Automation

Automation is where you build the logic that drives your workspace. It has three layers: Workflows define how conversations flow, Actions respond to events in real time, and Routines run tasks on a recurring basis.

Together, they let you design systems where the AI handles conversations, data flows between systems automatically, and humans get pulled in only when they're needed.

Workflows

The backbone of every AI conversation. Workflows define multi-step processes that guide your agent through conversations.

• Visual workflow builder with connected tasks
• Each task has its own prompt, goal, data collection, and abilities
• Branching logic for conditional conversation paths
• Clone and iterate without starting from scratch

Example: A patient intake workflow with tasks for greeting, screening questions, data collection, and appointment scheduling — each step with its own AI instructions and form.

Actions

Event-triggered automations that fire when things happen in your workspace.

• Trigger on task events, form submissions, member changes, or chat messages
• Filter with CEL expressions for precise targeting (e.g., "vip" in member.labels)
• Actions can send messages, notify team, switch tasks, update labels, fire webhooks, or modify data
• Delayed execution for follow-ups and time-based escalation

Example: When a form is submitted with priority == "high", immediately notify the support team and switch the chat to the priority support task.

Routines

Time-based automation for recurring work that doesn't depend on a specific event.

• CRON-style scheduling: daily, weekly, monthly, or custom intervals
• Target all members, filtered groups, or specific individuals
• Chain multiple actions to a single routine
• Manual trigger option for testing or one-off runs

Example: Every weekday at 9 AM, send a check-in message to all members labeled "active-patient" who haven't responded in 48 hours.

Experiments (A/B Testing)

Member-level A/B tests with weighted Groups, lazy CEL assignment, and journey goal reporting.

• Create control and variant Groups with allocation weights
• Branch in Actions and task edges with member.experiments["slug"]
• Filter who gets assigned with optional eligibility CEL
• Compare Journey Goal outcomes by Group (exposed Members only)
• List filtering for archived experiments; small-number suppression in stats

Example: Test two opening-message variants in a onboarding journey, then compare goal completion rates by Group on the journey Goals tab.

How They Work Together

The real power comes from combining all three:

1. A Workflow guides a patient through intake, collecting their information
2. An Action fires when the intake form is submitted, adding a "new-patient" label and sending a webhook to your EHR
3. A Routine runs daily, checking on patients who haven't completed intake and sending a reminder

This combination lets you build fully automated engagement systems that handle thousands of conversations while keeping humans in the loop where it matters.