Skip to main content

Workflow Concepts

What is a Workflow?​

A workflow in XY is an automated sequence of steps that accomplishes a specific business outcome—like processing an insurance claim, extracting data from documents, or managing patient scheduling. Think of it as a digital assembly line where each station (step) performs a specific task, and the work flows smoothly from one station to the next.

Core Components​

1. Steps​

The fundamental building blocks of any workflow. Each step represents a single action or decision point:

  • Agent Steps: AI-powered agents that handle complex tasks (e.g., extracting data from a PDF, checking claim codes)
  • Integration Steps: Connect to external systems (e.g., pull data from DrChrono, send to a billing system)
  • Decision Steps: Conditional logic that routes work based on rules (e.g., if claim > $10,000, route for manual review)
  • Human Steps: Points where a person needs to review, approve, or provide input

2. Agents​

Specialized AI workers that execute intelligent tasks within workflows:

  • Each agent has a specific expertise (claims processing, data entry, scheduling, etc.)
  • Agents can use tools to interact with external systems
  • They process inputs and produce structured outputs for the next step
  • Think of them as your digital workforce—each trained for specific jobs

3. Data Flow​

Information moves through your workflow like water through pipes:

  • Inputs: Starting data (files, form submissions, API calls)
  • Variables: Store and pass data between steps
  • Transformations: Modify data as it flows (extract, format, validate)
  • Outputs: Final results (updated records, processed documents, notifications)

4. Dependencies​

Steps can depend on other steps completing first:

  • Sequential: Steps run one after another in order
  • Parallel: Multiple steps run simultaneously when possible
  • Conditional: Steps run only when certain conditions are met

Workflow Types​

Template Workflows​

Pre-built, reusable workflow patterns for common tasks:

  • Invoice processing
  • Claims submission
  • Patient intake
  • Appointment scheduling

Think of these as blueprints—you can use them as-is or customize them for your needs.

Custom Workflows​

Workflows built specifically for your unique processes:

  • Tailored to your organization's specific requirements
  • Combine multiple agents and integrations
  • Include your business rules and logic

Instance Workflows​

Active, running copies of workflows processing real data:

  • Created from templates when triggered
  • Track their own progress and state
  • Can be monitored and managed individually

Execution Modes​

Manual (Co-pilot)​

  • You initiate each workflow run manually
  • Great for testing or one-off tasks
  • Full visibility into each step's execution
  • Ability to pause, modify, or stop mid-flow

Supervised (HitL - Human in the Loop)​

  • Workflows run automatically but pause for human input at key points
  • Ideal for high-stakes decisions or exception handling
  • Humans approve critical steps before proceeding
  • Balances automation with human judgment

Automated (Autopilot)​

  • Fully autonomous execution from start to finish
  • Triggered by schedules, events, or API calls
  • Runs 24/7 without intervention
  • Best for repetitive, well-defined processes

Key Concepts to Remember​

Idempotency​

Workflows are designed to be safely re-run without causing problems. If a workflow fails halfway through, it can often be restarted without duplicating work already completed.

Error Handling​

Every workflow includes built-in error handling:

  • Retries: Automatic attempts to recover from temporary failures
  • Fallbacks: Alternative paths when the primary path fails
  • Notifications: Alerts when human intervention is needed
  • Logging: Complete audit trail of all actions taken

Queue Management​

Workflows use queues to manage work:

  • Each workflow gets its own dedicated queue for isolation
  • Items wait in queues until resources are available
  • Priority levels ensure urgent work gets processed first
  • Failed items can be retried or sent to error queues

State Management​

Workflows maintain their state throughout execution:

  • Pending: Waiting to start
  • Running: Currently executing
  • Paused: Waiting for human input or scheduled time
  • Completed: Successfully finished
  • Failed: Encountered an unrecoverable error

Real-World Example​

Let's say you need to process insurance claims:

  1. Trigger: A claim PDF arrives via email
  2. Step 1: Data Gateway Agent extracts the attachment
  3. Step 2: Data Entry Agent reads and structures the claim data
  4. Step 3: Claim Validation Agent checks for required fields and valid codes
  5. Decision Point: If validation passes, continue; if not, route to human queue
  6. Step 4: Database Agent saves the claim to your system
  7. Step 5: Notification Agent alerts the billing team
  8. Complete: Claim is processed and ready for submission

This entire process happens automatically, taking minutes instead of hours, with full visibility and control at every step.

Why Workflows Matter​

Workflows transform manual, error-prone processes into reliable, scalable operations:

  • Consistency: Every task follows the same proven process
  • Speed: Parallel execution and 24/7 operation
  • Accuracy: AI agents don't make typos or miss fields
  • Visibility: Track every step and decision
  • Scalability: Handle 10 or 10,000 items with the same workflow
  • Compliance: Built-in audit trails and approval gates

Next Steps​

Now that you understand the concepts, you're ready to:

  • Start building automated processes for your organization
  • Connect workflows to your existing systems and tools
  • Work with your XY customer success representative to create workflows tailored to your needs