How capture works
Alpheris does not capture classic browser tracking events and sets no marketing cookies and no tracking pixels on the visitor. Instead, the dashboard analyzes exactly what arises during normal operation anyway: the conversations themselves and the AI usage behind them. Both are stored server-side in your tenant and aggregated depending on the time period.
Specifically, all metrics are fed from two sources:
- Messages – every message in a conversation with its sender role (
visitor,ai,agent,system) and timestamp. Chats, distribution, and history are calculated from this. - AI usage – for each AI response, the model, provider, token consumption, and a blocked flag (e.g. a limit being reached) are logged. Token, request, and limit metrics are derived from this.
window.dataLayer/fbq push from the widget. If you want to report clicks or page views to GA4 or the Meta Pixel, you continue to use your own analytics setup on the website for that – independently of Alpheris.What gets captured
Alpheris builds the following metrics automatically – with no extra configuration, as soon as the widget or an inbox channel is active.
| Metric | Meaning | Source |
|---|---|---|
chats | Conversations with at least one visitor message | Messages |
visitor_messages | Messages that came from the visitor | Messages |
ai_requests | Answered AI requests (not blocked) | AI usage |
tokens | Tokens consumed by the AI responses | AI usage |
blocked | Blocked requests (e.g. limit reached) | AI usage |
From these raw values, the dashboard additionally derives further analyses:
- Sender distribution – share of visitor, AI, and staff across all messages.
- Channel distribution – by webchat, email, WhatsApp, Instagram, etc. (see Inbox).
- Time series – a series per hour, day, or month – depending on the selected time period.
- Top questions – the most frequent visitor questions, grouped by topic.
- Drop-off points – where conversations end without completion (funnel stages).
- Knowledge gaps – topics with low AI confidence where answers were missing.
Dashboard analysis
You'll find the full analysis in the dashboard under Analytics (/dashboard/analytics). At the top you choose two things:
The most important cards at a glance:
Close knowledge gaps directly
When the AI repeatedly detects questions it couldn't answer confidently, these land in the dashboard as open knowledge gaps – sorted by frequency. There you can add an answer directly, which is adopted into the knowledge base as an FAQ. The next time, the AI answers grounded.
AI topic summary
For a linguistically condensed topic summary of the most frequent questions, the dashboard uses the AI. This feature is limited to Business and Enterprise; on smaller plans the card stays locked. The summary is cached (Business 24 h, Enterprise 6 h) so that it doesn't create additional load.
Inbox analysis
If you select the Inbox area at the top, the metrics are based on the email and messaging threads instead of the webchat. Among the things analyzed are:
- New messages and processed threads in the period.
- Average response time (first customer message to first reply).
- Status distribution (open / in progress / closed).
- Workload per staff member and history by hour/weekday.
- AI learning progress – how many completed email tickets the AI has already learned from and how many are still missing until the next level (see Inbox & AI Replies).
Limits & Usage
Token and chat usage are measured against your plan limits – shown as a daily and monthly value with the remaining allowance. When a limit is reached, Alpheris logs the affected request as blocked with a reason instead of silently discarding it. This way you can see in the log exactly when a limit took effect.
# Example entries from the usage log
2026-06-18 09:14:02 UTC usage 812 tokens (mistral:mistral-medium-latest)
2026-06-18 09:15:41 UTC usage 640 tokens (mistral:mistral-large-latest)
2026-06-18 09:22:10 UTC limit monthly token limit reachedPrivacy-compliant capture
The capture is deliberately designed to minimize data:
- No tracking pixels, no marketing cookies. No third-party pixels (GA4, Meta) are set by Alpheris and no visitors are tracked across pages.
- No IP storage in the analytics data. The analyzed message and AI usage data contain the conversation ID, sender role, text, timestamp, and token consumption – but no visitor IP.
- Tenant isolation. Every analysis is strictly bound to your shop; other tenants never see your data.
- Data location Germany. Tracking events are stored and evaluated exclusively in Germany (Hetzner Online GmbH, data centers in Nuremberg/Falkenstein); they are not sent to AI providers.
Consent & Privacy Notice
Since Alpheris does not use any consent-requiring tracking technologies for the conversation itself, the widget's consent prompt is disabled by default. You can enable it per shop if you want to obtain explicit consent before the first chat – then the widget requires a confirmation before a conversation starts.
Additionally, you can set a privacy URL that is linked in the widget. You'll find both in the dashboard under Install (widget settings).
You'll find the data processing agreement (DPA/GDPR Art. 28) at alpheris.com/avv. Details on data location and subprocessors can be found under Security & Privacy.