How do I route different customer requests?

Use Router when one AI agent needs to handle several different types of customer requests.

Router identifies the customer's intent and sends the conversation to the appropriate path.

For example:

Pricing question
→ Data Search

Policy question
→ Document Search

Appointment request
→ AI Answer with Scheduling enabled

Speak to someone
→ Human Handoff

When do I need Router?

Router is useful when the AI agent has several distinct jobs.

For example, a customer-facing agent may need to handle:

• General questions
• Product search
• Pricing
• Booking
• Support
• Human handoff

Without routing, all of these requests may end up in the same part of the workflow.

When might I not need Router?

You may not need Router when the AI agent has one simple job.

For example:

Start → AI Answer

may be enough for an agent that only answers general service questions.

Do not add Router just because it is available.

Use it when different requests genuinely need different paths.

Route by intent

An intent represents what the customer is trying to accomplish.

Examples include:

• Book appointment
• Ask about pricing
• Find product
• Ask policy question
• Request support
• Speak with person

The customer does not need to use exact wording.

For example, all of these may express the same booking intent:

“I want to book.”

“Can I come in Friday?”

“Do you have anything tomorrow?”

“I need an appointment.”

Keep intents distinct

Good intents should represent meaningfully different jobs.

For example:

• Book appointment
• Product search
• Support issue
• Human handoff

Avoid creating several intents that mean nearly the same thing.

For example:

• Book
• Schedule
• Appointment
• Reserve a time

These may be better represented as one booking intent.

Route to the node best suited for the job

Once an intent is identified, send it to the right destination.

For example:

Written knowledge

→ Document Search

Structured records

→ Data Search

Conversational guidance

→ AI Answer

Structured intake

→ AI Form

Appointment request

→ AI Answer with Scheduling enabled

Human request

→ Human Handoff

Example: service business

A service business could use:

Start → Router

Then:

Service question
→ AI Answer

Price estimate
→ AI Answer with Calculations enabled

Book service
→ AI Answer with Scheduling enabled

Speak with team
→ Human Handoff

Example: hotel

A hotel AI agent could route:

Hotel policy
→ Document Search

Room or package question
→ Data Search or AI Answer

Current local information
→ AI Answer with Live Web Search enabled

Book an appointment or service
→ AI Answer with Scheduling enabled

Human assistance
→ Human Handoff

Example: e-commerce

An e-commerce agent could route:

Find product
→ Data Search

Compare products
→ AI Answer

Show products
→ AI Answer with Product Cards enabled

Returns policy
→ Document Search

Talk to sales
→ Human Handoff

Do not route based on keywords alone

Customers can express the same intent in many ways.

A good routing setup should focus on meaning.

For example:

“Can someone call me?”

may be a human-handoff request even though the customer did not say:

“Human agent.”

Test different natural phrases for every route.

Add a fallback path

Not every customer request will match a route perfectly.

Plan what should happen when the intent is unclear.

The fallback might:

• Send the request to AI Answer
• Ask a clarifying question
• Use Human Handoff
• Provide a general support path

Choose the fallback that makes sense for the agent's role.

Avoid too many routes

A Router with too many overlapping intents becomes harder to test and maintain.

Start with the main jobs.

Add new routes only when:

• The destination is genuinely different
• The business process is different
• The knowledge source is different
• The action is different

Test each route with several phrases

For every intent, test more than one example.

For booking:

• “I want to book.”
• “Any openings Friday?”
• “Can I schedule something?”
• “Do you have time tomorrow?”

For human handoff:

• “Can I speak with someone?”
• “I need a person.”
• “This is not helping.”
• “Can someone from your team contact me?”

Test overlapping requests

Customers may ask several things at once.

For example:

“How much is it and can I book Friday?”

Test whether the workflow handles the message appropriately.

Depending on your design, the agent may:

• Answer both
• Handle one and continue to the other
• Route to a path that can manage both

Test ambiguous requests

Try short messages such as:

“Help”

“Price”

“Booking”

“I have a problem”

The workflow should either route correctly or ask for clarification.

Review routing when the workflow changes

Update Router when you:

• Add a new service
• Add a new skill
• Add a new workflow path
• Change handoff behavior
• Add a new data source
• Remove a previous capability

Old routing rules can send customers to the wrong destination after the workflow changes.

Tips for routing customer requests

• Use Router only when the agent has several real jobs.
• Define intents around customer goals.
• Keep similar intents together.
• Send each intent to the node best suited for the task.
• Include a fallback path.
• Test natural customer wording.
• Test ambiguous and multi-intent messages.
• Avoid creating too many overlapping routes.
• Review routing after major workflow changes.

Related guides

For more information, see:

• Which AskHandle node should I use?
• How do I plan an AI agent?
• How should I test my AI agent before publishing?
• How does Human Handoff work?