Skip to content
Chatbotscape
Editorial flat-vector illustration for How to Route Live Chats to the Right Agent
Published 7 min read

How to Route Live Chats to the Right Agent

Quick answer: Write down which team owns which kind of question, then route on inputs you reliably have: the page, a topic from the pre-chat form or the chatbot, and the visitor's language. Put specific rules above general ones and end with a fallback team. Inside each team, use an assignment method that respects a per-agent chat limit, decide what happens when nobody is available, and track transfers and wait times so you can fix the rule that misfires.

Step 1: Map questions to teams before touching settings

Make a short table: each type of question, the team that answers it, and who is in that team. Two or three teams are plenty for most small businesses.

Question typeTeamPeopleNotes
Before buying: pricing, fit, demosSales1-2Fast reply matters most
Orders, accounts, how-toSupport2-3Most of the volume
Billing and refundsSupport, billing skill1Rare, needs access
Anything unclearSupport (fallback)AllCatches what rules miss

If one person sits in every team, routing will not create capacity; it only sorts the queue. That is fine, but set expectations accordingly.

Step 2: Choose routing inputs you actually have

A rule is only as good as its input. Common inputs, from most to least reliable:

  • Page or domain. Chats from the pricing page go to sales, chats from the help center to support. LiveChat's routing rules match on URL (contains or exact) and domain.
  • Topic chosen by the visitor. A short choice list in the pre-chat form, written in the visitor's words. See designing a pre-chat form. Remember that proactive and manually invited chats can skip the form.
  • Topic from the chatbot. The bot asks or infers the need, answers simple questions and passes the rest; see the handoff design guide.
  • Language. From the visitor's choice, the bot or automatic detection; chatbot language detection explains why browser settings and message text can disagree.
  • Country or customer tier. LiveChat documents location-based routing on its Business and Enterprise plans. VIP routing needs the platform to know who the visitor is, usually from a logged-in session or CRM data.

Pick two inputs to start. Every extra input is another rule to maintain.

Step 3: Order the rules and add a fallback

Rules are usually checked in list order, so order decides which rule applies when several could. LiveChat's article says to put more specific rules before general ones, and it ends the list with a fallback group that catches chats no rule matched; by default that is the group containing all agents. Do the same wherever you set it up:

  1. Specific: billing topic → support with the billing skill.
  2. Page: pricing page → sales.
  3. Language: Spanish → Spanish-speaking agents.
  4. Fallback: everything else → support.

Test each rule by starting a chat yourself from the right page with the right answer, and check which team receives it.

Step 4: Pick the assignment method and set chat limits

Once a chat reaches a team, the platform picks a person. The main choices:

  • Round robin takes turns. Intercom's round robin skips teammates set to Away by default but does not consider how many conversations each person already has. Intercom recommends it for sales teams, so leads are spread evenly.
  • Capacity-based gives the chat to whoever has the most room. Zendesk's default for omnichannel routing is highest spare capacity; Intercom's balanced assignment, which it documents for the Expert plan, gives the next conversation to the eligible teammate with the fewest open ones; HubSpot chatflows send chats to the least busy available user. This is usually the better fit for support.
  • Agents pick from a queue. LiveChat's manual selection mode notifies accepting agents and lets one take the chat. Useful when chats vary a lot, slower to start.

Then set a concurrent chat limit per agent. LiveChat stops auto-assigning once every accepting agent has reached their limit. New chats then wait in a queue; what visitors see while they wait is covered in managing a live chat queue. Start low enough that people can answer well, and raise it only if waits are long while agents say they have room.

Skills narrow the pool further. Zendesk lists skills-based routing for Professional and above and offers a skills timeout that lets a chat go to an agent without the skill after a set wait, which helps when only one person has it.

Step 5: Plan for away agents, timeouts and nobody available

Decide what happens in each of these cases before visitors find out:

  • Assigned agent does not respond. LiveChat moves the chat to another agent after an inactivity timeout, fixed at 3 minutes on its Starter plan and adjustable on Team and Business.
  • Agent goes away mid-shift. Intercom's round robin skips Away teammates by default, and its article notes that conversations assigned while everyone was away are not assigned automatically when someone returns. Someone has to check that inbox.
  • Nobody is available. HubSpot leaves the chat unassigned for the team to triage. Pair this with an after-hours chatbot that sets a clear reply expectation and collects an email.
  • Wrong team. Agents need a quick transfer with context; see warm transfer and summarizing conversations for handoff.

Step 6: Measure transfers, waits and missed chats

Track three numbers weekly: the share of chats transferred to another team after they start, the time chats wait before an agent accepts them, and chats that ended with no agent. Transfers point at rules or unclear topic options. Long waits while agents are idle point at limits, status settings or a skill held by one person. Missed chats point at Step 5. See chatbot metrics and chatbot first response time.

Check the arithmetic

An invented example, not a measurement. A shop has four support agents with a limit of 3 chats each, so the team can hold 4 × 3 = 12 chats at once. Two of them speak Spanish, so Spanish chats have 2 × 3 = 6 slots, half the team's capacity. In one week, 18 of 120 chats were transferred to another team: 18 ÷ 120 = 15%. Most were billing questions arriving through the generic "Help with my order" option. The shop adds a "Billing or refund" option to the pre-chat form and a rule for it. The next week, 6 of 125 chats were transferred: 6 ÷ 125 = 4.8%. Two weeks with different traffic do not prove cause, so read the remaining transfers before changing more. These numbers are made up to show the method.

Common mistakes

  • Routing on inputs you do not collect, such as a topic field that proactive chats skip.
  • No fallback, so unmatched chats sit nowhere.
  • Round robin for support without checking how many chats each person already has.
  • Limits set once and never revisited.
  • A skill held by one person with no timeout or backup.
  • Internal team names in the topic list visitors see.

FAQ

How do I route live chats to a specific department? Create a team or group per department, then add rules that send chats there by page, pre-chat topic or bot answer, with a fallback team for everything else.

Is round robin or skills-based routing better for a small team? They answer different questions. Skills decide who is eligible; round robin or capacity-based assignment decides who among them gets the chat. For support, a capacity-aware method usually works better than plain round robin.

How many chats should one agent handle at once? No vendor page read for this guide gives a recommended number. Start low, watch quality and wait times, and adjust.

How do I route chats by language? Give each language its own team or skill and route on the visitor's choice, the bot's question or detected language. Check how detection works on your platform.

What happens when no agent is available? It depends on the platform: a queue, a fallback group or an unassigned inbox. Plan a bot message and a follow-up path for those chats.

Can my chatbot do the routing? It can collect the topic and language and answer the simple questions; the platform's rules then pick the team and agent.

Sources