How many chats can one agent handle at once? Set a limit from your data

How many chats can an agent handle at once? Many vendors offer one starting number, yet the same person copes differently with delivery questions and with a complaint that means checking three systems. A concurrent chat limit is a setting you derive from your own data and adjust over time.

This guide is about one person and their load at a given moment. How many people you need per hour is covered in our guide to customer service staffing, and shifts and cover in the article on SLA staffing for a small team. Here you set a baseline, spot overload and test whether to raise or lower the limit.

What a concurrent chat limit depends on

Parallel chats only work because customers do not type nonstop. While one person reads a reply, finds an order number or takes a product photo, the agent can answer someone else. When customers reply instantly and the topic needs focus, every extra chat makes everyone else wait longer.

↔ Scroll the table sideways to see all columns.

The table shows a direction, not a number. Do not adopt an outside norm such as “three to five chats”: those figures describe other teams, topics and tools. Start from a cautious baseline and measure what happens when you change it.

Measure a baseline on your own conversations

One week of normal traffic and a few dozen closed conversations are enough. Take the data from your conversation history and log it in a spreadsheet.

1. Group conversations by type.Pick the three or four most common topics, such as delivery, product choice and complaints.
2. Measure the customer’s pauses.For a few chats of each type, note how long the customer usually takes to reply. That is time the agent can give to another chat.
3. Measure the agent’s work.Note how long one reply takes, including checking data in other systems.
4. Calculate the baseline.Divide the typical customer pause by the time per reply and add one. Round down, and go lower for topics that need focus.

Take a fictional accessories shop: a customer asking about delivery usually replies after two minutes and a reply takes the agent one minute, so the calculation gives three chats. For complaints the agent spends three minutes on each reply and the customer answers quickly, so the result drops to one. That is a starting point for a test, not a norm. Ways to cut the work on each reply are covered in our guide to chat handling time.

A 2018 study by Goes, Ilk, Lin and Zhao, based on logs, transcripts and surveys from a large service firm’s chat, found that agent multitasking led to longer in-service delays and lower problem resolution, which reduced customer satisfaction. Management Science, RePEc record. The study does not set a safe number of chats for your team.

Website chat versus Messenger and Instagram

Website chat is usually synchronous: the customer sits at the window and sees that nobody is replying. Messenger and Instagram messages more often arrive with gaps, because people write between other tasks and come back an hour later. One person can keep more of these threads open, as long as they do not compete for attention with a live website chat.

The Messenger Platform policy states that businesses have up to 24 hours to respond to a user in the standard messaging window. Meta for Developers. That rule frames when you can send, not how fast your customer expects an answer.

In Meetlead, the optional workload limit counts open conversations assigned to a person across all channels. An Instagram thread waiting for the customer therefore takes a slot just like an active chat. Close resolved cases straight away and consider auto-closing conversations that are waiting for the customer. Also note that Meetlead sends Messenger replies only within the standard 24-hour window.

Signs the limit is too high

Overload rarely shows in the chat count. It shows in how long customers wait for the next reply and how many sent messages need fixing. Review these signals per person, because a team average hides one overloaded agent.

• Next-reply time rises with the number of chats one person has open, even though topics are similar.
• A reply lands in the wrong thread or contains another customer’s details.
• A customer repeats a question because the last reply skipped part of it.
• More customers come back about a case that was marked as done.

The first signal matters most because you can measure it. In Meetlead, the first and next response report shows the median and p90 per agent, counting only the periods when the conversation was assigned to that person. How to read median and p90 is explained in our guide to first response time in chat.

What happens when the limit is reached

A limit only makes sense if you know where the fourth chat goes when the limit is three. In Meetlead, an admin can set an optional limit of open conversations per agent for a website. Automatic assignment skips people who have reached it, and if nobody has a free slot, the chat stays in the department as unassigned. Claiming a chat above the limit is blocked; an admin can override it, and the override is recorded.

Note that assignment is attempted when the chat opens and does not repeat on its own when someone frees up. Someone has to watch unassigned chats sorted by longest wait, and an optional alert after a set number of minutes helps. Which case to take first is covered in our guide to queue priorities, and what the customer sees meanwhile in the article on waiting messages in live chat.

Round robin or workload-based assignment

A Meetlead department assigns chats either in turn among available members or by workload, meaning to the person with the fewest open conversations. Round robin is predictable and works when cases are similar and take about as long. With uneven cases, it may give one more chat to someone stuck on three hard ones.

Workload-based assignment suits a team where some chats end in a minute and others take a quarter of an hour. It counts open chats, not difficulty: three complaints weigh the same as three delivery questions. Route hard topics to a separate department staffed by people who know them. Setting up departments and reply owners is covered in our guide to assigning customer requests, and a single owner also prevents two people replying to the same customer.

A two-week limit test

Change the limit in a controlled test: one bad shift proves nothing.

1. Record a baseline.Before the change, note each person’s median and p90 next-reply time, wrong-thread mistakes and how long unassigned chats wait.
2. Change one thing.Raise or lower the limit by one chat. Do not change the rota, departments or saved replies at the same time.
3. Review after a week.Compare the data with the baseline and read a few chats with the longest waits.
4. Decide.After two weeks, keep the new value, go back to the old one or plan another test.

Write the criteria down before you start. Keep the higher value, or raise it in the next test, if next-reply p90 did not rise noticeably, there were no wrong-thread replies and unassigned chats waited less. Lower it if next-reply time grows, even one reply landed in the wrong thread or customers repeat questions more often. If the queue grows while reply quality stays good, you are short of people, not of slots. Departments, alerts and the limit are set up in Meetlead customer support.

Questions about concurrent chat limits

Should a new agent handle fewer chats at once? Yes, in the first weeks they need a lower threshold than the rest of the team. In Meetlead the workload limit is one value for the whole website, so at first you can leave the new person out of departments with automatic assignment and pass chats to them manually.

Will AI let agents handle more chats at once? It can shorten reply preparation, for example through Copilot suggestions that the agent reviews before sending. Less work per reply raises the baseline calculation, but it does not replace the test.