AI in Automations

Intelligent automation actions that analyze, classify, and transform data using artificial intelligence

Elementum’s AI-powered automation actions process unstructured information, make classification decisions, and transform data within your automation workflows. Each AI action receives data through variables and produces structured outputs that subsequent actions in your sequence can reference.

For a full list of every available automation action (including AI actions), see the Automation Actions Reference. For trigger configuration, see the Automation Triggers Reference.

Adding AI Actions to an Automation

AI actions are added the same way as any other action in the automation builder. If you haven’t built an automation before, start with the Automation System guide.

1. Open an App and navigate to the Automations tab
2. Create a new automation or edit an existing one
3. Add a trigger (or use the existing one) to define when the automation runs
4. Click Add Action at the point in the sequence where you want AI processing
5. Select an AI action from the action list — AI Classification, AI Summarization, Transform Data with AI, or AI File Analysis
6. Configure the action’s input by mapping variables from the trigger or previous actions
7. Use the action’s output variables in subsequent steps for routing, record updates, or notifications

Start with a single AI action before chaining multiple together. Validate that its output matches your expectations using the automation history, then build on it.

AI Actions

AI Classification

AI Classification analyzes content and assigns categories, tags, or labels based on patterns and context. It returns a classification result along with a confidence score that you can use to route workflows or flag low-confidence results for human review.

Common applications:

• Categorize support tickets by type, priority, and department
• Qualify leads as hot, warm, or cold based on form data
• Classify uploaded documents by type (contract, invoice, report)
• Analyze customer feedback for sentiment and topic

Example flow:

Support Email Received → AI Classification → IF category = "Bug" → Route to Engineering

Output variables:

ai_classification.category = "Bug Report"
ai_classification.confidence = 0.92
ai_classification.priority = "High"

AI Summarization

AI Summarization condenses long-form content into key points and insights. It extracts the most relevant information from text fields, documents, or conversation histories.

Common applications:

• Convert detailed reports into executive summaries
• Extract action items and decisions from meeting notes
• Summarize customer feedback to identify recurring themes
• Create digests of lengthy documents for quick review

Example flow:

Monthly Report Generated → AI Summarization → Send Email to Executive Team

Output variables:

ai_summarization.summary = "Key project milestones achieved with 15% budget savings"
ai_summarization.key_points = ["Budget savings achieved", "Timeline on track", "Resource optimization"]
ai_summarization.sentiment = "Positive"

Transform Data with AI

Transform Data with AI cleans, normalizes, and standardizes data by recognizing patterns and applying context-aware corrections. This is especially useful when ingesting data from external sources with inconsistent formatting.

Common applications:

• Standardize addresses: "123 main st, NYC, ny" becomes "123 Main Street, New York, NY"
• Format phone numbers: "5551234567" becomes "[phone]"
• Normalize company names: "Microsoft Corp" becomes "Microsoft Corporation"
• Enrich records with consistent formatting and structure

Example flow:

Customer Data Imported → Transform Data with AI → Update Customer Records

Output variables:

transform_data.standardized_address = "123 Main Street, Suite 200, New York, NY 10001"
transform_data.formatted_phone = "[phone]"
transform_data.confidence_score = 0.95

AI File Analysis

AI File Analysis extracts structured information from uploaded documents by understanding the content and context of different document types. It identifies key data points automatically based on the document format.

Common applications:

• Extract parties, dates, terms, and obligations from contracts
• Pull vendor information, line items, and due dates from invoices
• Parse skills, experience, and education from resumes
• Identify key metrics and trends from business reports

Example flow:

Contract PDF Uploaded → AI File Analysis → Create Contract Record → Set Renewal Reminders

Output variables:

ai_file_analysis.contract_parties = ["Company A", "Company B"]
ai_file_analysis.effective_date = "2024-01-15"
ai_file_analysis.contract_value = "$50,000"
ai_file_analysis.renewal_date = "2025-01-15"

Combining AI Actions

You can chain multiple AI actions within a single automation for more comprehensive processing. Each action’s output variables are available to all subsequent actions in the sequence.

Sequential processing — Process a document through multiple AI stages:

Document Uploaded → AI File Analysis → AI Classification → AI Summarization → Create Record

Conditional routing — Use AI results to branch your workflow:

Customer Email → AI Classification →
  IF category = "Complaint" → AI Summarization → Route to Manager
  OTHERWISE → Standard Response → Route to Support Team

Multi-action enrichment — Combine classification with transformation:

Lead Form Submitted → Transform Data with AI → AI Classification →
  IF quality = "Hot" → Immediate Follow-up
  IF quality = "Warm" → Add to Nurture Campaign

Example: Document Processing Pipeline

This example shows how multiple AI actions work together in a single automation to process an incoming invoice end-to-end.

Scenario: A vendor emails an invoice PDF. The automation extracts the data, standardizes it, classifies the invoice for routing, and creates a record.

Email Received (invoice PDF attached) →
  AI File Analysis → Transform Data with AI → AI Classification →
  Create Record → IF amount > $5,000 → Start Approval Process
                  OTHERWISE → Update Record Fields (auto-approved)

Step-by-step data flow:

1. AI File Analysis extracts raw data from the PDF:

   • ai_file_analysis.vendor_name = "ACME corp."
   • ai_file_analysis.amount = "$2,450.00"
   • ai_file_analysis.due_date = "3/15/2026"

2. Transform Data with AI standardizes the extracted values:

   • transform_data.vendor_name = "ACME Corporation"
   • transform_data.amount = "$2,450.00"
   • transform_data.due_date = "2026-03-15"

3. AI Classification categorizes the invoice:

   • ai_classification.category = "Office Supplies"
   • ai_classification.department = "Operations"
   • ai_classification.confidence = 0.89

4. Create Record uses all three outputs to populate an invoice record with clean, classified data — then routing logic handles approval based on the amount.

Tips for AI Actions

These tips apply specifically to AI actions in automations. For general automation design guidance, see Automation Best Practices.

Provide sufficient context in inputs. AI actions produce better results when they receive relevant background information. When configuring an AI Classification action for support tickets, pass the full email body and subject line rather than just a snippet — the additional context improves categorization accuracy.

Use confidence scores to gate decisions. AI actions return a confidence score with their results. Use IF conditions to check confidence before acting on the output. For example, route results with confidence above 0.85 through the automated path, and send lower-confidence results to a human reviewer.

AI Classification → IF ai_classification.confidence > 0.85
  → Automated routing
  OTHERWISE → Make Assignment (manual review queue)

Validate outputs before writing to records. Add IF conditions to verify that AI action outputs contain expected values before using them in Update Record Fields or Create Record actions. This prevents empty or malformed data from entering your system when an AI action produces an unexpected result.

Match the action to the task. Each AI action is designed for a specific type of processing:

Task	Action
Categorize data into groups	AI Classification
Condense long content	AI Summarization
Standardize or clean data	Transform Data with AI
Extract data from documents	AI File Analysis
Research, reasoning, or multi-step judgment	Run Agent Task

Agent Tasks in Automations

Beyond the four AI actions above, you can assign tasks to AI agents directly within automations using the Run Agent Task action. This bridges structured automation steps with autonomous, reasoning-based task completion — the automation gathers context and the agent handles analysis, research, or decision-making that goes beyond predefined rules.

Example flow:

Order Approved → Search Records (customer history) → Run Agent Task ("Research customer industry") → Update Record Fields → Send Email

The agent works autonomously in a headless environment (no user interaction), then returns structured output that subsequent automation actions can reference as variables.

AI Actions vs. Agent Tasks

AI actions and agent tasks serve different purposes within an automation. Use this as a quick guide for choosing between them:

Use AI actions when:

• The task fits a well-defined pattern (classify, summarize, transform, extract)
• You need fast, deterministic processing
• The expected output structure is consistent across runs

Use Run Agent Task when:

• The task requires reasoning, judgment, or synthesis of multiple data points
• You need the agent to research or evaluate something that doesn’t fit a predefined template
• The output depends on contextual interpretation rather than pattern matching

You can combine both approaches in a single automation — for example, use AI File Analysis to extract contract data, then hand the extracted data to an agent task for risk assessment that requires business judgment.

For a complete guide on configuring agent tasks, writing task definitions, designing structured output fields, and best practices for headless operation, see Agent Task Automation.

Agent Conversation Ended Trigger

The Agent Conversation Ended trigger fires when an AI agent completes a conversation, making the conversation context available for further automation processing. Use it to generate conversation summaries, create follow-up tasks, update customer records, or trigger post-conversation workflows.

Agent Conversation Ended → AI Summarization → Create Follow-up Tasks → Update Customer Record

When this trigger fires, your automation can access the full conversation transcript (trigger.transcript), conversation type (Teams, Voice), and all field values from the record the conversation is associated with. Pass the transcript to AI Summarization or AI Classification to extract structured insights for downstream actions.

For trigger configuration details including conversation type filtering, timeout settings, and where conditions, see Agent Conversation Ended in the Triggers Reference.

Next Steps

Automation Actions Reference

Full configuration details and variable outputs for every action, including AI actions

Agent Task Automation

In-depth guide to bridging structured automation with agent intelligence

Automation System

Understand the event-driven framework that powers all automations

Automation Best Practices

Design principles, performance strategies, and proven patterns