How each metric is calculated

Every metric in Konvo is counted from your own conversations, using the rules below rather than an industry average or an estimate. Because you can put any metric on a dashboard you build, knowing exactly what each one counts is the difference between a number you can act on and a number you can only quote.

The rules that apply to everything

• A conversation is a ticket: one customer thread, created at their first message. It can close and reopen, but it only counts as new once.
• Your AI counts as an agent, so it appears alongside your people rather than in a separate report.
• Spam is excluded from every metric except the spam count.
• Time metrics are medians, not averages, so one nightmare conversation does not drag your whole week. They currently run on plain clock time around the clock, which means a message arriving on Friday night and answered on Monday morning counts the whole weekend. Read your response times against when your team actually works.
• Two ways of counting. Conversation metrics count in the period the conversation was created. Event metrics (closes, reopens, replies, handovers) count in the period the event happened.

The five outcomes everything else is built on

Every conversation ends in exactly one of these, and most AI metrics are ratios between them:

• Resolved. The conversation actually resolved: your agent or a person marked it resolved, the customer confirmed they were happy, or they simply stopped replying without asking for a person. It counts as resolved whoever got it there.
• Transferred. A person took over a conversation that had been routed to your agent — by your agent's choice or the customer's request, and whether or not your agent had replied — and it has not resolved yet.
• Human-handled. The conversation never went to your agent, and has not resolved yet.
• Constrained. Your agent is blocked waiting on your team: it asked for input, or created a task nobody has picked up.
• Still automated. Your agent is still on it and none of the above applies yet.

The pair worth watching is Resolved against Transferred: your agent finishing the job, against your agent meeting something it could not. Transfers are not failures, they are a list of the things worth teaching it next, and they are the cheapest improvements available to you because each one is already written down as a real conversation.

Volume

• New conversations. Created in the period. Counted once, on their creation date; a reopen never re-counts them.
• Open conversations. Still awaiting resolution right now. A snapshot, not a flow.
• Closed conversations. Closed in the period, counted once per close.
• Reopened conversations. Closed conversations that came back. A quality signal: resolutions that did not stick.
• Conversations by channel. Grouped by where the first message arrived, even if the thread later moved.
• Messages received. Inbound customer messages, which measures contact intensity rather than demand.

Speed

All medians.

• First response time. First customer message to the first reply, from your agent or a person.
• Response time. Across every customer-message-to-reply pair, not just the first.
• Response time after transfer. From your agent handing off to the first human reply. This is the one your customers feel most and the one most teams never look at.
• Resolution time. First message to the final close. Because it uses the final close, reopen loops make it longer.
• Time to first close. The same, but to the first close, ignoring later reopens.
• Handle time. A person's actual working time on a conversation, not counting time waiting on the customer.

Resolution time and time to first close being far apart is worth a look: it means things are being closed and reopened rather than solved.

AI performance

These three multiply, which is the most useful relationship in the whole set:

Involvement rate = AI-involved ÷ all new conversations
Resolution rate  = AI-resolved ÷ AI-involved
Automation rate  = AI-resolved ÷ all new conversations
                 = Involvement × Resolution

• Involvement rate. How often your agent even gets to try. Low means channels or hours are keeping it out, not that it is bad.
• Resolution rate. How often it finishes once it engages. Low means it lacks knowledge or tools.
• Automation rate. The headline. Because it is the product of the other two, a disappointing number always decomposes into one of two very different problems, and it is worth knowing which before you act.
• Transfer rate. Conversations a person took over, over all conversations routed to your agent.
• Automated conversations. The count your agent replied in and resolved without a person taking over. An approved draft still counts; a takeover does not.
• Handovers. Conversations that moved from your agent to a person, counted once each.

Effort

• One-touch rate. Closed with a single reply, over all closed.
• Zero-touch rate. Closed with no human reply at all. This is the truest measure of what your AI is saving you.
• Replies sent. Customer-facing replies from your team and your agent.
• Messages per conversation. Median, counting both sides. Read it against resolution time: fewer messages with steady resolution time means answers are landing first time, while fewer messages and rising resolution time usually means conversations are stalling rather than being resolved efficiently.

Cost and usage

• Time saved. Automated conversations times the median human handle time.
• Cost saved. Automated conversations times the difference between what a conversation costs you when a person handles it and what it costs when your agent does. The human side comes from your own median handle time, so this figure moves with your team's real speed rather than a fixed assumption.

Your usage against your plan lives on the Plans and billing page rather than in Analytics.

A conversation becomes billable when your agent sends more than one message in it, so a greeting that immediately hands off to a person does not count.

Quality and topics

• Sentiment distribution, and how it moves over time.
• Negative-sentiment conversations. The count worth a view of its own.
• Conversations by category, and which categories are growing fastest against the previous period.
• Language distribution and priority distribution.

Your team

• Agent performance. One row per person, with your AI as a benchmark row: assigned, closed, first response time, resolution time, one-touch, replies.
• Assigned and closed per agent. Workload and throughput.

Reassignments count for each holder, so team totals can exceed the number of unique conversations. That is expected rather than double counting.