Overview

Fibi AI Attributes let Fibi understand and classify a conversation using your existing conversation attributes. Fibi applies structured values like issue type, sentiment, urgency, or product area, so your team can route conversations faster, trigger the right actions, and analyze support trends without manual tagging.

Key benefits

• Custom attributes that reflect your business: Train Fibi to detect attributes like issue type, urgency, sentiment, or spam status.

• Adaptive detection: Fibi continuously evaluates context and updates values as the conversation evolves.

• Smarter routing: Use detected values in Workflows to branch, route, assign, or trigger follow-up actions

• Reporting-ready structure: Detected values are saved as conversation attributes, so you can use them in filters, views, and reporting

How Fibi applies attributes

By default, Fibi classifies enabled attributes when Fibi's involvement in a conversation ends, such as when:

• Fibi hands the conversation over to a teammate

• The conversation is resolved by Fibi

• The conversation is closed after inactivity

If an attribute is used in live Workflow branch conditions, Fibi can also re-check that attribute during the conversation so your Workflow can react to the latest context.

If ‘Detect on close’ is enabled, Fibi re-checks the value again when a teammate closes the conversation.

Fibi does not overwrite values set by a teammate or a Workflow. If Fibi previously detected a value, it can update that AI-detected value later as the conversation changes.

How to set up Fibi Attributes

1. Decide what to classify

Start by deciding what structured information you want Fibi to detect from support conversations.

Good Fibi Attributes are useful for routing, filtering, reporting, or follow-up automation. Common examples include:

• 💡Issue type – Billing, Bug report, Feature request, Account access

• 😃 Sentiment – Positive, Neutral, Negative

• 🚨 Urgency – Urgent, High, Normal, Low

• ⚠️ Spam detection – Spam, Legitimate

• 📦 Product area – Inbox, Workflows, Billing, Integrations

Before creating the attribute, make sure each value is distinct enough that a teammate could confidently choose between them. If the values overlap too much, Fibi will have a harder time classifying conversations accurately.

2. Create a Fibi AI Attribute

Fibi AI Attributes are created from conversation attributes. To create one:

1. Go to Settings → Conversations → Attributes

2. Click New attribute (or edit an existing one)

3. Set field type to “Select”

4. Check “Let Fibi detect”

5. Add the possible attribute values as Options

6. Add clear descriptions for each value

7. Click Create attribute

Only select-type conversation attributes can be detected by Fibi. Other field types can still be used as regular conversation attributes, but Fibi will not automatically classify them.

3. Configure detection rules (optional)

Conditions

Use Conditions to detect an attribute only when another AI-detected attribute has a specific value. This creates a parent and dependent relationship. Fibi will only attempt to detect the dependent attribute after it has first identified the parent attribute and its value.

For example:

• If Type of query is Billing, detect Billing reason

• If Sentiment is Negative, detect Urgency

• If Product area is Integrations, detect Integration type

The controlling attribute must also have Let Fibi detect enabled. A conditional attribute can depend on one AI-detected parent attribute, and that parent attribute cannot itself be conditional.

Detect on close

Enable Detect on close when you want Fibi to re-check the attribute when the conversation is closed. This is useful when the customer shares important context later in the conversation, or when the final outcome is clearer at the end than it was during the first reply.

4. Preview detection

After creating the attribute, reopen it to test detection against sample messages.

Use Preview to:

• Test example customer messages

• Check which value Fibi would choose

• Review Fibi's reasoning for the test result

• Refine your attribute name, description, and option descriptions before using it on real conversations

If the preview does not return the value you expected, update the option names or descriptions to make the difference between values clearer.

Use Fibi Attributes across Featurebase

Fibi AI Attributes are saved as conversation attribute values. Once detected, they can be used in the same places as other conversation attributes: Use Fibi AI Attributes in Workflows & Inbox

You can use detected values to:

• Build Workflow branches based on what Fibi detected

• Route conversations to the right team

• Create filtered Inbox views

• Add conversation attribute columns to your Inbox

• Segment reporting by issue type, urgency, sentiment, or other attributes

For example, you could create a Type of query attribute with values like Billing, Bug report, and Feature request. Then, in a Workflow, branch conversations where Type of query is Billing to your Billing team.

Fibi Attribute examples

Below are examples of attributes and values you could use to classify support conversations with Fibi.

Issue type

Use an Issue type attribute to understand what the customer is asking about.

• Projects – Conversations about managing projects, tasks, milestones, collaboration, status, deadlines, or other work tied to a specific project

• Billing – Conversations about subscriptions, invoices, payment methods, plan changes, discounts, trials, account limits, refunds, or other payment-related questions

• Account management – Conversations about creating, deleting, updating, transferring, or managing a customer account

Sentiment

Use a Sentiment attribute to understand how the customer feels in the conversation.

• Positive – The customer sounds happy, satisfied, appreciative, or generally positive about the experience

• Negative – The customer sounds frustrated, unhappy, dissatisfied, blocked, or generally negative about the experience

• Neutral – The customer is asking a question or describing a situation without a clear positive or negative emotion

Spam detection

Use a Spam detection attribute to separate real support conversations from messages your team can ignore.

• Spam – Automated or irrelevant messages, such as auto-replies, newsletters, sales pitches, guest post requests, or unrelated promotional content

• Legitimate – Real customer conversations where the sender has a support question, product issue, billing question, or another request your team should handle

Best practices

Clear attribute values make detection more accurate. Treat your attribute setup like a taxonomy that Fibi needs to understand from text alone.

Use short, descriptive value names

Keep values easy to scan and understand. A good name immediately signals the attribute's purpose.

• Keep names under approximately 25 characters.

• Use short, descriptive language (e.g., "Login Issues", "Billing Questions").

• Avoid internal jargon, ticket IDs, or ambiguous terms.

Write detailed option descriptions

A detailed description is crucial for helping Fibi understand the nuances of each attribute value. For each option, explain:

• What the value means: describe what the value represents and when Fibi should choose it.

• When Fibi should choose it: include keywords or phrases the customer might use.

• Common words or phrases customers might use (e.g., "How do I reset my password?").

• What should not be included in that value

Keep values distinct

Avoid values that overlap too much. If a teammate would struggle to choose between two values, Fibi will likely struggle too.

Before enabling detection, review the full list and ask yourself:

• "Would a human find it hard to choose between these attributes?"

• "Would a customer message clearly belong to one value over the others?"

Avoid creating redundant or overly narrow attributes that could be combined.

Add an Other value when needed

For broad attributes, we recommend including a general-purpose Other or Other / Uncategorized value so Fibi has a safe fallback when none of the specific values apply.

This works well for attributes like Issue type, Product area, or Reason for contact.

For attributes where the values are already exhaustive, such as Sentiment with Positive, Neutral, and Negative, an Other value is usually not needed.

Examples of good attribute value descriptions

Example 1: Account access

Account access

Use this value when the customer cannot get into their account, is blocked during sign-in, forgot their password, is locked out, or cannot complete verification.

Applies if the customer:

• Says they cannot log in or access their account

• Mentions password reset, sign-in errors, SSO, 2FA, or email verification

• Says they are locked out, suspended, or unable to complete authentication

Does not apply if the customer:

• Is asking about billing, invoices, refunds, or subscription changes

• Wants to delete, merge, or transfer an account

• Is reporting a product bug after they have already logged in

Likely keywords: login, password, locked out, sign in, SSO, 2FA, verification email

Example 2: Refund request

Refund request

Use this value when the customer is asking for money back, disputing a charge, or asking about the status of a refund.

Applies if the customer:

• Directly asks for a refund or says they want their money back

• Mentions being charged twice, charged incorrectly, or overcharged

• Asks when a refund will arrive or why a refund has not been processed

Does not apply if the customer:

• Reports a failed payment or declined card without asking for a refund

• Asks about changing plans, updating billing details, or downloading invoices

• Mentions cancellation but does not mention a charge, refund, or payment issue

Likely keywords: refund, money back, charged twice, overcharged, dispute, refund status

FAQs