Design chat routing rules that send each conversation to the right bot, person or shift. Covers website chat, WhatsApp and Messenger, plus a setup guide.
Most chat teams don't have a routing problem until they have a missed-chat problem. A WhatsApp message sits unanswered overnight, or the wrong person gets pinged on a public holiday, or a billing question lands with someone who can't see invoices. This article explains how to design routing rules that match your team, and how to set up the parts of that design in AI SNS CHAT.
What a routing rule actually decides
A routing rule answers three questions about every conversation:
- Who answers first? The AI agent, a human, or a specific team.
- When does it move? The trigger that hands a chat from one party to another.
- Where does it go? A named person, whoever is on shift, or a shared queue.
Many teams only define the second question ("escalate when the customer is angry") and leave the other two to chance. That is how chats end up in a shared channel where everyone assumes someone else has it.
Write your rules as plain sentences before you touch any settings. "If a customer asks for a refund on WhatsApp, the AI collects the order number, then hands off to whoever is on the support shift." If you can't say it in one sentence, the rule is probably two rules.
The four inputs you can route on
Good routing uses information you already have at the moment a chat arrives or changes state.
Channel
Website chat, WhatsApp and Messenger behave differently. Website visitors expect an answer within a minute and leave if they don't get one. WhatsApp customers tolerate a gap but expect the thread to still be there tomorrow. Messenger sits in between. Your rule can reflect that, for example by letting the AI handle first contact on all three but giving WhatsApp threads a longer human response window.
Topic or intent
Order status, billing, technical faults, sales questions and complaints usually need different people. Topic is the most useful routing input and also the hardest to keep clean, because customers rarely label their own messages. The AI agent is good at detecting intent from free text, which is why it is a sensible first stop.
Time and availability
Time zone matters more than most teams expect. If your customers are in Sydney and your agents are in London, "business hours" means something different to each side. A rule that depends on who is actually working right now will outperform one that depends on a static list of names.
Customer signal
Sentiment, repeat contact and explicit requests ("I want to talk to a person") are all signals. The explicit request should always win. Making customers argue with a bot to reach a human costs more goodwill than it saves in labour.
Common routing models
| Model | How it works | Best for | Weak spot |
|---|---|---|---|
| Fixed owner | One named person receives every handoff | Solo operators, very small shops | Single point of failure |
| Shift-based | Handoff goes to whoever is on duty now | Teams across time zones or split hours | Needs an accurate rota |
| Team by topic | Billing, tech and sales each have owners | Mid-sized teams with distinct skills | Needs reliable topic detection |
| Separate agents by brand or language | Each brand or market has its own AI agent and its own people | Multi-brand businesses | More agents to maintain |
Most teams end up combining two of these. A common pairing is separate agents per brand, with shift-based handoff inside each one.
- 01Customer message arrives
- 02AI agent answers or collects details
- 03Trigger fires (request, topic, sentiment)
- 04Handoff to person on shift
Decide when the AI hands off
Handoff triggers are where routing rules succeed or fail. Keep the list short enough that your team can remember it.
Reasonable triggers include:
- The customer asks for a person, in any wording.
- The question is outside what the knowledge base covers and the AI has no confident answer.
- The topic involves money movement, account changes or complaints.
- The same customer has asked the same thing twice without getting a useful answer.
Avoid triggers that fire on single words. A customer writing "this is not a problem" contains the word "problem" and is not asking for escalation.
Make handoffs honest
A routing rule that sends a chat to a human is only half the job. The customer needs to know what happens next, and the line you give them must be true.
Do not promise "an agent will reply in two minutes" unless your rota can guarantee it. A better message states the real situation: someone is on shift and will pick it up, or the team is offline and the customer will hear back when it reopens. Customers forgive a clear wait far more readily than a broken promise.
Also decide how your team learns about the handoff. A chat that is routed correctly but announced nowhere is still a missed chat. Alerts should reach people where they already work, whether that is Slack, Teams, email or the messaging app they use on their phone.
Measure whether the rules work
Routing is never finished. Review it monthly using a small set of questions:
- How many handoffs arrived outside staffed hours?
- How many chats were reassigned after the first person looked at them? A high number means the rule is sending work to the wrong place.
- How many handoffs were unnecessary, meaning the AI could have answered?
- Which topics produce the most handoffs? These are candidates for new knowledge-base content.
Here is a purely illustrative example of how a month of handoffs might look for a small shop, to show what to look for. These are not real figures.
In this made-up month, the "outside knowledge base" group is the quickest win. Adding the missing answers removes those handoffs entirely, and the shift rota only has to cover the rest.
Doing this in AI SNS CHAT
AI SNS CHAT builds handoff routing around shifts, so the person who receives a handoff is the person actually working at that moment.
Set the rota. Go to Helpdesk → Scheduling. Here you set shifts per person and per time zone. Handoffs are auto-assigned to whoever is on shift. Helpdesk requires the Standard plan or higher.
Choose where alerts go. Handoff alerts can be sent to Slack, Microsoft Teams, LINE WORKS, WhatsApp, email or a webhook. Pick the tool your team keeps open all day rather than a new one they must remember to check.
Let the system word the next step. When a chat is handed off, the server reads the rota and adds the "what happens next" line to the customer message itself. It never invents a wait time, so the message reflects who is actually scheduled.
Add an escalation trigger. Under Actions, the built-in Escalations action lets you define when the agent passes a conversation to a person. Actions are available from the Hobby plan, and the number per agent depends on your plan.
Use live-chat handoff targets where you need them. If your team already works in Sunshine or Salesforce, those are supported handoff targets, as is AI SNS live chat.
Work the conversation. Conversations from website chat, WhatsApp, Messenger and other chat channels appear in Activity → Chat logs. The Inbox is for email only. In either place, a person can take over a chat, which pauses the bot. They can also use live translation, saved replies, and edit a reply into the ideal answer so the agent learns from it. In Chat logs you can filter, flag, escalate and resolve conversations, in bulk if needed.
Split by brand or market. If you need different rules for different audiences, create separate AI agents and deploy each to its own channels under Channels. The number of agents depends on your plan, from 1 on Free up to 10 on Corporate, and extra agents can be added for $10 per agent per month.
Control who can see what. Roles (Owner, Co-owner, Member, Support, Viewer) with per-member permission overrides let you keep billing conversations away from people who shouldn't see them.
Review. Use the Chats, Topics and Sentiment views under Analytics to see which topics cause handoffs. Advanced analytics need the Pro plan.
You can compare plans at pricing and see the full feature list at features. Channel setup is covered at connections.
Conclusion
Good routing rules are short, written in plain language and tied to real availability. Start with who answers first, define a handful of honest triggers, put a rota behind the handoff and review the results monthly. Tools can carry the rules, but the thinking has to come first.
Frequently asked questions
Should the AI or a human answer first on WhatsApp?
For most businesses the AI should answer first, because it responds instantly and can gather details such as an order number. The rule that matters is that a customer who asks for a person gets one. Keep the handoff trigger simple and always honour a direct request.
Does AI SNS CHAT route chats by keyword or topic automatically?
Handoff routing is built around shifts: a handoff goes to whoever is on shift in Helpdesk → Scheduling. If you need topic-based separation, the practical approach is separate AI agents per brand or market, plus escalation triggers in the agent's Escalations action and guardrails. Check the features page for the current list of capabilities.
Where do I find conversations from Messenger and website chat?
They appear in Activity → Chat logs, along with all other non-email channels. The Inbox is email-only. Both let a person take over a chat, which pauses the bot.
What does the customer see when a chat is handed off?
The server reads your rota and adds the "what happens next" line for you. It never invents a wait time, so the message reflects who is scheduled rather than an optimistic guess.
Which plan do I need for shift-based handoff?
The Helpdesk section, which includes Scheduling, needs Standard or higher. Integrations such as escalation Actions need Hobby or higher.


