Live Chat Concurrency: Chats Per Agent at Once

Live chat concurrency is the number of conversations one agent handles at the same time. It is the single setting that decides whether your chat feels fast or feels abandoned, and most teams never set it on purpose: they let agents take whatever arrives until someone complains. This guide explains how to pick a concurrency target, how it interacts with response time and quality, and how an AI assistant that answers the routine questions changes the number you need.

There is no universal right figure. A team answering “where is my order” questions can run four or five chats per agent; a team troubleshooting software can barely manage two. What follows is a way to find your own number and keep it honest.

What Live Chat Concurrency Means in Practice

Unlike a phone call, a chat has gaps: the visitor types, reads, checks their order, walks to another room. An agent can use those gaps to answer another visitor. Concurrency is how many such conversations the agent juggles at once. A concurrency of three means three open conversations with visitors who are all waiting on, or typing to, the same person.

The gain is capacity: three parallel chats triple the throughput of one agent, at least on paper. The cost is attention. Every extra chat adds switching, slows each reply a little, and raises the chance of answering the wrong visitor or missing a message. Past a certain point the extra chat makes every conversation slower, and the throughput gain disappears into longer handle times and repeat contacts.

Typical Live Chat Concurrency by Conversation Type

The table below is a starting point, not a benchmark. Measure your own handle times before committing to a number.

Conversation type	Typical length	Reasonable concurrency	Why
Order status, delivery, opening hours	2–5 minutes	4–5	Short, scripted, long visitor pauses
Pre-sales product questions	5–10 minutes	3–4	Some lookup, moderate thinking
Account, billing, returns	8–15 minutes	2–3	Needs records, careful wording
Technical troubleshooting	15–40 minutes	1–2	Step-by-step, mistakes are costly
Complaints and escalations	10–30 minutes	1	Tone matters; divided attention shows
After an AI assistant handles routine questions	Mostly the longer types above	Lower than before	What reaches humans is harder on average

How Concurrency, Response Time and Handle Time Connect

Three numbers move together. First-reply time is how long a visitor waits for the first human answer; our guide to first response time in live chat covers targets. Average handle time is how long a conversation takes from start to close. Concurrency is how many run in parallel. Raise concurrency and first-reply time drops at first, because more visitors get picked up, but handle time rises because each reply is slower. Raise it further and first-reply time climbs again, because agents are too busy to notice a new chat.

Queueing theory describes this precisely, and you do not need the maths to use the idea. Little’s law says the number of conversations in progress equals the arrival rate multiplied by the average time each one stays open. If 30 chats arrive per hour and each stays open 10 minutes, you have on average five conversations open at any moment. With a concurrency target of 2.5 per agent, that is two agents on the desk. The Erlang unit is the same idea expressed for contact centres, and the same formulas size chat teams once you account for parallel handling.

A worked example

A shop receives 240 chats on a weekday between 09:00 and 18:00, roughly 27 an hour, with a peak of 45 an hour around lunch. Average handle time is 8 minutes. At the peak, 45 × 8 ÷ 60 = 6 conversations are open at once. With concurrency 3, two agents cover the peak; with concurrency 2, three are needed. If the AI assistant answers half of those chats without a hand-off, the peak that reaches humans is about 3 open conversations, and one agent with concurrency 3 covers it, with a second one as backup for the harder cases.

Finding Your Own Number

Measure handle time by type. Export a month of conversations and group them by subject. The long types set your ceiling.
Start at two. Run a week with a maximum of two chats per agent and record first-reply time, handle time, ratings and repeat contacts.
Raise to three and compare. If handle time per chat rises by more than a quarter, or ratings fall, three is too many for that agent on that mix.
Set per agent, not per team. An experienced agent on simple questions can run four; a new hire on returns should run two. Concurrency is a skill setting, not a policy.
Review monthly. Product launches, sales and policy changes shift the mix. Our article on busiest hours and live chat staffing shows how to read the hourly pattern.

How Talkmio Supports Concurrency

Talkmio does not set a hard cap on open conversations per agent; the number is a team decision you enforce through assignment and status. What the product does is reduce how much reaches an agent and make juggling safer.

Mio answers first. The AI assistant answers from your website, FAQ and uploaded documents, in the visitor’s language, and hands over only when it cannot answer from your content, when the visitor asks for a person, complains or asks about a specific order. Routine questions never reach the queue, which is where the biggest concurrency gain sits.
Inbox filters. Mine, Unassigned and Unanswered views show each agent what is theirs and what is waiting, so an agent at their limit can leave Unassigned alone and a teammate picks it up.
Assignment and status. Conversations are assigned to a team member from the header, with priority from Low to Urgent and status Open, Pending or Solved. Pending is the right status for a chat waiting on the visitor; it stops counting against the agent’s attention.
Overdue badge. A response-time target under Settings marks conversations without a first reply as overdue, which is the earliest signal that concurrency is too high for the current staffing.
Quick replies, Suggest, Translate and Polish. A slash inserts a quick reply, Suggest drafts an answer from your knowledge base, Translate turns a draft into the visitor’s language and Polish fixes grammar and tone. Each one shortens the typing part of a reply, which is the part that scales badly across parallel chats.
Operator status. Online, Away and Offline let an agent step out of rotation without closing chats. When the whole team is offline, visitors leave an e-mail before the first message and the reply goes out as an e-mail.

Reports show conversations per day, the share answered by Mio alone, first-reply time, busiest hours and team performance, which are the inputs for the monthly review.

Signs Your Concurrency Is Too High

First-reply time is fine but handle time keeps growing.
Ratings drop on afternoons when volume peaks, with no change in staff.
Repeat contacts rise: visitors come back because the first answer was rushed or wrong.
Agents paste the wrong canned reply into the wrong chat. This is the classic multitasking error and it is a concurrency signal, not a training problem.
Internal notes get shorter or disappear, so the next agent has no context.

Signs it is too low

Visitors wait while agents sit on one chat where the visitor has been idle for minutes.
Unassigned conversations pile up during peaks while every agent shows one open chat.
Agents report boredom in the quiet hours and panic in the peaks, which usually means the cap is too rigid rather than too low; let it rise during peaks for the simple types.

Concurrency Across Chat, Tickets and Messaging

The number only applies to live conversations. A ticket, which any Talkmio conversation can become, has no waiting visitor and does not count toward concurrency; the same goes for e-mail that arrives through the e-mail channel on Pro and above. Messages from Facebook, Instagram and WhatsApp on the Ultimate plan sit in between: the sender expects a reply within minutes, not seconds, so one agent can carry more of them than website chats. A sensible rule is to count a messaging conversation as half a chat when setting the cap. The distinction between the modes is covered in tickets vs live chat.

Concurrency and Quality Reviews

When you review conversations for quality, note the agent’s open-chat count at the time of each reply. A reply that reads as curt or off-target often turns out to have been written while four other chats were waiting, and the fix is the cap, not a coaching session. Sampling ten conversations per agent per month, split between quiet and peak hours, is enough to see whether quality holds at the concurrency you have set. If peak-hour samples score lower, lower the cap during peaks or add a second agent for those hours only.

Training Agents for Parallel Chats

Concurrency is a skill, and it improves with a few habits. Acknowledge every new chat within seconds with a real sentence, not a canned “an agent will be with you”, then return to the chat that needs the next step. Keep an internal note of where each conversation stands so switching back costs nothing. Set a conversation to Pending the moment it is waiting on the visitor. Never answer from memory of a different chat; re-read the last two messages before replying. And when a conversation turns into a complaint, drop to one: close the others properly or hand them to a colleague through assignment rather than leaving them hanging.

Frequently Asked Questions

What is a good live chat concurrency per agent?

Two to three for a mixed queue is the usual safe range. Simple order-status chats can run at four or five, technical troubleshooting at one or two, and complaints at one. Measure handle time by conversation type and test two against three for a week before setting a number; it should differ by agent and by the mix of questions.

Does an AI assistant change how many chats an agent can handle?

Yes, in two directions. Mio answers routine questions without a hand-off, so fewer chats reach agents and the peak is smaller. But the chats that remain are the harder ones, so the right concurrency per agent usually drops. Teams typically need fewer agents online with a lower per-agent cap.

Can I set a maximum number of chats per agent in Talkmio?

Talkmio has no hard per-agent cap setting. Teams enforce concurrency through assignment, the Mine and Unassigned filters, the Pending status for chats waiting on the visitor and the Away status when an agent is full. The overdue badge from the response-time target shows when the current staffing cannot keep up.

How do I calculate how many agents I need at the peak?

Multiply chats per hour at the peak by average handle time in hours to get concurrent conversations, then divide by your per-agent concurrency. For example, 45 chats an hour at 8 minutes each is 6 open conversations; at a concurrency of 3 that needs two agents. Subtract the share Mio answers alone first.

Why does handle time go up when concurrency goes up?

Each reply waits while the agent types to someone else, so every conversation stretches even if the agent is never idle. Past about three parallel chats the stretching usually cancels the throughput gain, and errors such as replying in the wrong chat increase. Watch handle time, ratings and repeat contacts together, not throughput alone.

Do WhatsApp and Messenger conversations count toward concurrency?

Count them at roughly half a website chat. Senders on messaging channels expect replies within minutes rather than seconds, so one agent can carry more of them. Tickets and forwarded e-mail do not count at all, because nobody is waiting live. Keep the cap for website chats, where the visitor is on the page right now.

The Bottom Line

Live chat concurrency is a number you should choose deliberately, per agent and per conversation type, and revisit monthly against first-reply time, handle time and ratings. Two to three parallel chats is the safe range for a mixed queue, and an AI assistant that answers routine questions lowers both the volume that reaches people and the cap those people should carry. Create a free Talkmio account, let Mio take the simple questions and see how many chats your team really needs to juggle.

Live Chat Concurrency: How Many Chats Per Agent at Once?