Detect project risks with AI

AI risk detection reads a project’s current state, its schedule, task statuses, dependencies, and workload, and proposes concrete risks: things like a critical-path task with no assignee, a milestone with everything bunched in the final week, or one person carrying half the plan.

Run an analysis

1. Open the project, then Analytics > Risks.
2. Select Run Analysis. It writes the detected risks straight to the project’s risk register. It replaces only risks Onplana added on its own (an earlier analysis, the background scan, an import, a template or an AI-drafted plan) that nobody has edited, accepted or dismissed; risks you or an agent added, and any risk a person has touched, stay. Each proposed risk comes with a severity, a category, and a recommendation.
3. Act on each risk. On the Risks tab, Promote to Issue drafts an issue from a risk that has happened, and Dismiss takes one that does not apply off the open register (the dismissal stays on record, though a later run can detect the same thing again). On the Gantt chart, the risk panel above the chart also offers Accept & create task, which creates a mitigation task and closes the risk.

Tip

Every detected risk lands on the register the moment analysis runs, and feeds the project’s health indicators, whether or not you’ve accepted it yet. A risk that materializes into a real problem can be promoted to the issue log from the Risks view.

Add or edit a risk yourself

Not every risk comes from an analysis, and you don’t need the AI to keep a register. Adding and editing risks works on every plan, and keeps working when your workspace has no AI tokens left. The full guide, for teams that keep their register by hand, is Keep a risk register.

1. Open the project, then Analytics > Risks.
2. Select Add risk. Describe what could go wrong, pick a category (Schedule, Resource, Budget, Scope, or External) and a severity (Low to Critical), and optionally write a recommendation: what would reduce the risk, or what to do if it happens.
3. To change a risk later, select Edit on its card. You can change any of the four fields; clearing the recommendation removes it.

Severity is the only rating a risk carries. There is no separate probability, impact, or owner.

You need permission to add content to the project, the same permission that lets you log an issue. People without it can read the register but don’t see Add risk or Edit. A risk that has been accepted or dismissed is closed and can no longer be edited.

Risks your agents record by hand

When a connected agent spots a risk during its work, it records it as a real risk-register entry (description, severity, category, and an optional recommendation) instead of burying it in a comment. Hand-recorded risks, whether you or an agent wrote them, work on every plan. Recording a risk never touches the risks already on the register, and a later analysis leaves hand-recorded risks in place.

FAQs

Does the analysis change my project? Yes. Running an analysis writes the detected risks straight to the register, in place of the earlier automatic ones nobody has touched (see step 2 above). Accept and Dismiss don’t decide whether a risk is tracked, they decide what happens next: Accept (on the Gantt chart’s risk panel) creates a mitigation task and closes the risk, Dismiss takes it off the open register and records the dismissal. Dismissing does not stop a later run detecting the same risk again.

How is this different from project health (RAG) status? Health status is a quick computed indicator. Risk detection produces specific, explainable items you can act on, each with its own severity and category.

Which plan includes risk detection? Risk detection is part of the advanced AI features, available on the Pro plan and above.

Best practices

1. Run analysis after major changes. New tasks, schedule shifts, member adds. Fresh analysis catches new risks before they materialize.
2. Accept selectively; dismiss irrelevant. Don’t accept every suggested risk. The dismiss feedback improves future suggestions.
3. Promote materialized risks to issues. The risk→issue handoff is a first-class pattern. See Track issues in the issue log.
4. Re-run weekly during execution. Risk landscape changes weekly. Weekly cadence keeps the radar fresh.
5. Review with the team. Don’t act on AI risks unilaterally. The team knows context the AI doesn’t.

Troubleshooting / common pitfalls

• Analysis returned no risks. Either project is healthy or analysis lacks context. Verify project has tasks, dates, and assignees.
• Same risk suggested again after dismiss. Dismiss is per-suggestion; recurring patterns may surface a semantically similar risk.
• Risk severity feels off. AI’s severity is an estimate. Select Edit on the risk and change it before you accept it; an accepted risk is closed and can’t be edited.
• Analysis timed out. Large projects (1000+ tasks) may exceed time limit. Filter to subset.
• Risk category unexpected. AI categorizes by detected pattern. Change the category with Edit while the risk is still open.
• AI tokens used up. Run Analysis stops, because detection is an AI call. The risks already on the register stay readable, and you can still accept or dismiss them. See Understand your AI token budget.

How this combines with other features

• Risks + Issues. Promote risks to issues when they materialize. See Track issues.
• Risks + Project health. Risks feed RAG health computation.
• Risks + EVM. Critical risks affect SPI/CPI predictions. See Read the EVM dashboard.
• Risks + Workflows. Workflow can notify on new risk acceptance.

Coming from another tool

Tool	Mapping
Microsoft Project Online risk register	Direct concept
ServiceNow PPM risk module	Direct
Custom risk spreadsheet	Onplana automates the detection
Custom Excel risk matrix	Direct
Jira risk plugin	Direct

Related

• Keep a risk register, adding and editing risks by hand on every plan
• Track issues, promote risks to issues
• Read the EVM dashboard, risk impact on cost/schedule
• Chat with Onplana AI, ad-hoc risk questions