/dashboard gives you a real-time view of everything your AI support operation is doing — how much it costs, how fast it resolves tickets, where it struggles, and where your agents spend their time. You don’t need to dig through logs or export data; the metrics update continuously as tickets flow through the system.
What the dashboard shows
The dashboard is organized into sections, each focused on a different aspect of your AI operation. You can use them together to spot patterns: for example, a spike in cost per ticket alongside a drop in cache hit rate usually means a new category of question is coming in that the response cache hasn’t seen before.Cost per ticket
The average AI cost for each ticket resolved, with a daily trend line. Use this as your primary efficiency metric — it factors in token usage across all operations (generate, translate, rewrite, refine).
Mode breakdown
A split between Agent Mode (fully autonomous resolutions) and Copilot Mode (AI-assisted, agent-reviewed). Shows ticket counts, average response time, and cost for each mode side by side.
Token usage
Total token consumption broken down by operation type: generate, translate, rewrite, and refine. Each operation shows its cost share so you can see what’s driving spend.
Activity heatmap
A 7-day × 24-hour grid that shows your busiest periods. Darker cells mean more ticket activity. Use this to plan staffing and to understand when Agent Mode handles the most volume autonomously.
CSAT tracking
Satisfaction scores, good/bad rating counts, average resolution times, and reopen rates over time. Drops here are the earliest signal that something needs attention.
Knowledge gaps
A live list of topics where the AI had low confidence or couldn’t match a procedure. Each gap links to example tickets so you can see exactly what customers are asking. See Detect and close knowledge gaps for the full workflow.
Resolution stats
P50, P95, and P99 response time percentiles, with counts broken down by ticket type and status. Percentiles give you a more honest picture of performance than averages alone.
Agent usage
Per-agent breakdown: ticket distribution, token consumption, cost attribution, and last active timestamp. Useful for understanding how the AI load is distributed across your team.
Response cache
Cache hit rate and the total cost saved by serving cached responses instead of running the full AI pipeline. Cached replies have near-zero latency and cost nothing in AI tokens.
Reading key metrics
Cost per ticket
Cost per ticket is the clearest measure of your AI operation’s efficiency. A rising trend without a corresponding increase in ticket complexity usually means one of three things: the response cache hit rate has dropped, a new ticket category is bypassing your procedures, or token-heavy operations like translation are running more frequently.Token costs are based on your own AI provider account (OpenAI, Anthropic, or Gemini). raum doesn’t mark up what you pay for AI — the cost figures in the dashboard reflect your direct provider costs.
Mode breakdown
The mode breakdown tells you how much of your volume the AI is handling autonomously versus with agent involvement. A healthy operation typically sees Agent Mode handling the high-confidence, repeatable ticket types while Copilot Mode covers the more complex or sensitive ones. If Agent Mode response times are creeping up, check your procedure coverage — the AI may be falling back to slower reasoning paths because it doesn’t have a matching procedure.Resolution stats and percentiles
The P50 is your median resolution time — half of tickets resolve faster than this. The P95 and P99 show your tail latency: the slowest 5% and 1% of tickets. If P99 is dramatically higher than P50, you likely have a category of complex tickets that are pulling down the overall experience.Response cache
The response cache stores AI-generated replies and serves them instantly when a similar question comes in again, without re-running the AI pipeline. A high cache hit rate is one of the biggest cost reducers available to you.Tips for optimizing based on data
Cost is rising — where do I start?
Cost is rising — where do I start?
Open the token usage section and look at the operation breakdown. If “generate” is the dominant cost, check whether Agent Mode is running the full pipeline on tickets that could be handled by the response cache. If “translate” is high, you may have a language segment that needs dedicated procedures in that language to reduce the back-and-forth.
CSAT dropped — what caused it?
CSAT dropped — what caused it?
Cross-reference the CSAT trend with the activity heatmap and knowledge gaps list. A CSAT drop that coincides with a busier period may point to slower response times. If the drop is steady rather than event-driven, new knowledge gaps are a likely cause — the AI is handling topics it doesn’t have good procedures for.
Agent Mode resolution time increased
Agent Mode resolution time increased
Check whether new ticket types have appeared in knowledge gaps. If the AI can’t match a procedure, it runs a slower reasoning path. Adding a procedure for the new topic typically brings resolution times back down.
One agent shows high token consumption
One agent shows high token consumption
The agent usage section shows per-user token costs. High token consumption for one agent in Copilot Mode may mean they’re triggering regeneration frequently. Check whether the AI’s first draft is consistently off for that agent’s ticket types — this often points to a missing procedure or a guideline that needs updating.