--- title: AI Features excerpt: "The intelligent layer of Expert Statistics: AI Chat, AI Analytics, AI Alerts, AI Dashboard, and per-host AI Learning, powered by Mistral AI." order: 3 --- ## AI Features AI Features is the intelligent layer of Expert Statistics. It is enabled per phone system, and its pages appear in the portal only when AI has been activated for your host. Each host uses its own securely stored AI key, so AI usage is isolated per customer and metered. AI Features is being progressively deployed to customers from September 2026. The AI features are powered by Mistral AI. AI Features is organised into five capabilities. ### AI Chat AI Chat lets a manager ask questions about their phone activity in plain language and get an answer grounded in their own data. The assistant has tool access to the underlying statistics: KPIs, CDRs, queue and user reports, DID reports, caller reports, trends, and agent status. Instead of returning only text, it returns structured answers that can include a written explanation, a chart, a table, or a KPI card, along with highlighted insights and clickable follow-up suggestions. The chat interface uses a two-panel layout: conversation history on one side and the active thread on the other. A conversation is bounded to a handful of exchanges so answers stay focused. Crucially, an answer is a launch point: from a chat result the user can jump straight into the matching Expert Statistics page with the period, time range, and numbers already filled in. For example, a question that used the queue-KPI tool links through to the queue KPI page for the same scope. A conversation can also be turned into a scheduled AI report. ### AI Analytics AI Analytics is a set of precomputed analytical views calculated directly from the call data (these views are pure data queries and do not consume AI credits). They accept a date range, defaulting to the last thirty days, and cover: - **Call loss**: totals and rates for answered, abandoned, abandoned-during-pre-answer, and not-answered calls, broken down by queue, by hour, by agent, and with peak hours highlighted. - **Peak hours**: hourly and weekday distributions, with highlights for the busiest and quietest hour, the busiest day, and the hour with the highest abandonment. - **Period comparison**: current versus previous period with deltas and a trend indicator (improving, stable, or declining), broken down by queue. - **Agent performance**: a ranked view of agents with answered calls, talk time, average talk time, share of total, and comparison to the team average, plus the bottom performers. - **Agent status**: per-agent time in each status, availability percentage, time in queue, and total tracked time, with team-level averages and a count of agents spending unusually high time in do-not-disturb. These analytics are reached through AI Chat and its deep links into the statistics pages, and the analytical results are fully available. ### AI Alerts AI Alerts is scheduled anomaly detection. Rather than requiring you to state exact thresholds for every condition, it watches your metrics and raises alerts when something deviates from the expected pattern. Alerts carry a severity (info, warning, or critical) and a category (call loss, volume, performance, agent, or queue), and each records the metric involved, its value, and the threshold it crossed. Alerts can be filtered by severity and category and can be dismissed. AI Alerts are configurable. Settings include whether alerts are enabled, how often to check, the notification language, and an optional notification email. Thresholds can be tuned for abandonment rate (warning and critical), not-answered rate (warning and critical), pre-answer abandonment rate, wait time (warning and critical), volume change percentage, and abandonment-rate change percentage. In the portal, active anomaly cards are shown alongside summary KPI cards for the total number of alerts and the counts by severity. ### AI Dashboard The AI Dashboard presents pre-computed insight panels so a manager can open one screen and immediately see what deserves attention. Panels come in several types, including an alert summary, a trend, a detected pattern, a recommendation, a suggestion, a KPI snapshot, and a quick action. Each panel carries a severity and a category, and unread panels are marked. Insights can be refreshed, marked as read individually or all at once, and dismissed. ### AI Learning Underneath the AI features is a per-host learning profile. The system builds a baseline for each phone system over time and uses it to inform anomaly thresholds and to personalise what the AI surfaces. This is why AI Alerts and the AI Dashboard become more tailored to your operation the longer they run. AI Alerts complement, but do not replace, the rule-based alerts described in [Core Features](/en/docs/expert-statistics/core-features). The rule-based system is explicit and deterministic; the AI system learns a baseline and flags deviations.