Reports

Admin Panel → Reports shows how the widget is doing: conversation volume, resolution rate, satisfaction, the mix of questions and your agents' workload.

Headline numbers

Measure	Window
Conversations	the date range you picked
Open chats	right now
Human takeover	the date range you picked
Closed conversations	the date range you picked; closures by an agent and automatic closures are shown separately, and automatic closures don't count in the resolution rate

Performance

Measure	Target	Window
Resolution rate	above 80%	the date range you picked
Escalation rate	below 15%	the date range you picked
Average response time	under 3s	the date range you picked

The targets are references, not commitments; your own good/bad threshold may differ.

Customer satisfaction

The 1–5 score collected at the end of a widget chat and through the satisfaction survey (email, WhatsApp, Instagram, Telegram): average, satisfaction rate (the share of 4s and 5s) and total ratings. With no ratings the rate is not computed — printing "0% satisfaction" would paint a false picture when nobody has voted.

Two boxes below split the score: Answered by AI and Answered by the team, each with its average, satisfaction rate and number of ratings. If an agent took over the conversation, the score belongs to the team; if nobody took over, to the AI. A side with no ratings says “No ratings in this range.” To hear about low ratings (1–2): Alert me about low ratings.

Net Promoter Score (NPS)

The card below satisfaction adds up the 0–10 scores given to satisfaction survey (NPS) emails over the last 90 days: the NPS, the number of responses and the split into promoters (9–10), passives (7–8) and detractors (0–6). NPS is the percentage of promoters minus the percentage of detractors (from −100 to +100). The card doesn't follow the date range you pick. With no responses yet, it suggests sending an email whose type is Satisfaction survey (NPS) from Campaigns › Email.

Question categories

Incoming questions are classified automatically: Order tracking · Product query · Return / cancel · Customer info · General support · Escalation · Other. The distribution follows your selected range and shows what customers actually ask — a good guide to where your knowledge base needs weight.

Human handoff

How many widget conversations went to a person, how many the AI resolved alone, and how fast the team picks up and resolves handoffs. The table below shows RAG and e-commerce on separate rows; next to it are the reasons conversations went to a person and the rules that hand off most.

Measure	What is counted
Went to a person	The share of conversations handed to the team or taken over by an agent.
Resolved by the AI alone	Conversations an agent marked Resolved without a person stepping in, or that got a thumbs-up or a 4-5 rating; auto-closed ones don't count.
First human reply	Time from the handoff to the agent's first message: the median and “9 in 10 answered within …”. Handoffs that arrive after hours are shown separately.
Handoff to resolution	Time from the handoff (or the takeover, if there was no handoff) until an agent marks the conversation Resolved: the median and “9 in 10 were resolved within …”. Conversations that closed on their own don't count.
Reply time target	If you set a target, a line such as “Target 15 minutes: 29/33 handoffs answered on target (87.9%)” appears under the first human reply. A handoff is on target if an agent wrote before the target time ran out; it is missed if they wrote later, or the time ran out with no reply; an unanswered handoff whose time hasn't run out isn't counted yet. With working hours on, the clock only runs during working hours. Without a target, the line says so and links to Set a target.

Agents

Your team's workload and speed, person by person. All channels (widget, email, WhatsApp, Instagram, Telegram) are counted by default; use the Channel filter to narrow it down. Counts follow the work done in the range, even in conversations that started earlier.

Column	What is counted
Assigned	Conversations assigned to this person in the range that are still with them.
Replied · Messages	Conversations and messages this person wrote in the range.
First reply	Time from an assignment in the range to this person's first message: the median, and below it “9 in 10” (how fast 9 in 10 assignments were answered).
Resolved	Conversations this person marked Resolved in the range; automatically closed ones count for no one.
Resolution time	The median time from assignment until this person marks the conversation Resolved. It's in the CSV exports too.
CSAT	Average 1–5 rating given in the range to conversations assigned to this person, and the number of ratings.
Open	Conversations with this person that aren't resolved yet, regardless of the range.

• Filters: channel (All channels by default), widget type (RAG · E-commerce), team (its members are listed even with no work) and agent.
• Download as CSV: Download summary gives the table itself, Download conversations one row per conversation (assignee, assigned by, first reply, resolved by, CSAT). The files open directly in Excel. Customer names and emails appear only for roles that can see Conversations.
• If someone's role changed, their row stays with a “no longer an agent” note.
• A selection with no work doesn't leave the list empty: if the person or team you picked had no work in that range, their row shows zeros and the reason (“The selected person has no conversations assigned, replied to or resolved in this range.”).
• Older records: agent messages have recorded who wrote them since 13 September 2026, and conversations their assignee since 12 September 2026. Older messages without a recorded author count for the conversation's assignee; those with no assignee either sit on the “Unknown” row and can't be linked to a person afterwards. With a person or team filter selected these messages are left out, and the screen says how many.

Channels

Channels is the first section of the Reports screen. The Channels table splits the selected range into widget, email, WhatsApp, Instagram and Telegram: the number of conversations and their share, how often a conversation went to a person, how often the AI resolved alone, the first human reply (in-hours median; hover the cell for the out-of-hours value), the AI's reply time, and customer messages left unanswered in 24 h. A channel with no conversations and no unanswered messages in the range isn't listed. Definitions match the handoff report; the table can be downloaded as CSV.

• Unanswered in 24 h (WhatsApp and Instagram): customer messages whose window closed in the range with no reply reaching the customer and no new customer message within 24 hours. Reactions, history imported when the number was connected, and conversations marked resolved within the window don't count. You can still write to these customers with an approved template on WhatsApp; on Instagram nothing can be sent until the customer writes again (Messaging › The 24-hour window).
• AI alone is measured only on channels where the AI gives the first reply in every conversation. Email is therefore left out and its cell shows “—”: on email the AI is optional per inbox (the Let AI reply setting on the email channel, see Email channel). The AI reply time shows “—” for email too.
• Only WhatsApp and Instagram have a 24-hour window; for the widget, email and Telegram the Unanswered in 24 h column shows “—”.
• WhatsApp message charges (Meta): this month so far and last month, by category (service replies, utility, marketing, authentication…), with the number of charged and free messages. This is an approximation of what Meta charges your WhatsApp account; the exact amount is in WhatsApp Manager › Billing. Amounts are in your WhatsApp account's currency and don't use SemAgent credits. This card doesn't follow the date range: it always shows this month and last month.

Product cards

The section below Channels counts the product cards the AI showed in the e-commerce widget for the selected range: Cards shown, Product page clicks and Added to cart (both also as a share of cards shown), and how many different products were shown. The table below lists shown, clicks, adds to cart and click rate per product, starting with the most clicked.

• Each card in a reply counts once for a click and once for an add to cart: the same visitor pressing the same card ten times still counts once.
• No personal data is kept, only daily counts. Deleting a conversation doesn't change the numbers.
• Counting started on 24 October 2026; earlier days show nothing.

Weekly report email

Turn it on with the Weekly report email card on the Reports screen; it's off by default. When it's on, last week's summary arrives by email every Monday at 09:00: conversations and the change from the week before, the share the AI resolved alone, the share that went to a person, first human reply, handoff-to-resolution time, satisfaction, questions without an answer and your most active agents.

• Who gets it: people who can edit Reports, custom roles included (by default the account owner and admins; change it in Roles & Permissions), up to 20 people.
• The time zone comes from the working hours in Widget Settings; the card shows the next send time and the number of recipients.
• Send me a sample sends a sample of the current week to you only (at most 5 per hour).
• One email goes out per week; even if the server restarts, the same week isn't sent twice.