Build Guide: Intent and Sub-Topic Routing

Quick links in this article:

In this guide, you’ll build a Workflow that uses AI to automatically detect the reason for contact (Intent) and a more specific Sub-Topic, then route the interaction to the correct queue.

We’re using Email as the inbound channel in this example, but this setup works for most message-based channels.

What This Workflow Will Do

• Detect the Intent from the customer’s message (e.g., Tracking, Refund, Website Issues)

• Identify a Sub-Topic based on the detected Intent (e.g., Damaged item, Checkout Errors)

• Route the interaction to the appropriate queue

• Store both the Intent and Sub-Topic in custom data fields for reporting and agent context

Before You Start

Before building your flow, you’ll need to:

• Plan your intents and sub-topics, and identify queues

• Create two custom fields to store detected values

• Identify or create your trigger event

Let’s dive into each of those tasks below.

Plan your intents and sub-topics

Before you can start hooking things together in workflow, you need to decide how you want it to work!

Start by listing the common reasons your customers get in touch.
These are your Intents — things like:

• Parcel Tracking

• Website Issues

• Return and Refund

Then break each intent into a few Sub-Topics to capture more detail, and finally, identify the queue(s) you’ll want to route to for each combination. For example:

Create Your Custom Fields

You’ll need two new fields to store the results from the AI actions - be sure to add them to the data set you want to use so they’ll be visible on the interaction:

• Intent – stores the overall reason for contact
  Suggested values:

  • Tracking

  • Website

  • Returns

  • Unknown

  • etc.

• Sub-Topic – stores the more specific issue or context
  Suggested values:

  • Late

  • Chasing

  • Redelivery

  • Change Address

  • Discount Codes

  • Checkout Errors

  • Stock

  • Damaged

  • Wrong Item

  • Doesn’t Fit

  • Unknown

  • etc.

Open

Create your custom fields - and add them to a data set to show them on the interaction!

Identify a trigger event

You’ll attach your flow to a Workflow event later — either an existing New Message Received or Response Received event, or a new one.

Head to Settings > Workflow > Events to check which event applies to your use case — or create a new one for the right channel if needed.

Building the flow

With your planning, custom fields, and event ready, it’s time to build the flow itself. 

Create a new flow

Head to Configuration > Workflow > Builder and click Create Flow.

• Give your flow a clear name and label — for example: "AI Routing".

• Save early — and save often! (You can come back and edit the flow as you refine your logic.)

Once you’ve created your flow shell, you’re ready to start adding actions.

Open

Create a new flow, then give it a name and a label!

Create the interaction

If the interaction hasn’t been created yet in a previous flow (remember multiple flows can be attached to an event, and run consecutively from top to bottom!), you’ll need to start the flow by creating the interaction.

If that’s the case, at the very start of your flow:

1. Click the + button and add a Create Interaction action.

2. Set the Initial State to Queued

3. Select your default holding queue — this is where the interaction will temporarily sit before it's re-routed. Flows run near-instantly - so it’ll move swiftly on to the relevant queue!
   (e.g. Emails Queue or a general queue)

4. You don’t have to, but you could preset any data fields you like at this point - such as an Email field with the sender’s email address Echo.Person.Email.Address).

5. We’re also assigning the data group we want to display to agents here.

Open

Create the interaction in a queued state - we’ll update the queue dynamically later!

Extract Intent

Next, add the Extract Intent action to your flow. This is where the AI identifies the customer’s reason for contact — like Tracking, Refund, or Website Issue — and stores the result for use later.

You can leave the Message and Output Name fields as they are.

Now add each of your planned Intent categories one by one using the + Add button. These are the options the AI will choose from when trying to match the message to a reason.

Open

Add the Extract Intent action, and add in each of your primary reasons for contact

Split the Flow by Intent

Now that the AI is detecting intent, let’s build the logic to route each interaction based on the result.

Instead of building your Decision manually, use the built-in shortcut:

1. Go back to your Extract Intent action.

2. Click the Copy to Decision button at the bottom right of the action editor.

3. Open the Library panel and look for the new Decision it just generated.

4. Insert it into the flow directly after your Extract Intent action.

5. Rename the Decision to Split by Intent so it’s clear what this logic is doing.

Open

Copy your Intents to a Decision, then select it from the action library

Once inserted, the Decision will already contain branches for each of your defined Intents, plus a No Criteria Met fallback for when the Intent is Unknown.

Route to queues

For each branch in your Split by Intent decision:

1. Add an Update Interaction action.

2. Select Update queue and map each branch to the corresponding queue from your routing plan

3. Select Preset Data.

4. Find and select your Intent custom field.

5. Enter the matching Intent label as a static value (e.g. “Parcel Tracking” for the Parcel Tracking branch).

6. (Optional): Relabel the action, too, to keep your Flow easy to scan. i.e. Routing: Tracking

Repeat this step for each branch in the flow.

Open

Update the queue, and store the detected Intent in your custom data field on each branch

Once done, every interaction will be routing into the appropriate queue and will carry a clean, standardised record of why the customer got in touch.

Add Sub-Topic Detection

Next, you’ll use the Extract Sub-Topic action to add nuance to your routing and reporting — giving the AI specific clues to detect why, in more detail, the customer is getting in touch.

This action needs to be added beneath each Intent branch. Each instance should contain the Sub-Topics relevant to that Intent only.

Open

Add an Extract Sub-Topic for each branch, and add your sub-topic and prompt pairs

Here are suggested prompts for each of the example Sub-Topics used in this guide:

✨ Website Issues

• Discount codes → “If the customer is trying to apply a discount code and it isn’t working.”

• Checkout Errors → “If the customer is getting an error message when checking out.”

• Stock Issues → “If the item the customer wants is out of stock.”

📦 Parcel Tracking

• Late → “If the customer says their parcel hasn’t arrived on time.”

• Chasing → “If the customer is asking for an update on delivery progress.”

• Redelivery → “If the parcel was missed and they want it delivered again.”

• Change Address → “If the customer wants to change the delivery address.”

🔁 Return and Refund

• Damaged → “If the customer says the item arrived damaged.”

• Wrong Item → “If the customer got the wrong item or size.”

• Doesn’t Fit → “If the item doesn’t fit the customer or recipient.”

Add each Sub-Topic and its matching prompt into the Extract Sub-Topic action for that Intent.

Once completed, your AI will be able to detect not just the broad reason for contact — but the precise issue behind it.

Handle the No Criteria Met Branch

If the AI can’t confidently assign an Intent, the flow will land in the No Criteria Met branch of your Decision. What happens next is up to you — but there are two common approaches.

Option A: Route to a General Queue

If you want things quick and easy, simply add an Update Interaction action here that:

• Presets the Intent field as “Unknown” (or “Other”, if you prefer)

• Assigns the interaction to a generic “Catch-All” queue (like General Inbox)

This approach is fast, easy to maintain, and ensures the interaction still lands with an agent.

Option B: Add a Catch-All Sub-Topic Action

If you don’t mind your flow getting a bit more complex, then try this.

• Add an Extract Sub-Topic action to the No Criteria Met branch.

• Paste in all the Sub-Topic pairs from your other Intent branches — yes, all of them! — into a single action.

• Use the detected Sub-Topic to decide which Intent the customer probably meant, and route accordingly.

This effectively gives the AI a second shot at getting the routing right, using more specific language cues.

Open

Add a catch-all Sub-Topic action, and build a custom Decision to group back into Intent categories!

• Add a custom Decision to group Sub-Topic values back into Intent categories.

• In each branch, use KeywordGrouper.Output to check for your Sub-Topics.

• Group multiple Sub-Topics with OR conditions to reflect their shared parent Intent.

Open

Completed catch-all Decision

Once your Decision is set up, simply drag the matching branches to join the existing flow beneath the first Intent detection — that way, the interaction continues exactly as if Intent had been detected correctly the first time.

Open

Drag new branches back to match the original Intent branches

The new ‘No criteria met’ branch is where queries have failed to find Intent or a Sub-Topic. We’d recommend you add an Update Interaction action here that:

• Presets the Intent field as “Unknown” (or “Other”, if you prefer)

• Assigns the interaction to a generic “Catch-All” queue (like General Inbox)

Open

Handle the No Criteria Met branch by routing to a general queue

Store the Detected Sub-Topic

You’ve already got your Extract Sub-Topic actions in place — now it’s time to store the results for reporting, visibility, and further routing if needed.

You’ve got two options here:

Option A: Structured, Reportable Approach (Recommended)

1. Copy to Decision from each Sub-Topic action.

2. Insert that Decision directly under the Sub-Topic action to split the flow by Sub-Topic.

3. Under each new branch:

   • Add an Update Interaction action.

   • Use Preset Data to map the Sub-Topic value to your Sub-Topic custom field.

Repeat this process for each Intent branch — you’ll end up with a small Decision + Update combo under each.

This method gives you:

• Cleaner data

• Easier-to-read reports

• Consistent interaction tagging

Open

Split each branch by Sub-Topic, and update Sub-Topic data field to match appropriately

Your flow at this point, should look look a bit like this:

Open

Your completed AI Routing flow!

Option B: Raw Output Storage (Faster, Less Flexible)

Alternatively, you can skip that final Sub-Topic split and just store KeywordGrouper.Output directly in an open text field using a single Update Interaction action.

It’s quicker, but:

• The data will be harder to filter or group

• Agents won’t be able to easily correct/update it if needed

Final Step: Save, Publish and Attach to an Event

Once your flow is complete, it’s time to put it live.

✅ Save and Publish

• Click Save to store your flow

• Then click Publish to make it available for use

✅ Attach to a Trigger Event

• Go to Configuration > Workflow > Events

• Find the New Message Received event that matches your Email account (or create a new one)

• Attach your Intent & Sub-Topic Routing flow to it