Intelligence & Analytics

The Intelligence tab is your chatbot's report card: what visitors asked, where Orac answered confidently, and — the part that actually matters — where it didn't. Instead of reading transcripts hoping to spot problems, you get them detected, grouped, and ranked.

Every query is scored before Orac answers, based on how well your knowledge base actually covers it — retrieval strength, agreement between sources, and topic match. High-confidence answers cite their sources; low-confidence queries are logged as potential gaps. In strict mode, queries below the confidence gate get your fallback response instead of a guess — zero-hallucination by construction, not by prompt engineering.

Confidence scoring also drives routing. In Settings → Auto-Escalation you choose a threshold: below it, the visitor still receives Orac's best-effort reply — no dead air — while an escalation lands in your team's inbox and Orac goes quiet on the follow-up so nobody talks over your reply. The AI handles what it's sure about; people handle the rest.

Low-confidence queries are clustered into named gaps ("refund policy for annual plans", "API rate limits") with counts, so you fix the most-asked gap first. Orac drafts a suggested answer from the nearest related content where it can; you edit, approve, and it becomes a verified knowledge entry that answers that question directly from then on. Replies you write in conversations can be saved the same way — every human answer makes the bot smarter.

Repeated rephrasing, negative feedback, and frustration signals in visitor messages are detected and can trigger escalation to your team — before the visitor gives up, not after. Each trigger is visible on the conversation, so you can see why a handoff happened.

A weekly email summarizes conversations, resolution quality, top questions, and new gaps — set the frequency (or turn it off) under Settings → Notifications. Ten seconds of reading that tells you exactly what to add to your docs this week.