Recommendations and AI insights
Recommendations and AI insights direct your attention to the AI governance work that matters most, so you can resolve high-impact issues without searching for them.
AI Control Tower continuously evaluates your managed AI assets against governance, security, value, risk, and compliance criteria. When AI Control Tower detects an issue or an opportunity, it surfaces that finding as either a recommendation or an AI insight, depending on whether the user is expected to take action or to read contextual information.
Key benefits
- Focus on the highest-priority governance work without searching the product for issues that need your attention.
- Resolve routine issues hands-off by letting an unsupervised agent run the remediation workflow for you.
- Work through complex issues in a guided conversation with a supervised agent that proposes actions you can accept, refine, or reject.
- See the recommendations and insights most relevant to your role, with the severity, confidence, and age of each finding factored into the order.
Recommendations
A recommendation is an action item that AI Control Tower has identified as worth your attention. Each recommendation names the condition that was detected, the outcome it supports, and how you can resolve it.
Recommendations appear throughout AI Control Tower.
- Your top recommendations in Home
- Your top recommendations on the Home page provides recommendations across your portfolio. Select See all Recommendations in Activity Center to open the full list.
- Recommendations view in Activity Center
- The full list of recommendations across your portfolio, with filters for category, priority, confidence, and outcome. Use this view when you want to work through recommendations at scale or focus on a specific category such as Security or Risk and compliance.
- Needs attention section on an asset record
- Recommendations scoped to a single asset, visible on the Overview tab of the asset record. The recommendations shown here are the subset that apply to that asset. For more information about the Overview tab, see Reviewing AI asset status.
AI Agent Advisor recommendations
When you install AI Agent Advisor, its highest-value automation opportunities also surface as recommendations in AI Control Tower, so you can discover agent-building opportunities without checking AI Admin Center separately. AI Control Tower shows your top automation opportunities, ranked by projected return on investment, up to a limit you can configure with the
sn_aict_genai.aaa_automation_opportunities_limit property (10 by default).
Each recommendation card shows the opportunity's name and a short description. Selecting the recommendation's action opens the Automation details page for that specific opportunity in AI Admin Center, where you can act on it, such as by building an agent.
This integration requires no setup. The connection activates automatically when both AI Agent Advisor and AI Admin Center are installed on your instance. AI Agent Advisor must also complete at least one mining run before its opportunities appear. If AI Admin Center isn't installed, AI Agent Advisor doesn't send any recommendations to AI Control Tower.
Recommendations stay current on their own, refreshing daily. If an opportunity is still among the top group the next day, AI Control Tower updates the existing recommendation instead of adding a duplicate. If an opportunity drops out of the top group, or no longer exists in AI Admin Center by the time you select it, AI Control Tower removes the recommendation.
For details on AI Agent Advisor, see AI Agent Advisor.
AI insights
An AI insight is a contextual fact that AI Control Tower adds to a dashboard to give you a richer reading of the data. Unlike a recommendation, an AI insight does not ask you to take action. It appears as a short line of text near the data it explains, such as "Total productivity gains in last 30 days" or "Three AI systems delivered 82% of total value." AI insights appear inside widgets on the Home page, the Value page, the Govern page, and other AI Control Tower dashboards.
When an AI insight is associated with a supervised agent, you can select Manage with AI to start a conversation that expands on the insight with deeper analysis.
Resolving recommendations
Each recommendation includes a primary resolution that AI Control Tower has selected as the most efficient path to address the underlying issue. Alternatively, you can mark the recommendation as complete if you address the issue another way. For the full procedure, see Resolving AI recommendations in Activity Center.
The following options are available in the Actions menu:
| Action | Description |
|---|---|
| View details | Provides the reasoning behind the recommendation, the asset or scope it applies to, and the history of earlier runs. |
| Complete | Marks the recommendation as resolved without running the primary resolution action. Use this option when you have addressed the underlying issue in another way. |
| Delete | Removes the recommendation. A deleted recommendation might reappear if AI Control Tower detects the same condition again. |
Depending on how the recommendation was produced, you can use one of the following resolution options:
| Primary resolution | Description |
|---|---|
| Automate with Otto | An unsupervised agent runs the resolution workflow on your behalf. When you confirm, the agent executes the steps and updates the recommendation status to In progress, then to Complete once the work is done. This resolution pattern is used for recommendations where the steps are routine and don't require your input, such as populating missing asset descriptions. |
| Manage with Otto | A supervised agent opens a conversation in the AI Control Tower chat interface. The agent proposes actions, and you accept, refine, or reject each one. This resolution pattern is used for recommendations that benefit from your judgment, such as reviewing the evidence behind a security threat before resolving it. |
| Open | AI Control Tower redirects you to the record, page, or workflow where you can resolve the issue directly. This resolution pattern is used for recommendations that require changes in another part of the product, such as creating tasks in an onboarding playbook. |
When you select View details from the action menu, the side panel offers two views:
- Focused
- Shows the current status, the steps the agent will take or has taken, and an estimated time to complete. Use this view while an agent is running so you can monitor progress without leaving the page.
- Details
- Shows the full reasoning behind the recommendation, including the data sources that contributed to it and the history of earlier runs.
Prioritizing recommendations
AI Control Tower does not show every recommendation it detects. Recommendations are prioritized according to the following factors:
- Your role
- An AI steward and a product owner see different recommendations in different order. Each role is responsible for different outcomes. For example, a product owner sees recommendations that help drive productivity for their own AI assets first, while an AI steward sees recommendations that improve inventory signal and AI governance across the portfolio first.
- Severity
- Each recommendation carries a priority level of 1-Critical, 2-High, 3-Medium, or 4-Low. Higher severity items rank above lower severity items when other factors are equal.
- Confidence
- Each recommendation carries a confidence level of Low, Medium, or High. Confidence reflects how certain AI Control Tower is that the detected condition is accurate. A high-confidence recommendation ranks above a low-confidence recommendation at the same severity.
- Age
- Recommendations become less prominent over time if no one acts on them. This prevents a long-standing issue from permanently occupying the top of the list and gives newer recommendations a chance to surface.
In Activity Center, use the Refine list panel to filter recommendations by priority, confidence, category, or outcome, and override the default ranking to focus on a specific area.
Recommendation outcomes
Every recommendation is tagged with an outcome that identifies the business goal the recommendation supports. The outcome appears on the recommendation card.
| Outcome | Description |
|---|---|
| Drive AI productivity | Helps you increase the value that your AI assets deliver, such as identifying high-performing assets to scale or underused assets to retire. |
| Improve AI governance | Helps you strengthen oversight of your AI portfolio, such as completing lifecycle tasks or enforcing policy across managed assets. |
| Improve inventory signal | Helps you make your inventory more complete and accurate, such as populating missing asset descriptions or reviewing newly discovered assets. |
| Monitor security | Helps you respond to AI security threats and access issues, such as reviewing high-severity security events or addressing agent output deviation. |
| Monitor risk and compliance | Helps you maintain regulatory and policy compliance, such as reviewing overdue risk assessments or addressing non-compliant controls. |
Recommendation status
Each recommendation moves through a lifecycle as you or an agent works on it. The status appears in the side panel that appears when you select View details.
| Status | Description |
|---|---|
| Not started | The recommendation has been generated but no one has acted on it. |
| In progress | An agent is running or the assignee is working on the recommendation. |
| Complete | The recommendation has been resolved. Completed recommendations are removed from the list. |
| Failed | An agent could not complete the recommendation, or could only complete it for some of the assets it targeted. You can retry the action or resolve the recommendation manually. |
| Cancelled | The recommendation was stopped before it completed, either by the user or by the system. |
You can manually mark a recommendation as Complete or delete it from the action menu on the recommendation card, regardless of its current status.