---
title: "Live Chat Metrics: 14 KPIs, Formulas and What to Track"
description: "The live chat metrics that matter for sales and support, with formulas, Zendesk and Intercom definitions, and the AI chatbot metrics older lists leave out."
canonical: https://luckpanda.app/blog/live-chat-metrics
author: "Dhrumil Barot"
author_bio: "Dhrumil is a small-business operations and compliance expert. Dhrumil built LuckPanda so small businesses can run invoicing, inventory, accounts and compliance in one place."
published: 2026-11-01
updated: 2026-11-01
category: "Customer support and chat"
keywords: "live chat metrics"
language: en
---

# Live Chat Metrics: 14 KPIs, Formulas and What to Track

The live chat metrics that matter for sales and support, with formulas, source definitions and the AI chatbot metrics older lists leave out.

By [Dhrumil Barot](https://luckpanda.app/about#founder), Founder. Published 1 Nov 2026. Updated 1 Nov 2026. 7 min read.

**TL;DR:** Live chat metrics are the numbers that show whether website chat is answering people fast, solving their problem and turning visitors into customers. Track four groups: speed (first response time), quality (resolution rate, CSAT), sales (chat-to-lead and chat-to-sale conversion) and, if a bot answers first, bot metrics (resolution rate, handoff rate, unanswered-question rate). Pick five to eight, define each formula once, and review them weekly.

Live chat metrics are measurements of a website chat channel's speed, quality and business result. The most useful set for a small business is short: first response time, resolution rate, customer satisfaction (CSAT), chat-to-lead conversion, chat-to-sale conversion, handoff rate and the list of questions nobody could answer. Most published lists (Comm100's goes back to 2014, MetricNet's to 2020) were written for teams of human agents, so they skip what changes when an AI chatbot takes the first reply. This page gives each metric a formula and a source definition, then covers the bot metrics.

## Which live chat metrics should you track? (lookup table)

The table below is the reference. Formulas use counts from one period, usually a week.

| Metric | Formula | Group | Watch for |
| --- | --- | --- | --- |
| Chat volume | count of chats started | Demand | Spikes after a price or policy change |
| Chat rate | chats / website sessions x 100 | Demand | Pages with unusually high chat rates |
| First response time (FRT) | median seconds from first visitor message to first reply | Speed | Use the median, not the mean |
| Missed chat rate | chats with no reply / chats started x 100 | Speed | Out-of-hours gaps |
| Average handle time (AHT) | total chat duration / chats handled | Effort | Too low can mean rushed answers |
| Resolution rate | chats resolved without follow-up / chats x 100 | Quality | How "resolved" is defined |
| First contact resolution (FCR) | issues solved in the first chat / issues x 100 | Quality | Repeat chats from the same person |
| CSAT | ratings of 4 or 5 / all ratings x 100 (5-point scale) | Quality | Low response rate to the survey |
| Transfer or handoff rate | chats passed to a person / chats x 100 | Bot and routing | Rising rate after content changes |
| Unanswered-question rate | chats where the answer was not in your content / chats x 100 | Bot | The questions themselves |
| Chat-to-lead conversion | chats that left contact details / chats x 100 | Sales | Asked too early or too late |
| Chat-to-sale conversion | chatters who bought within N days / chatters x 100 | Sales | Fix N (7 or 30 days) and keep it |
| Revenue per chat | revenue from chatters / chats | Sales | One large order skews it |
| Proactive invite acceptance | invites answered / invites shown x 100 | Sales | Invites firing too soon |

Which five to eight to start with depends on what chat is for. | If chat is mainly for | Track |

| Sales | Chat volume, chat-to-lead conversion, chat-to-sale conversion, unanswered questions, handoff rate |
| --- | --- |
| Support | Median FRT, missed chat rate, resolution rate (confirmed and assumed), CSAT with response rate, handoff rate |
| Both, with an AI bot first | Bot resolution split, handoff rate and reason, unanswered questions, chat-to-lead conversion, CSAT |

Write each formula down once (what counts as resolved, which days count toward a sale, whether autoreplies stop the clock) and do not change it without noting the date. A metric whose definition drifts is worse than no metric. For lead-focused setups, see [chatbots for lead generation](https://luckpanda.app/chatbot/use-cases/lead-generation) and [customer support use cases](https://luckpanda.app/chatbot/use-cases/customer-support).

Exact values for a few thresholds worth knowing:

- Zendesk example first reply SLA for messaging: 30 seconds
- Zendesk minimum SLA target: 15 seconds
- NPS: % promoters (9-10) minus % detractors (0-6), range -100 to +100

## What are the speed metrics, and how are they defined?

Speed metrics measure how long a visitor waits. First response time is the time between the visitor's first message and the first reply. Zendesk defines first reply time as "the time between ticket creation and the first public comment from an agent (or autoreply)" ([Zendesk: Defining SLA policies](https://support.zendesk.com/hc/en-us/articles/4408829459866)). Note the "or autoreply": an automatic "thanks, we'll be with you shortly" stops that clock, which flatters the number. Decide whether an autoreply counts and write it down.

Zendesk lets you set SLA targets for `First reply time`, `Next reply time`, `Periodic update`, `Requester wait time`, `Agent work time` and `Total resolution time`, in business or calendar hours. It suggests that "messaging customers may want to define an SLA where the first reply time has a 30 second target", and the minimum target you can set is 15 seconds. An SLA only applies when the ticket's system `Priority` field (`Low`, `Normal`, `High` or `Urgent`) has a value, so unprioritised chats fall out of the report.

Report FRT as a median. One chat that waited three hours overnight will drag a mean far above what most visitors experienced. Report missed chats separately: a chat with no reply has no response time at all and disappears from an FRT average.

## How do you measure chat quality: resolution, FCR and CSAT?

Resolution rate is the share of chats where the visitor got what they came for and did not need to come back. The hard part is the definition, because most chats end with the visitor simply closing the window.

Intercom's definition for its Fin AI agent is the clearest published one. A resolution is counted when, after Fin's last answer, the customer either confirms the answer helped ("Ok thanks") or "exits the conversation without requesting further assistance" ([Intercom: Fin AI Agent outcomes](https://www.intercom.com/help/en/articles/8205718-fin-ai-agent-resolutions)). Intercom calls the first a `confirmed resolution` and the second an `assumed resolution`. Two exclusions matter: replying to a greeting is not a resolution, and if the customer comes back to the same conversation later, the resolution is deducted. If the bot's last message was a clarifying question and the visitor left, no resolution is recorded.

Use the same split yourself: report confirmed and assumed resolutions as two numbers. An assumed resolution rate that rises while CSAT falls usually means visitors are giving up, not getting answers.

First contact resolution (FCR) is stricter: was the issue solved in the first chat, with no repeat chat, email or call about it. Comm100 cites an average FCR rate of 66% from research it does not name ([Comm100](https://www.comm100.com/blog/measuring-live-chat-effectiveness.html)); treat that as a rough reference, not a benchmark.

CSAT is the share of post-chat ratings that are positive, usually 4 or 5 on a 5-point scale. Always show the survey response rate next to it. A CSAT of 95% from 3% of chats tells you about the happiest 3%.

## Which live chat metrics show sales, not just support?

Sales metrics connect a chat to money. Comm100 split its list into sales and support goals for this reason: a sales-led business should watch visitor-initiated chats, proactive invitation acceptance rate, sales conversion rate and average sale per conversion.

- **Chat-to-lead conversion:** chats that ended with contact details or a booked call, divided by all chats. If it is low, check when the chat asks for an email. Asking before answering the question loses people; asking once the visitor wants something (a quote, a call, a sample) keeps them.
- **Chat-to-sale conversion:** visitors who chatted and bought within a fixed window. Pick 7 days or 30 days and do not change it, or month-to-month numbers stop being comparable.
- **Revenue per chat:** revenue from chatters divided by chats. It answers "is chat worth staffing?" in one number.
- **Lead quality:** the share of chat leads your sales process accepts. Ten leads a week that never answer the phone are worth less than three that do.

To count chat leads in Google Analytics 4, send the recommended `generate_lead` event (with `currency` and `value` if you know them, plus a custom `method` parameter such as `website_chat`) when a chat captures contact details, then mark it as a key event; Google renamed "conversions" to "key events" in March 2024 ([Google Analytics: recommended events](https://support.google.com/analytics/answer/9267735)).

When the chat widget captures a lead, the call is: `gtag('event', 'generate_lead', { currency: 'USD', value: 50, method: 'website_chat' });`

## What changes when an AI chatbot answers first?

An AI chatbot answering first changes which numbers mean anything. First response time drops to a few seconds for every chat, so it stops telling you much. Three metrics take its place.

**Bot resolution rate** uses the confirmed and assumed split above. Intercom prices Fin per outcome on this basis: $0.99 per resolution, procedure handoff, disqualification or self-serve routing, and $9.99 per qualified prospect, charged once per conversation. That pricing is also a useful reminder that vendors who bill per resolution have a reason to define it generously.

**Handoff rate** is the share of chats passed to a person. Neither zero nor high is good. Zero can mean the bot is answering things it should not; high means your content is missing answers. Track why each handoff happened: the visitor asked for a person, the answer was not in the content, or the request needed an action (refund, custom quote).

**Unanswered-question rate** counts chats where the bot said it did not have the answer. The rate matters less than the list. Each unanswered question is either a gap in your help content or a product people want that you do not sell.

MetricNet lists chat-specific metrics for service desks that still apply here: the percentage of contacts that start in chat, the percentage resolved in chat, the failover rate from chat to another channel, and average and maximum concurrent chats ([MetricNet](https://www.metricnet.com/resources/metrics-that-matter-chat-metrics)).

Worked example for one week, so the formulas are concrete:

| Input | Metric | Result |
| --- | --- | --- |
| 4,000 sessions, 120 chats | Chat rate | 120 / 4,000 = 3.0% |
| 6 chats with no reply | Missed chat rate | 6 / 120 = 5.0% |
| 18 leads captured | Chat-to-lead | 18 / 120 = 15.0% |
| 9 bought within 7 days | Chat-to-sale | 9 / 120 = 7.5% |
| 24 handed to a person | Handoff rate | 24 / 120 = 20.0% |
| 40 of 50 ratings were 4 or 5 | CSAT (response rate 41.7%) | 80% |

These are example numbers, not benchmarks.

## How do you read chat transcripts as customer feedback?

Chat transcripts are the one metric source written in the customer's own words. Counts tell you something went wrong; transcripts tell you what. A weekly 20-minute read beats a dashboard nobody opens:

1. Sort last week's chats by outcome: resolved, handed off, unanswered, lead.
2. Read every unanswered chat and group them by topic (delivery, pricing, a missing feature, a missing product).
3. Count how many different people asked each thing, not how many messages.
4. Fix the top content gap the same day: add the answer to the page or FAQ the bot reads.
5. Send product requests (things people want that you do not offer) to whoever decides what you sell.

[LuckPanda Chatbot](https://luckpanda.app/chatbot) does steps 2 and 3 for you: every chat is saved word for word with the page it started on, leads arrive with a summary of the need, timeline and size, and with request logging on, things visitors asked for that you do not offer are grouped on a Requests board, ranked by how many different people asked, with a weekly digest. Chats that need a person are flagged as waiting so handoffs are counted, not lost. See [how human handoff works](https://luckpanda.app/blog/chatbot-human-handoff) and [how a knowledge base chatbot answers from your content](https://luckpanda.app/blog/knowledge-base-chatbot).

## Questions

### What are the most important live chat metrics?

For most small businesses: median first response time, resolution rate, CSAT, chat-to-lead conversion and the list of unanswered questions. Add chat-to-sale conversion if chat is part of how you sell.

### What is a good first response time for live chat?

Zendesk's own example for messaging is a 30-second first reply target, and Comm100 suggests 30 seconds or less to prevent abandoned chats. Measure it as a median and decide whether automatic replies stop the clock.

### How do you calculate chat conversion rate?

Divide the number of chats that led to the outcome (contact details left, or a purchase within a fixed number of days) by the number of chats, then multiply by 100. Keep the time window fixed between reports.

### How is chatbot resolution rate measured?

Intercom counts a resolution when the customer confirms the answer helped or leaves without asking for more help after the bot's last answer. Report confirmed and assumed resolutions separately, because assumed ones include people who gave up.

### What is the difference between CSAT and NPS for chat?

CSAT rates one chat, usually on a 1 to 5 scale. NPS asks how likely someone is to recommend the business, from 0 to 10, and is the share of promoters (9 to 10) minus detractors (0 to 6) (Bain: measuring NPS). Use CSAT for chat quality.

### Which live chat metrics matter for an AI chatbot?

Bot resolution rate (confirmed versus assumed), handoff rate with the reason, and the unanswered-question list. First response time stops being informative because the bot replies in seconds.

## More guides

- [Customer Feedback Categories: A Codebook With Examples](https://luckpanda.app/blog/customer-feedback-categories)
- [How to Add a Chatbot to Squarespace (Step by Step)](https://luckpanda.app/blog/chatbots-for-squarespace)
- [Knowledge Base Chatbot: How It Works and How to Build One](https://luckpanda.app/blog/knowledge-base-chatbot)

## Related tools

- [Invoice generator](https://luckpanda.app/tools/invoice-generator): Bill a client in any currency, with tax worked out as you type.

## About the author

**Written by [Dhrumil Barot](https://luckpanda.app/about#founder)**, Founder. Dhrumil is a small-business operations and compliance expert. Dhrumil built LuckPanda so small businesses can run invoicing, inventory, accounts and compliance in one place. [LinkedIn](https://www.linkedin.com/in/bdhrumil) · [X](https://x.com/fyidhrumil)