Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
An agent outcome is a completed result of security work performed by a Perception agent. Examples include an alert verdict, an investigation report, or a posture assessment.
As a security administrator, use this article to understand what each Perception agent outcome represents, what its output includes, and what factors can affect its cost.
Agent outcome types
The unit differs by outcome type. An alert verdict relates to one alert, while a report can bring together multiple findings or recommendations.
| Outcome in the calculator | What one outcome represents |
|---|---|
| Alerts triaged | One alert verdict classified as a true positive (TP) or false positive (FP). |
| Investigations | One investigation report about an incident or alerts |
| Detection reports | One report of custom detections from threat intelligence specific to your tenant |
| Threat intelligence reports | One TI report with ATT&CK-mapped TTPs, indicators, vulnerabilities, hunting queries etc. along with remediation recommendations |
| Posture assessments | One report that includes prioritized risks, supporting evidence, affected assets, and remediation plan |
| Reconnaissance runs | One report on exposed assets and attack paths. |
Alerts triaged
An alert triage outcome is one verdict for one successfully triaged alert. The Triage Agent classifies the alert as a true positive (TP) or false positive (FP).
Each verdict includes:
- A written explanation of the classification.
- A decision graph showing the reasoning and supporting evidence.
The outcome is the verdict for the alert, not the incident that contains it or each piece of evidence reviewed. Triage determines whether an alert represents a real threat or a false alarm. A deeper investigation can follow to establish the attack story, impact, and response.
What affects the cost
The cost of triaging an alert can vary based on:
- Alert complexity: The evidence available and the analysis needed to reach a verdict.
- Reasoning depth: Straightforward alerts require fewer analytical steps. Ambiguous alerts require more evidence evaluation and reasoning.
- Workload: Different alert types require different context, signals, and analytical approaches.
For example, an alert with clear evidence might require fewer analytical steps than one with incomplete or conflicting evidence.
For capabilities and supported alert types, see Triage Agent.
Investigations
An investigation outcome is one report for one completed investigation. The Attack Investigation Agent correlates alerts, incidents, entities, and underlying telemetry to reconstruct what happened, the scope, and the impact.
Each report includes:
- A verdict and executive summary.
- An attack graph and timeline.
- Compromised entities, indicators of compromise, and observed MITRE ATT&CK techniques, with supporting evidence.
- Prioritized response recommendations explaining what to do, where to act, why it matters, and operational considerations.
The outcome is the investigation report, not each alert, entity, or timeline event examined during the investigation.
What affects the cost
The cost of an investigation can vary based on:
- Distinct activity: Multiple alerts can describe the same behavior. Different behaviors and affected entities require additional assessment, so alert count alone doesn't determine the work involved.
- Investigation breadth: New findings can extend the investigation beyond the starting incident to other identities, devices, applications, and cloud resources.
- Evidence to examine: Connecting activity across data sources and time periods requires analysis. More evidence to interpret means more investigative work.
- Uncertainty: Ambiguous or conflicting findings require additional checks to distinguish legitimate activity from threats and establish what happened.
For example, an isolated alert with limited activity and few entities might require less analysis than one that uncovers activity across multiple identities, devices, applications, and time periods.
For report details, see Attack Investigation Agent.
Detection reports
A detection report outcome is one report created by the Detection Authoring Agent. The agent turns threat intelligence into custom detection recommendations specific to your tenant. It maps attacker techniques to behavioral patterns and writes detection queries using Microsoft Defender XDR and supported Microsoft Sentinel data.
Each report includes:
- Proposed detection logic, configuration settings, and supporting explanations.
- Estimated alert volumes based on testing against tenant data and up to 14 days of history.
- Tuning recommendations to help reduce noise before deployment.
The outcome is the report, not each proposed detection or each alert a deployed rule might generate. One report can contain multiple proposed detections. A successfully completed report can also contain no proposed detections.
The agent doesn't deploy or enable detection rules. An analyst reviews the recommendations and chooses which detections to adjust and activate.
What affects the cost
The cost of creating a detection report can vary based on:
- Candidate detections: The number and complexity of detections to assess, rather than tenant size.
- Threat intelligence complexity: Articles covering more attacker techniques produce more behavioral patterns and candidate detections to analyze.
- Detection validation: Each candidate requires query authoring, execution, and historical testing. Noisy detections require additional tuning.
- Data source complexity: Detections spanning multiple sources, including third-party tables, require more work to understand the data schemas and write queries.
For example, a focused threat might require one straightforward detection using a familiar data source. A more complex threat might require several detections across multiple sources, with more historical testing and tuning.
For report details and the review workflow, see Detection Authoring Agent.
Threat intelligence reports
A threat intelligence report outcome is one report created by the Threat Intelligence Agent. The agent analyzes a threat intelligence article, extracts relevant information, and enriches it with Microsoft threat intelligence.
Each run produces:
- A short threat summary.
- A full analysis report identifying the source article and detailing threat actors, attacker techniques mapped to MITRE ATT&CK, indicators, vulnerabilities, malware, tools, and hunting queries.
The outcome is the analysis report, not each extracted indicator or supporting query file. The summary and full report are outputs of the same run, not separate outcomes.
The agent provides recommendations but doesn't block indicators, change your threat intelligence platform, or perform remediation. It can analyze Microsoft articles available in Microsoft Defender and third-party articles provided through HTTPS URLs.
What affects the cost
The cost of creating a threat intelligence report can vary based on:
- Article complexity: Longer, more detailed articles covering multiple actors, techniques, and indicators require more extraction and analysis.
- Intelligence enrichment: More extracted information requires more work to add context and supporting intelligence.
- Indicator comparison: More indicators require more checks against existing Microsoft Sentinel threat intelligence to identify gaps and recommend additions.
For example, a short article about one threat actor with a few clearly documented indicators might require less analysis than a detailed campaign report covering multiple actors, techniques, vulnerabilities, and indicators.
For report details, see Threat Intelligence Agent.
Posture assessments
A posture assessment outcome is one report created by the Posture Prioritization Agent. The agent identifies and ranks security posture risks based on supporting evidence and affected assets to help your team decide what to address first.
Each report includes:
- Prioritized security posture risks.
- Supporting evidence and affected assets.
- A remediation plan.
The outcome is the report, not each risk, asset, or recommended action it contains. The agent provides recommendations but doesn't remediate findings or make changes to your environment.
The agent supports the Plan for posture remediation and Protect against a threat playbooks. The first prioritizes current posture findings for remediation planning. The second analyzes posture findings in the context of a specific threat.
What affects the cost
The cost of creating a posture assessment can vary based on:
- Scope of posture data: Broader environments and more relevant posture signals require additional collection and analysis.
- Evidence and relationships: More evidence and relationships to evaluate require more work to identify, correlate, and prioritize risks.
- Scenario complexity: Threat-specific assessments require additional analysis of how the threat relates to your environment. Routine remediation planning varies with the available posture data and the complexity of prioritizing it.
Cost depends on the breadth of data analyzed and the complexity of prioritization, not necessarily the number of risks in the final report.
For example, a targeted assessment might evaluate a limited set of risks and affected assets with clear remediation paths. A broader, threat-specific assessment might require correlating more findings, assets, and relationships to determine which risks matter most.
For report details, see Posture Prioritization Agent.
Reconnaissance runs
A reconnaissance outcome is one report created by the Recon Agent. The agent provides a point-in-time assessment of potential attack paths across Azure resources and Microsoft Entra identities, showing where an attacker could go from a starting asset or identity.
Each report includes:
- Reachable assets and potential attack paths.
- Supporting evidence and an attack-path graph.
- Choke points where changes could interrupt attack paths.
The outcome is the report, not each asset discovered or attack path identified. The report's value comes from relevant findings and actionable evidence, rather than the number of paths found.
The agent performs a read-only assessment. It doesn't execute attacks or remediate findings.
What affects the cost
The cost of creating a reconnaissance report can vary based on:
- Assessment scope: More starting points, interconnected identities, and permission relationships require more data collection and analysis.
- Connectivity: Highly connected identities and permissions create branching paths that require more collection and reasoning for each starting point.
- Analysis complexity: Cost depends on the breadth and complexity of the assessment, not simply the number of resources. A smaller, highly connected environment can still require substantial analysis.
For example, an identity with multiple branching permission paths across Azure resources might require deeper analysis than one with limited connections. A smaller, more interconnected environment can therefore cost more to assess than a larger, simpler one.
For report details, see Recon Agent.
Outcomes across a playbook
A playbook can use multiple agents, with the output from one agent becoming input for another. One playbook can therefore produce more than one outcome type.
For example, the Protect against a threat playbook can produce a threat intelligence report and a detection report. The threat intelligence report assesses the threat's relevance; the detection report proposes detections. These reports remain different outcome types within the same playbook.
Completed outcomes and findings
A completed outcome doesn't necessarily identify a threat or contain findings. For example:
- A detection report can contain no proposed detections.
- A reconnaissance scan can complete without finding exposed assets or attack paths.
- A posture assessment can complete without identifying risks or exposures.
A completed analysis with no findings differs from a run that couldn't complete its analysis or produce a valid result.
These descriptions explain the outcome units shown in the calculator. They don't define charging rules for failed runs or reruns.