Bot Analytics
The Analytics tab gives you a real-time view of how your bot is performing — how many conversations it is handling, which platforms are most active, how fast it responds, and where errors are occurring. Use this data to understand usage patterns and guide your bot improvement efforts.
What analytics shows
Analytics is available on the bot cockpit for every bot you own. Navigate to your bot in Studio and click the Analytics tab. The page updates automatically and shows data as recent as the last polling interval.
The page is organized into four sections:
- KPI cards — four headline metrics at a glance
- Message activity chart — daily message and conversation volume over time
- Platform breakdown — share of traffic by connected channel
- Performance metrics — detailed latency, token, cost, and error data
KPI cards
The four KPI cards at the top give you an immediate health read on your bot.
| Card | What it measures |
|---|---|
| Total Conversations | The number of unique conversation threads your bot has participated in. A conversation is a distinct session with one user. |
| Total Messages | The total number of individual messages processed — both incoming user messages and bot responses. |
| Avg Response Time | The mean time the bot takes from receiving a message to sending a reply. Shown in milliseconds (ms) for fast responses and seconds (s) for slower ones. |
| Uptime | The percentage of time the bot was available and responding. Values above 99% show in green; 95–99% show in amber; below 95% show in red. The error rate is shown as a subtext below the uptime value. |
Message activity chart
The message activity chart shows your bot's daily volume over either the last 7 days or 30 days. Use the 7d / 30d toggle in the top-right of the chart panel to switch between ranges.
The chart plots three series:
- Messages (solid teal line) — total messages per day, including both inbound and outbound
- Conversations (blue line) — unique conversation threads started that day
- Errors (dashed red line) — message processing errors; a flat zero line means everything is healthy
Look for patterns in the Messages and Conversations lines. If conversations are flat but messages per conversation are rising, your users are sending longer exchanges — which may signal ambiguity in responses that the knowledge base could resolve. If errors spike on a specific day, correlate that with any configuration changes you made.
Platform breakdown
The Platform Breakdown panel shows you how traffic is split across your connected channels — Telegram, Discord, Widget, API, Twitter/X, and any others you have configured.
Each platform row shows:
- Platform name with a color-coded dot
- Conversation count — unique conversation threads from this platform
- Message count — total messages from this platform
- Percentage — this platform's share of total traffic, shown as both a number and a progress bar
If one platform dominates, you may want to prioritize knowledge base content and response style for that audience. If a platform you expected to be active shows low numbers, check that the bot is connected and that the Primary Channel is configured correctly in the bot settings.
Performance metrics
The Performance Metrics panel gives you the detail behind the KPI cards.
| Metric | Description |
|---|---|
| Avg Response Time | Mean latency across all messages in the selected period. A healthy bot typically responds in under 2 seconds. |
| p95 Response Time | The 95th percentile latency — the time under which 95% of all responses complete. High p95 with low average signals occasional slow outliers, often caused by long knowledge base retrievals or complex tool calls. |
| Avg Tokens / Message | Average number of LLM tokens consumed per message exchange. Higher values mean longer prompts or more context being injected — which increases cost. |
| Total Cost | Total estimated LLM cost for the period in USD, calculated from token usage. This is an estimate based on the configured model pricing and may not exactly match your invoice. |
| Error Rate | Percentage of messages that resulted in an error. Below 1% is healthy (shown in green); 1–5% is a warning (amber); above 5% requires investigation (red). |
The Last updated timestamp at the bottom of the panel tells you when the data was last refreshed.
Time range and data freshness
The message activity chart supports 7-day and 30-day views. The KPI cards and performance metrics always show totals across all available data — they are not filtered by the chart's time range selector.
Analytics data is refreshed periodically in the background. If you need the most current numbers, refresh the page.
Tips for interpreting metrics
High p95 but low average: A small number of slow requests are pulling the 95th percentile up. Check whether certain types of questions trigger heavy knowledge base lookups or multiple tool calls. Simplifying knowledge base structure or reducing system prompt length can help.
Rising token count over time: This often means the conversation history being injected into each prompt is growing. Review your bot's context window settings in the AI Configuration tab — consider reducing the number of historical turns retained.
Error rate above 1%: Open the trading dashboard or check the audit logs in Hub to identify the failure type. Common causes include malformed tool responses, rate limit hits on LLM providers, or expired platform tokens (e.g., a Telegram bot token that was regenerated).
Platform traffic lower than expected: Verify the bot's primary channel assignment in Bot Settings. Ensure the platform integration (Telegram bot token, Discord webhook, etc.) is active and that the bot has not been paused or put in maintenance mode.
Uptime below 99%: This metric reflects backend availability, not network issues on the user's end. Persistent downtime below 95% should be reported through the troubleshooting guide.
Related pages
- AI Configuration — adjust model, context window, and system prompt to reduce token usage
- Knowledge Base — improve response quality by strengthening retrieval content
- Bot Settings — check primary channel and platform integration status
- Audit Logs — cross-reference error spikes with platform-level events
- Troubleshooting — steps to diagnose persistent uptime or error rate issues