Skip to content

Metrics

Live Chat Metrics: 9 KPIs for Agent Productivity and Quality

Live chat metrics for support leads: 9 chat KPIs for agent productivity and quality, each with a formula, a 2024 benchmark and how to improve it.

· Updated · 18 min read

Reviewed by Dominik Blattner, founder of Kaizo

Part of: IQS Meaning: Internal Quality Score in Customer Service

Live chat metrics are the numbers that show how well your chat support performs, on speed, resolution and customer sentiment. The nine worth tracking are below, each with a formula, a sourced benchmark where one exists, and how to move it. Together they are the agent productivity metrics for chat that a support lead needs, as long as each one is paired with a quality score.

In short

  • Start with First Response Time, First Contact Resolution and CSAT.
  • LiveChat’s 2024 data puts the average first response at 35 seconds.
  • Compare chat numbers within chat, never against phone or email.
  • Pair every speed metric with a quality score on the same chats.
  • Use productivity numbers such as chats per agent for staffing, and coach from the conversations behind them.

Chat metrics are how you stop guessing. Measured properly, they tell you where agents are fast and where they stall, which questions keep coming back, and whether customers leave a chat happy or frustrated. This guide covers the nine that earn their place on a dashboard, then the one thing a metrics dashboard alone will never tell you. The numbers only pay off when they feed agent coaching from QA and support insights that explain why a number moved, so every section ends with what to change, not just what to watch.

Kaizo live chat metrics across multiple channels in one dashboard

Live chat metrics at a glance

The nine live chat metrics below cover speed, resolution, volume, sentiment and workload, each with a formula and a benchmark where one exists. Benchmarks come from the LiveChat Customer Service Report (2024 data), SQM Group for FCR and Bain’s NPS ranges as summarized by Qualtrics.

Benchmarks are only listed where we could check the original source. Where there is none, the target is your own trend line.

Chat metricFormulaWhat good looks like
First Response Time (FRT)Average of (first agent reply − customer’s first message)Average across LiveChat customers: 35 seconds (LiveChat, 2024)
First Contact Resolution (FCR)(Chats resolved on first contact ÷ total chats) × 10070 to 79% is good, 80% or more is world class for contact centers (SQM Group)
Average Resolution Time (ART)Total resolution time ÷ number of resolved chatsAverage chat lasts 8 minutes 25 seconds (LiveChat, 2024)
Total Number of ChatsCount of opened, missed and resolved chats per periodNo target. Watch for spikes and dips
CSAT(Positive responses ÷ total responses) × 100Average for rated chats: 64.2% (LiveChat, 2024). Beat your own baseline
NPS% promoters − % detractorsAbove 0 is good, above 50 is excellent, per Bain (Bain, via Qualtrics)
Missed and abandoned chats(Missed or abandoned chats ÷ total incoming) × 100Average queue dropout: 27.4% (LiveChat, 2024). Lower is better
Tags (contact reasons)Categorized count of tagged chatsNo target. Track the top drivers over time
Chats per agentTotal chats ÷ number of agentsAverage: 84.1 chats per agent per day (LiveChat, 2024). Use for staffing, not scoring

The metrics you choose will depend on your business, and more is not better. Pick the handful that map to your goals, get them reliable, and only then expand. For the wider picture of how these fit alongside phone and email measures, see our guide to customer service KPIs and metrics.

main live chat metrics to track

1. First Response Time (FRT)

What it measures: how long a customer waits between sending their first message and getting the first reply from an agent.

Formula: average of (time of first agent reply − time of customer’s first message) across all chats.

Benchmark: across LiveChat’s customer base the average first response is 35 seconds (LiveChat Customer Service Report, 2024). Treat that as a reference point, not a target for every team.

First response time sets the tone. It is usually the first impression a customer forms of your support team, and live chat carries a specific expectation: it is called live for a reason, so customers expect an almost immediate reply. Wait too long and frustration sets in before the conversation has even started.

Response-time expectations vary by platform and industry, so it is worth setting a target that matches yours rather than chasing a generic number. Our breakdown of what drives a high first reply time covers the usual culprits.

How to improve it:

  • Track FRT across days, weeks, and seasons. A sustained climb usually means agents are overwhelmed at specific times, which is a staffing signal, not a performance problem.
  • Use concurrency. Unlike phone, one agent can handle several chats at once, so tuning the number of concurrent chats per agent is one of the fastest ways to cut wait time.
  • Deploy saved replies for common openers so agents acknowledge the customer instantly while they work on the full answer.

first response time as a live chat metric

2. First Contact Resolution (FCR)

What it measures: the share of chats resolved in a single conversation, with no follow-up needed.

Formula: (chats resolved on first contact ÷ total chats) × 100.

Benchmark: SQM Group’s FCR benchmarks rate 70 to 79% as good and 80% or higher as world class. Those figures come from contact centers in general, so use them as a rough guide for chat.

FCR is one of the clearest predictors of customer satisfaction. When customers get a complete answer the first time, they walk away happy; when they have to come back, satisfaction drops with every repeat. A low FCR usually points to one of three things: agents need more training, your knowledge base is hard to search, or the issues themselves are genuinely complex.

How to improve it:

  • Coach and train agents so they have the tools and product knowledge to resolve issues on the spot rather than escalating.
  • Compare FCR across channels. A high FCR on phone but a low one on chat often signals a tooling or knowledge-access gap specific to chat.
  • Keep your internal knowledge base current so agents can find the right answer without putting the customer on hold.

3. Average Resolution Time (ART)

What it measures: the average time to fully resolve a chat, from the opening message to the moment the issue is closed.

Formula: total resolution time ÷ number of resolved chats.

Benchmark: the average chat lasts 8 minutes 25 seconds across LiveChat’s customers (LiveChat’s chat duration data, 2024).

ART shows how efficiently your team closes issues and highlights which agents or topics take longest, which is a coaching opportunity. But treat it as a context metric, never in isolation.

Consider two customers with the same issue: one is resolved in a day but kept informed the whole way and feels reassured, the other is resolved in an hour with a single terse reply. Faster is not automatically better. A longer conversation where the customer feels heard can beat a quick one that leaves them cold.

Because ART is so easy to misread, it is the metric where scores need to become coaching rather than a stick. Tie every slow chat back to a specific conversation and use it to teach, not to rank. Kaizo turns quality findings into per-agent AI coaching cards, so managers coach from real examples instead of compiling spreadsheets. More on that below.

How to improve it:

  • Meet customers on the channels they already use, so they respond quickly and conversations do not stall.
  • Organize canned responses for frequent questions to deliver faster, consistent replies.
  • Route chats automatically to the right team so customers are not bounced between agents.

Kaizo coaching card generated from chat quality data

4. Total Number of Chats

What it measures: the high-level count of opened, missed, and resolved conversations over a set period.

Formula: a simple count of chat conversations in your chosen window, broken down by status.

Benchmark: this one is trend-based rather than a fixed target. What matters is the shape of the curve and the anomalies in it.

Total chat volume is your engagement gauge and a productivity input, but its real value is in the surprises. A low number during a peak season can mean customers cannot find your chat widget. A high number of the same query means the answer is too hard to find on your site. Read the volume as a diagnostic, not just a tally.

How to improve (or interpret) it:

  • Publish answers to the most common queries on an FAQ or help page to keep repeat questions out of the queue.
  • Adjust staffing to match seasonal peaks so higher volumes do not degrade your other metrics.

live chat in customer service

5. Customer Satisfaction (CSAT)

What it measures: how satisfied customers are immediately after a chat, usually captured with a one-click post-chat survey.

Formula: (positive responses ÷ total responses) × 100.

Benchmark: LiveChat reports an average satisfaction score of 64.2% for rated chats, and only 6.7% of chats get a rating at all (LiveChat’s satisfaction data, 2024). With so few ratings, your own trend matters more than the industry figure.

CSAT is the staple satisfaction metric across every channel, and live chat suits it well because you can ask for feedback the moment the conversation ends, while the experience is fresh. Positive and negative ratings together paint a clear picture of where service is thriving and where it needs work. Measured regularly, CSAT reveals trends in customer behavior and gives you real-time signal to act on.

Speed and satisfaction are linked. Long waits put customers in a bad mood before the conversation starts, so fixing FRT is often the fastest route to a better CSAT.

How to improve it:

  • Improve First Response Time so customers are not waiting when the survey lands.
  • Coach and train to lift First Contact Resolution, since resolved-first-time chats score highest.
  • Use canned responses to speed up Average Resolution Time without sacrificing warmth.

For the full playbook, see how to improve customer satisfaction.

customer satisfaction survey for live chat

6. Net Promoter Score (NPS)

What it measures: how likely customers are to recommend you, on a 0 to 10 scale.

Formula: % promoters − % detractors. Scores of 9 to 10 are promoters, 7 to 8 are passives, and 0 to 6 are detractors.

Benchmark: Bain & Company, which created NPS, treats any score above 0 as good and above 50 as excellent (Qualtrics).

NPS is a single question that reveals a lot. The catch is that happy customers rarely volunteer their enthusiasm, which is where live chat helps: a survey that appears automatically at the end of a chat prompts a response in the moment, while goodwill is high. Detractors (0 to 6) are at risk of spreading negative word of mouth, passives (7 to 8) are satisfied but not loyal, and promoters (9 to 10) are the advocates worth nurturing.

How to improve it:

  • Engage detractors proactively. Their feedback is uncomfortable but it is the cheapest early warning you will get.
  • Follow up with promoters through chat, with an offer, a request for a review, or a simple thank you.
  • Route the feedback to the decision-makers who can actually act on it.

NPS survey as a chat metric

7. Missed and abandoned chats

What it measures: the requests you never answered (missed chats) and the ones customers gave up on before an agent replied (abandoned chats).

Formula: (missed or abandoned chats ÷ total incoming chat requests) × 100.

Benchmark: keep it as low as you can. Across LiveChat’s customers, 27.4% of people in a queue left before an agent picked up (LiveChat’s queue data, 2024).

Making a customer wait is one problem. Missing their message entirely is worse. If you offer live chat, customers expect availability, and a consistently high missed-chat rate, or sharp swings in it, is a clear sign something needs fixing. Unanswered chats do not just dent your metrics, they damage trust, so contacting the customers whose chats you missed matters more than the number itself.

How to improve it:

  • Use AI chatbots to handle simple questions when agents are unavailable, so no request goes completely unanswered.
  • Talk to agents about workload and multitasking, and make it normal to ask for help when the queue spikes.
  • Pressure-test your tooling. Is your platform actually provisioned for your peak chat volume?

8. Tags (contact reasons)

What it measures: why customers are chatting in, captured by categorizing each conversation.

Formula: a categorized count of tagged chats, ideally automated by your support software over time.

Benchmark: there is no target number. The value is in seeing your top contact drivers and how they shift.

Tags let whoever reviews a chat classify what it was about, and over time a good support tool automates that collection. The payoff is knowing which questions come up most, so you can route them to help-center articles, write new ones, or fix the underlying product or process issue. Tagging also builds a library of reference material for training on rare, complex cases.

How to improve it:

  • Make category creation a deliberate, top-down decision that maps to your KPIs, so you do not drown in tags that measure nothing useful.
  • Be consistent. Tagging only works if everyone applies the same categories the same way.

9. Chats per agent

What it measures: average agent workload and how efficiently you are using chat capacity.

Formula: total chats ÷ number of agents over the month.

Benchmark: LiveChat’s average is 84.1 chats per agent per day (LiveChat’s workload data, 2024), but your number depends on chat complexity and concurrency. Use it for staffing decisions rather than as a performance score.

Chats per agent tells you whether your investment in chat is being used well, but it is volatile and shaped by forces outside any team lead’s control, from seasonal spikes to one-off events. December almost always brings a surge, which teams cover with seasonal hires or outsourced support. High numbers can tip agents into overload, and a persistently low number can signal a deeper problem, so read this metric in context.

How to improve it:

  • To lower it, revisit workflow and staffing, and check in with agents. A high number can mean someone is in permanent hyper-drive, so make sure they have support and take their breaks. Sustained overload is a leading cause of call center burnout.
  • To raise it, match the work to the skill. Handling high chat volume rewards fast, accurate typists, so route heavy chat load to the agents suited to it, and support the rest.

Productivity numbers say how much work got done, not how well. For the wider playbook on lifting both, see our guide on how to improve agent performance.

Chat vs. email vs. phone: how to track metrics by channel

Track the same core metrics on every channel but set targets per channel: chat is judged in seconds and concurrency, email in hours and completeness, phone in queue time and handle time.

MetricLive chatEmailPhone
First responseSeconds, while the customer waits on the pageHours, set by your SLATime to answer in the queue
ResolutionWithin the session, often across concurrent chatsAcross several repliesWithin the call, plus callbacks
WorkloadChats per agent and concurrent chatsTickets solved per agentCalls per hour and handle time
AbandonmentCustomers who leave the queue before a replyRare: shows as no reply to a follow-upCalls abandoned in the queue
SatisfactionPost-chat survey at the end of the sessionSurvey after the ticket is solvedPost-call or IVR survey

Compare each channel with its own history, never with the others. A chat team with a lower FCR than phone may be handling different questions, while email often carries the complex cases. For the quality side of voice, our guide to call center QA metrics covers the numbers that predict CSAT on the phone.

Beyond the dashboard: what chat metrics do not tell you

Chat metrics measure speed, resolution and satisfaction, but not whether the agent’s answer was correct, compliant and on-brand. That gap is where most chat quality problems hide.

Chat metrics measure outcomes: how fast you replied, how often you resolved, how satisfied customers said they were. What they do not measure is whether the answer your agent gave was actually correct, compliant, and on-brand. A chat can close in 90 seconds with a delighted CSAT rating and still contain the wrong information.

Quality is judged separately, and traditionally it is judged on a small manual sample of a few chats per agent. Your speed dashboard covers every conversation while your quality view covers only a fraction of them, and that mismatch is where problems hide. This is also why measuring the Internal Quality Score (IQS) alongside your chat metrics matters: CSAT tells you the customer was happy, IQS tells you the work was actually good. Our guide to building a conversation quality score shows how to set that score up.

How full-coverage QA closes the gap

The fix is to stop sampling. AI-based auto-QA evaluates every chat against your scorecard instead of a handful. Kaizo’s AutoPilot scores every conversation against your own criteria and leaves scores and comments as soon as a case resolves, so coverage stays at 100% without anyone triggering reviews, and Auto QA drafts the reviews a person still needs to check. Your quality score stops being an estimate from a sample and becomes a measurement of every chat, and your QA team shifts from grading to coaching.

In a June 2024 survey by Gartner, 52% of service leaders who rate their QA program as highly valuable said that value comes mainly from voice-of-the-customer and CX insights, and fewer than 20% pointed to rep performance. The record of what happens in your chats is one of the richest sources of customer intelligence you have, and reading a small sample of it leaves most of that value on the table.

What this looks like in practice: at SteelSeries, a 30-person support team, AI summaries cut QA time in half on long, technical tickets. If you are comparing tools for this job, our roundup of the best software for monitoring live chat quality lays out the options.

See it in a live demo. Kaizo scores 100% of your support conversations and turns the findings into coaching. Let’s talk.

When 100% chat QA is not for you

Automated scoring is not always the right call. If your team handles a few dozen chats a week, one lead can read every conversation by hand, and a spreadsheet will do. If you have no written scorecard yet, start there: automating a standard nobody has agreed on only produces fast, confusing scores. The case for full coverage starts when volume outgrows what people can read.

productivity goals and performance tracking in Kaizo

Chatbot and AI agent metrics for live chat

Chatbots and AI agents need the same outcome metrics as human agents, plus three of their own: automated resolution rate, handover rate, and whether customers repeat themselves after a handover.

As chat teams deploy AI agents, someone has to grade the quality of those conversations too. Kaizo’s QA for AI agents scores 100% of human and AI-agent conversations against your own scorecard, so as your chat channel becomes a mix of human and automated responses, both are held to one standard.

Bot metricFormulaWhat it tells you
Automated resolution rateChats fully resolved by the bot ÷ chats the bot handled × 100Real resolution, not deflection
Handover rateChats passed to a human ÷ chats the bot handled × 100Where the bot runs out of answers
Repeat after handoverHandovers where the customer restated the issue ÷ all handovers × 100Whether context travels with the chat
Bot CSATPositive ratings on bot chats ÷ rated bot chats × 100How customers felt about the bot alone
Bot quality scoreBot conversations scored against your scorecardAccuracy, policy and tone, on the same standard as humans

Most of the nine metrics above apply to AI-handled chats too: resolution, CSAT and abandoned chats still tell you whether the customer got what they came for. For a benchmark, LiveChat measured average chatbot satisfaction at 64.7% across its customers (LiveChat’s chatbot data, 2024). Our guide to containment rate explains why a high automation number can hide customers who simply gave up.

Our guide on how to QA AI agents covers the scorecard, and AI customer service metrics covers the numbers to report to leadership.

Build your live chat metrics report

Start with three live chat metrics, First Response Time, First Contact Resolution and CSAT, report them weekly by team, and add a quality score before you add more speed numbers.

  1. Pick the core three. They cover speed, effectiveness and satisfaction with the least overhead, and improving one often lifts the others.
  2. Segment before you compare. Split every number by contact reason (your tags), hour of day and channel, so a busy Monday does not look like a performance problem.
  3. Pair speed with quality. Put the quality score for the same chats next to FRT and ART, so faster agents are not rewarded for weaker answers.
  4. Add the next layer once it is reliable. A coverage metric (missed and abandoned chats) and chats per agent come next, so you know who never got an answer and whether the workload is sustainable. Too many metrics at once causes a lack of focus.

You cannot manage what you cannot see, and that is the whole case for chat metrics. Too many team leads still stitch performance together from a stack of spreadsheets, spending more time compiling numbers than acting on them. The point of a metric is the decision it lets you make, so pick the handful that map to your goals, get them reliable, and change the strategy as you learn.

Just remember what the dashboard cannot show you. Speed and satisfaction numbers describe every chat, but the quality of the answers inside them is usually judged on a sliver. Close that gap, measure the content of your chats as completely as you measure their outcomes, and your metrics start reflecting reality instead of a sample of it.

To see how scores become coaching cards per agent, look at coaching cards built from QA scores in Kaizo. If you want to see what chat quality looks like when every conversation is scored automatically and turned into coaching, book a meeting for a live demo with realistic sample conversations.

Frequently asked questions

What are live chat performance metrics?

Live chat performance metrics are the measurements that show how well your live chat support performs, covering both speed (First Response Time, Average Resolution Time), effectiveness (First Contact Resolution, missed chats), and customer sentiment (CSAT, NPS). Tracked together, they tell you how your team is doing and where to improve, provided you pair the outcome numbers with a quality measure of whether answers were actually correct.

What is a good response time for live chat?

The average first response across LiveChat’s customers is 35 seconds (LiveChat’s 2024 report), so a reply inside a minute keeps you in a reasonable range. Speed matters because customers drop out of slow queues: LiveChat measured an average queue dropout of 27.4%.

What is a good CSAT score for live chat?

LiveChat’s 2024 satisfaction benchmarks show an average score of 64.2% for rated chats. Only 6.7% of chats get rated, so compare against your own baseline before the industry figure. The quickest way to raise CSAT is usually to cut First Response Time and lift First Contact Resolution.

How does chat support software improve CX metrics?

Chat support software improves CX metrics mainly through speed and context. Routing, saved replies and concurrency cut First Response Time, and a shared customer history helps agents resolve the issue in one chat, which lifts First Contact Resolution. Those gains reach CSAT and NPS only when answer quality keeps pace, so score the content of the chats, not just the clocks.

What are agent productivity metrics for chat?

The main agent productivity metrics for chat are chats per agent, First Response Time and Average Resolution Time. LiveChat’s 2024 average is 84.1 chats per agent per day. Read these numbers for staffing and workload, and pair them with a quality score so faster agents are not rewarded for weaker answers.

How do you measure chatbot or AI agent performance in chat?

Use the same core metrics as for human agents: resolution, CSAT and abandoned chats. Then add how often the AI agent hands a chat over to a human, and whether the customer had to repeat themselves afterwards. Kaizo scores human and AI-agent conversations against the same scorecard, so both are graded the same way and you can compare them fairly.

In KaizoDashboardsCoverage, quality trends and coaching impact report natively. No BI project, no monthly assembly job.See Dashboards

On this page

See this in a live demo

We will walk you through Kaizo with realistic sample conversations, so you see the example scored, explained and coached.

Trusted by global support teams

  • Foot Locker
  • Canva
  • GetYourGuide
  • Instacart
  • BlaBlaCar
  • UiPath