Overview

AI Actions allows you to embed AI models directly into Zingtree workflows either conversationally or programmatically via REST API.

AI Actions can be used to:

• Generate AI-powered decisions

• Summarize, validate, enrich, and transform data

• Drive workflow business logic

• Enhance end-user scripting and context

AI Actions supports two implementation approaches:

• Conversational AI Actions

• API (REST) AI Actions

Both approaches use the same underlying AI infrastructure. The difference is how and where the output is consumed within the workflow.

Accessing AI Actions

Navigate to Apps & Integrations & select AI Actions from the left navigation menu

 

Architecture

AI Actions combines the following components:

• Context → Data passed into the AI model

• System Settings → Defines model role and constraints

• User Settings → Defines the task or query

• Model Configuration → Controls determinism and output limits

The AI model processes the provided context according to the defined system and user instructions, then returns either text or JSON depending on configuration.

 

API (REST) AI Actions

API mode allows AI output to be consumed programmatically within workflows or external integrations.

Behavior

• Receives structured context data

• Applies system & user defined instructions

• Returns structured JSON

• JSON output can be referenced for branching, transformation and scripting

Output Type

• Structured JSON (recommended)

• Text (supported but not recommended for workflow logic)

Configuration (API Mode)

1. Create or edit an AI Action.

2. Toggle API Mode.

3. Set Output Format = JSON.

4. Configure model parameters.

5. Define expected JSON schema in your prompt.

6. Save.

 

Access AI Actions API response

AI Actions outputs will be placed on the actions namespace. Use the alias found in the AI Actions node configuration inside the workflow builder

 

 

Configuration Settings

The following settings impact output behavior:

System vs User Settings

System Settings

Defines the model’s role, constraints, and behavior.

Example:

User Settings

Defines the task to perform on the provided context.

Example:

Conversational AI Actions

Conversational mode inserts an AI interaction point within a workflow node.

Behavior

• Triggered when the workflow reaches the AI Actions node

• Receives configured context

• Returns text output

• Output is displayed to the end user

Output Type

• Plain text

Configuration (Conversational Mode)

1. Create or edit an AI Action.

2. Toggle Conversational Mode.

3. Set Output Format = Text.

4. Configure model parameters.

5. Define:

   • Context

   • System Settings

   • User Settings

6. Save.

Workflow Usage (Conversational Mode)

1. Add a new node to your workflow where an end user would interact with data

2. Select the AI Actions node type

   1. Add any additional context in the text inpt

3. Select the preferred AI Actions from the dropdown list

4. Preview the workflow to test the AI Actions Node

 

Common Use Cases

The following examples illustrate how AI Actions can be applied within Zingtree workflows.

Risk Evaluation

AI evaluates customer, transaction, or account data to determine risk level.

Example applications:

• Credit risk assessment

• Fraud likelihood scoring

• Compliance risk review

• Escalation prioritization

The AI analyzes structured data and returns a risk classification that can influence workflow progression.

Conversational Case Review

AI analyzes case history and provides a structured summary for human review.

Example applications:

• Support ticket analysis

• Claims review

• Incident summaries

• Account review preparation

The AI helps consolidate complex information so the user can make informed decisions before proceeding in the workflow.

Call Log Summaries for Agents

AI summarizes customer call transcripts or interaction logs.

Example applications:

• Post-call summaries

• Key issue extraction

• Action item identification

• Sentiment tagging

This reduces manual note-taking and standardizes documentation across agents.

Upsell Opportunity Identification

AI analyzes customer usage patterns and account data to identify potential expansion opportunities.

Example applications:

• Plan upgrade recommendations

• Feature adoption analysis

• Renewal opportunity detection

• Cross-sell identification

The AI evaluates behavioral or transactional signals and surfaces accounts that may warrant outreach.

Data Validation

AI evaluates structured data submissions for completeness, consistency, or anomalies.

Example applications:

• Form submission validation

• Data integrity checks

• Logical consistency review

• Compliance screening

This can be used to detect potential issues before allowing a workflow to proceed.

Data Enrichment

AI enhances existing records by adding structured metadata or classifications.

Example applications:

• Categorization of requests

• Sentiment classification

• Topic tagging

• Priority assignment

Enriched data can be used for routing, reporting, or downstream automation.

Decision Support

AI provides structured recommendations based on provided context.

Example applications:

• Recommended next best action

• Escalation recommendations

• Resolution suggestions

• Workflow routing guidance

This supports both automated decisions and human-assisted workflows.

Summary

AI Actions provides two operational modes for integrating AI into Zingtree workflows:

• Conversational mode returns text output at a workflow node.

• API mode returns structured JSON for automation and decisioning.

Both modes share the same configuration components: context, system settings, user settings, and model parameters.

Choose the mode based on how the output will be consumed within the workflow.