is lower than reality, or estimates are inflated | Adjust capacity or the overload ratio |
| No daily emails | The platform job is disabled, or notifications are off for the project | Ask your administrator; check the Notifications settings |
| "cached" advice after changes | The picture has not been recomputed | Recompute, then ask again |