Exporting AI Assistant Analytics
Dashboards are great for quick checks, but teams need raw data for modeling, attribution, and compliance. Offer exports that are easy to consume and safe to share.
Data to include
- Per-embed metrics: impressions, opens, chats, messages, containment.
- Fallback reasons and retrieval scores.
- Feedback events (thumbs, reason, free text).
- Billing usage: message counts, crawl minutes, LLM tokens.
- Metadata: tenant_id, embed_id, page_path, language, timestamp.
Format guidance
| Format | Use case |
|---|---|
| CSV | Quick Excel/Sheets reviews, manual reporting |
| NDJSON | Append-friendly ingestion into warehouses or data lakes |
| API | Low-latency integrations or scheduled pulls |
Delivery options
- Manual download: Admin UI button with date range filters and timezone selection.
- Scheduled exports: Allow admins to send periodic exports to email or cloud storage (GCS/S3).
- Webhook delivery: Post analytics batches to customer endpoints for real-time dashboards.
Security considerations
- Ensure exports respect tenant scoping and RBAC; only authorized users can download.
- Apply encryption at rest; if emailing, use signed URLs with short expiry.
- Log every export event (who, when, what filters) for audit purposes.
CrawlBot implementation
CrawlBot’s analytics service exposes CSV/NDJSON exports and an API, all scoped per tenant and embed. Start with these basics so customers can analyze assistant performance alongside their broader KPIs.***