When you need certainty,
not improvisation.
Conversation Strategy shapes the chat. Workflow Builder guarantees the actions: if a customer says X, the system runs Y — always, no exceptions. Draw the flow on a canvas, the agent executes it across every channel.
Discovery — Active Phase
Collecting: Budget, Timeline, Company size
Next trigger: When budget confirmed → move to Qualification
Strategy or Workflow? Here is the difference.
Conversation Strategy
Shapes the tone of the chat
You want the agent to follow a narrative funnel, ask qualifying questions, and close with a demo. The AI model writes the actual sentences based on context.
Examples: B2B lead qualification agent; e-commerce assistant with upsell; multilingual receptionist.
Workflow Builder
Guarantees the actions
You want certain situations to always run the same procedure — POST to Calendly, inventory check via webhook, escalate to operator on negative sentiment. Zero AI improvisation.
Examples: call booking (create Calendly event + write CRM); order refund (call Shopify API + email customer); ticket triage (classify + assign to queue).
They combine: Strategy guides the chat, Workflow runs the actions when the phase reaches "ready_to_book".
Without Workflow Builder
Every "guaranteed" automation needs developers, custom code, deploys. One misplaced comma and the entire flow breaks in production.
With Workflow Builder
Draw the flow on a canvas, test it in draft mode, publish with one click. The system runs it every time — same parameters, same results, zero surprises.
Real scenario
Dental practice, automated bookings 24/7
Practice with 3 dentists. Receives 40 calls a day for bookings. Workflow: trigger keyword "book" → look up patient in CRM → propose Calendly slots → confirm → write event + send email.
I would like to book a dental cleaning.
Happy to help. Can you confirm your full name?
Mark Reynolds
Found you in our records. I have Tue 10:30, Wed 4:00 PM, and Thu 9:30 AM open this week. Which works best?
Wed 4:00 PM.
✓ Workflow executed: event created in Calendly, confirmation email sent to mark.reynolds@email.com, lead updated in the practice CRM.
90 seconds
From request to confirmed — without calling the practice
Zero
Double bookings: the workflow checks availability before proposing
24/7
Operative at night and on weekends too
How it works
Three steps from blank canvas to live workflow.
Open the visual canvas
Go to the Workflow Builder from your agent settings. Start from one of 5 pre-built templates (lead qualification, appointment booking, FAQ with RAG, support triage) or from a blank canvas.
Drag and connect typed nodes
Add typed nodes: trigger, message, condition, tool_call, llm_call, handoff, end. Connect them with arrows — every branch is visible on the canvas and testable in draft mode.
Publish and the agent follows it
When the flow is ready, publish with one click. The agent immediately starts following it in every new conversation matching the trigger. Edit and republish anytime — no backend touch.
Start from a template
Five workflows ready to deploy
Click a template, customize the messages, publish. Three minutes from "new" to "live".
Blank canvas
Start from scratch with just a trigger node. Maximum freedom.
For custom workflows that don't fit a known pattern.
Lead qualification
Collects name, email, phone and submits the lead via tool_call. Handles the "phone skipped" case.
For agencies and B2B that want structured leads in the CRM, not generic forms.
Appointment booking
Triggers on words like "book/appointment", collects date/time/notes, submits booking via tool_call.
For dental practices, hairdressers, consultants — anyone managing a calendar.
FAQ with RAG
Classifies intent (FAQ/contact/other), searches the knowledge base, answers with citations or routes to contact CTA.
For customer support on complex products with extensive documentation.
Support triage
Asks issue type (billing/technical/other), routes billing to KB, technical to a specialized LLM, rest to the agent.
For SaaS and services with multiple support flows that must run deterministically.
The 10 node types you can use
All the workflow engine building blocks. Typed, deterministic, testable in dry-run.
Trigger
Workflow entry point. Fires on keyword ("book"), on any message, or on classified intent.
Message
Sends a text message to the customer. Optionally collects a field (name, email, phone) with validation.
Condition
Yes/no branch on variables: language = EN? field not empty? text contains "refund"? operators equals/contains/not_empty.
Tool action
Runs a typed action: submit_action (send lead/booking to the CRM), search_knowledge_base (search the docs), or custom webhook.
LLM call
Direct call to an AI model with custom prompt. Saves the output to a variable for later nodes to use.
Collect data
Collects structured data from the user (form-style) before continuing — useful for lead qualification and surveys.
Handoff to AI
Passes control to the standard AI agent when the workflow no longer knows what to do. The customer doesn't notice.
Delay
Waits a time interval before the next node. Useful for follow-up after X minutes/hours.
Goto
Jumps to another node in the workflow — loop pattern or "start over if the user changes their mind".
End
Terminates the workflow explicitly. Optionally with a closing message.
This feature works on
Workflows run behind the scenes on all channels — website, WhatsApp, voice, API. Same nodes, same deterministic execution.
Looking for how to define conversation phases, not actions?
Go to Conversation Strategy: it explains how to structure the narrative funnel (greeting → discovery → proposal → close) leaving the AI model to compose the actual sentences.
When you need this
Explore related topics
Critical actions should not be improvised. Draw them once, run them always.
Visual logic, infinite possibilities. Deploy instantly.