Intelligence & Reports

Conversations reports

Q: How do I see which channel performs best in HabariChat?

Use the Channel Performance report. It gives a per-channel breakdown of volume, share of total, unique senders, messages sent, AI resolution %, pending count and a 14-day sparkline, and you can click a channel to drill into its individual conversations.

Everything about the messages flowing through your inboxes, the channels they arrive on, how well they're answered, and the people answering them.

New to the shared terms used below (AI resolution %, CSAT, SLA, percentiles)? See How to read a report first.

Conversations reports

Message Activity

What you see. Daily incoming, sent, pending review, and failed message counts over time, plus a conversation flow funnel (Opened → AI replied → Resolved → Escalated → CSAT collected) with average messages per conversation, follow-ups sent, reopened-within-24h, and median time-to-close.

How to use it. Watch incoming-vs-sent to see whether replies keep up with demand. A widening gap, or a growing "pending" line, means work is piling up. The funnel shows how far conversations get before a human is pulled in.

How it's calculated: counts come from your inbox messages, bucketed by day. In the funnel, "Resolved" is an approximation = conversations opened − conversations escalated. Reopened-within-24h = resolved conversations that got a new customer message within 24 hours of closing; median time-to-close = median time from a conversation's first message to its resolution.

Channel Performance

What you see. A per-channel breakdown: volume, share of total, unique senders, messages sent, AI resolution %, average response time, pending count, connected inboxes, and a 14-day sparkline. Click a channel to drill into its individual conversations.

How to use it. Find your highest-volume and lowest-performing channels. High volume but low AI resolution is a good candidate for better knowledge base content or templates. The drill-down jumps straight into real conversations.

How it's calculated: grouped by channel over the selected range; AI resolution % per channel uses the same "un-edited AI replies ÷ outgoing" rule. Average response time = mean time between an incoming message and that channel's next outgoing reply.

Response Quality New

What you see. Median reply time, CSAT, auto-handled rate, and escalation rate; an SLA funnel (% replied within 1h / 4h / 24h); reply time by channel; the 20 slowest conversations; and rejection hotspots (the most common reasons reviewers reject AI drafts).

How to use it. Your service-quality scorecard. Use the SLA funnel to hold the team to a response promise, the slowest-conversations list to recover unhappy customers, and rejection hotspots to fix recurring AI mistakes at the source.

How it's calculated: reply time = reviewer's send time − customer's message time (median via percentile). Auto-handled rate = approved-without-edit ÷ all reviews. Escalation rate = escalations ÷ conversations in the period.

Customer Health New

What you see. Active customers and repeat escalations; a sentiment breakdown (positive / neutral / negative) and the negative rate; top companies and languages; and lists of low-CSAT conversations (rated 1–2) and conversations needing follow-up.

How to use it. Spot at-risk relationships before they churn. A rising negative-sentiment rate or a customer with repeat escalations is a prompt to reach out personally.

How it's calculated: sentiment, company, and language come from automatic conversation enrichment. "Needs follow-up" surfaces conversations escalated more than once.

Team Reviews

What you see. A per-reviewer table: number of reviews, approval / edit / rejection %, average review time, and a 7-day activity trend, plus your organization's top rejection reasons.

How to use it. See who is carrying the review load and how they handle AI drafts. A high edit rate suggests the AI's drafts need improvement; consistent rejection reasons tell you exactly what to fix.

How it's calculated: built from human review records; review time = when it was reviewed − when it entered the queue, averaged per reviewer.

Team Productivity New

What you see. Median review time, current queue backlog, and approval rate; a time-to-review histogram (under 5 min / 5–15 / 15–60 / over an hour); a day × hour heatmap of review activity; and pending by inbox, including the age of the oldest waiting item.

How to use it. Manage the review queue. A large backlog or an old "oldest pending" item means customers are waiting. The heatmap shows when your team actually works, so you can align staffing to demand.

How it's calculated: review time = reviewed − queued; backlog = messages currently pending review.

Escalations

What you see. A paginated list of escalations (reason, status, age) and a summary: counts by status (open / in progress / resolved / dismissed), average resolution time, number unresolved beyond 24h, dismissed rate, and top reasons.

How to use it. Your human-handoff queue. Work the unresolved-24h count down to zero, and use the top reasons to decide what the AI or knowledge base should learn to handle next.

How it's calculated: age = (resolved time or now) − created time. Unresolved-24h = not resolved/dismissed and created more than 24 hours ago.

Customer Satisfaction

What you see. CSAT average and number of ratings, the 1–5 distribution, and the trend; the AI self-assessment average and distribution alongside it; CSAT by channel, AI rating by conversation type, and a CSAT-vs-AI correlation view.

How to use it. Track whether customers are happy and whether it's improving. CSAT by channel shows where experience is weakest; the correlation view, comparing customer CSAT against the AI's self-rating, shows whether the AI's confidence matches reality.

How it's calculated: CSAT = average of customer 1–5 ratings; AI self-rating = average of the AI's own ratings; the correlation view plots each conversation's CSAT against its AI self-rating.

Common questions

How do I see which channel performs best in HabariChat?

Use the Channel Performance report. It gives a per-channel breakdown of volume, share of total, unique senders, messages sent, AI resolution %, pending count and a 14-day sparkline, and you can click a channel to drill into its individual conversations.

What does "unresolved beyond 24h" mean in the Escalations report?

It counts escalations that are not yet resolved or dismissed and were created more than 24 hours ago, meaning they have breached the 24-hour SLA. Work this number down to zero to keep customers from waiting.

What is the difference between CSAT and the AI self-rating?

CSAT is your customers' satisfaction on a 1 to 5 scale from ratings they submit. The AI self-rating (also called AI confidence) is the AI grading its own answer, also 1 to 5. CSAT is customer opinion; the self-rating is an early-warning signal from the AI.