AI Assistant
Audience: All users
Page Type: How-To Guide
Duration: 5 minutes
What is the AI Assistant?
The COS AI Assistant is a conversational AI powered by leading language models (OpenAI, Anthropic, etc.). It is embedded directly in the platform and has access to your organizational data — OKRs, strategies, KPIs, analyses — through a set of secure tool calls. It answers questions, generates content, and performs analysis in the context of your actual data.
AI is available to every entitlement set. Assistant menu visibility and available tools depend on your role and access permissions.
Opening the AI Assistant
- Click the robot icon (🤖) in the top navigation bar (top right area)
- A chat panel slides open from the right side
- Type your question or request and press Enter or click Send
What the AI Assistant Can Do
The assistant uses only tools offered for your role, scope and enabled modules.
Strategy & Analysis
- Summarize your SWOT, PESTLE, Porter's Five Forces, or BCG analysis
- Suggest focus areas based on your analysis results
- Compare current state assessments across cycles
OKR Management
- List your current OKRs and key results
- Summarize OKR progress and health scores
- Suggest key results for a given objective
- Identify at-risk OKRs that are falling behind
KPI & Performance
- Show progress of specific KPIs
- Compare KPI performance across units or time periods
- Identify performance gaps
KPI register tools. The assistant works with KPIs through the KPI register; the former per-report KPI tools were removed.
| What you can do | Note |
|---|---|
| Search KPIs and view details, measurement history, a report's KPI list | Only KPIs in your scope and reports you can see |
| View the Status Report KPI → Focus Area KPI → key result chain and its gaps | Goal Alignment Matrix data |
| Create KPIs, update definitions, retire KPIs | Plan limit and KPI Tracking authority rules apply unchanged |
| Add measurements | Only with measurement authority (responsible, owner/parent unit leader, administrator roles); not on retired KPIs |
| Add KPIs to a report (from library, from KPI register, from another report, from previous cycle), remove them, set report weights | Rejected for reports in a closed cycle |
| Set a KPI's success criteria | Measurement authority + KPI edit permission; rejected for retired KPIs |
| Link a key result to a Focus Area KPI | Same planning period rule applies |
- Every write, including destructive ones (retire, remove from report), asks for your approval on an approval card before it is applied.
- Changes made by the assistant appear in the KPI's History tab with the "AI assistant" channel.
- Setting success criteria: the assistant can propose a KPI's success criteria (basis, 5 levels, thresholds; the level is always calculated from the total) and save them with your approval. The authority rule is the same as in KPI Tracking (measurement authority + KPI edit permission). The assistant uses the criteria's current version; if someone changed it meanwhile, the save is rejected. Levels without a color get the standard palette.
- For the KPI list of a report in a closed cycle the assistant reads the closing snapshot.
- The assistant's KPI tools also check your access to the KPI Tracking menu and the relevant report menu.
Dashboard & Reporting
- Summarize the organization's strategic health
- Generate a narrative summary of your dashboard data
- Highlight units that are ahead or behind
Assessments (DMA & Excellence)
- Summarize open or completed assessment results
- List pending tasks for a given assessment
Findings
- List open findings and their current status
- Summarize findings by category or module
Roadmap
- Summarize the strategic roadmap
- Identify milestones that are at risk
Example Questions
"What are our top OKRs this quarter?"
"Which focus areas are performing below target?"
"Summarize our SWOT analysis for 2025."
"What key results would you suggest for an Operational Excellence objective?"
"Show me which units are behind on their OKRs."
"Are there any open findings in the Findings module?"
How It Works
- You send a message
- The AI selects the appropriate tools to call (e.g.,
list_my_okrs,get_swot_summary) - Tools are executed via secure HTTP API calls — the AI never accesses the database directly
- Results are interpreted and summarized in natural language
Tools operate in the selected company and user access scope. Confidential-record and role restrictions also apply through the assistant.
Limitations
| Limitation | Detail |
|---|---|
| Read-mostly | The AI can read and summarize data; some write operations (e.g., creating records) are also supported but may require confirmation |
| Context window | Very large datasets may be summarized or truncated |
| Language | The assistant responds in the language you write in |
| Access | Menu permission, data scope and provider availability apply |
Tips
- Be specific: "Summarize OKRs for Q1 2025" works better than "Tell me about goals"
- Mention module names when asking about specific data (OKR, SWOT, KPI, etc.)
- Ask follow-ups: The assistant remembers context within the same conversation
Conversations, models and history
- Send: Enter sends a message; Shift+Enter adds a line. Stop cancels the current response stream.
- New chat: Keeps the previous conversation in history and opens a new one.
- History: Open an earlier conversation to continue. A conversation chosen from history retains its own strategic-cycle context; distinguish it from a different top-bar cycle.
- Model selection: If your setup and access offer multiple models, use the picker beside the input. Options come from company, group and permitted platform providers.
- Assistant/agent selection: When multiple options are offered, choose one from the additional-actions menu. A conversation is pinned to one agent; changing it starts a new conversation context.
- Suggestions and context: The suggestions control and page/cycle/unit context help prepare requests. State which cycle and record you mean.
- Minimizing or closing the panel keeps the conversation in history. Changing company stops the previous company's stream and loads the new scope.
Usage pool and providers
AI is available to every entitlement set; it has no separate paid AI module. Menu access and permissions for the requested data still apply.
On SaaS, messages and web searches using the COS model count toward the group's monthly AI pool: active users × monthly share per user. It resets by calendar month with no rollover. A company's or group's own provider does not consume this pool; that provider's capacity/cost conditions still apply.
Web research is offered only when configuration and access allow it. Distinguish external-source results from company records. Chat retention is a platform setting; older conversations may be cleaned up after that period.
On-prem installations do not apply AI message-pool or AI rate quotas. An authorized administrator configures company/group providers. See AI providers and models.
Approvals and errors
When a write request becomes an approval card, review the proposed records and fields, then Approve or Reject. New messages are disabled while approval is pending. Ask for a new proposal if it expires or becomes stale after another change; a conversational response alone does not prove the record changed.
Members use the assistant's read tools. Being able to update progress on your own work in the UI does not grant every AI write tool. Permissions, cycle locks, retired KPIs and version-conflict checks still apply to tools.
If no provider is available, ask an administrator to check configuration. Follow the displayed message for pool/rate limits. If a response stops, check the tool result or current record before requesting the same write again.