Workflow guides

A workflow is one trigger, optional conditions and an ordered list of steps; only a published workflow runs, and this article explains how a run starts, forks, pauses, fails and stops.

The 34 events a workflow can start from — what each one actually watches, what it exposes to conditions and messages, and where it quietly does nothing.

How a condition decides whether a workflow starts or which path a contact takes, and how a {{variable}} is filled into the text a workflow sends.

What already stops a flow from re-triggering itself, and the two hard caps — on Jump To and on Trigger Another Workflow — that stop a deliberate loop from running forever.

Plan a workflow's path against a pretend conversation without sending anything, then publish it — and know what Publish refuses, what it only warns about, and what it locks.

Describe a flow in plain words and the assistant drafts it for you to review: the draft sits in the side panel until you put it on the canvas, nothing is saved until you press Save, and nothing runs on customers until the workflow is published.

Send free-form text — with several attachments if you need them, on the customer's own channel or a named one — inside the messaging window that channel currently has open.

Reach a WhatsApp customer with an approved Meta template regardless of whether the messaging window is open, checked for approval and marketing consent before anything is sent.

Show WhatsApp's native Share Contact Info button so a username-only chat can give you a real phone number to message on.

Ask with a Yes/No button pair inside an open conversation; a tap records the contact's decision, which every future marketing template send reads.

Send up to 5 voice notes, images, videos or documents from a public URL, in numbered order, inside the channel's current messaging window.

Send WhatsApp product cards from the number's own Commerce Manager catalog — one product, a list of up to 30, swipeable cards, or the whole catalog — so the customer can add to cart from inside the chat.

Send the one private DM Meta allows in answer to a Facebook or Instagram comment, with an optional public reply posted under it.

Open a WhatsApp thread with a number a flow just learned — a Messenger lead who typed it, a form, a webhook — using an approved template, since no messaging window is open yet on that number.

Hand this one turn to the thread's AI agent — the same generation and confidence gate as an inbound auto-reply — then continue the flow with what it decided.

Ask, then pause the whole run until the customer answers or a deadline passes — with typed answers matched deterministically against a fixed set of choices, never guessed by a model.

Split the run into up to twelve lanes plus a None lane, checked top to bottom — the first lane whose conditions all pass wins, and Condition tests one field against one value at a time.

Send the run straight to another step already on the canvas, forward or backward, capped by a per-step limit so a deliberate loop cannot run forever.

Route the run down an Open or a Closed path depending on business hours or a date range, checked the instant the step runs, with no branch treated as a dead end.

Pause the run for a set duration or until a time of day, optionally held to business hours or to a daily window so a customer is never messaged in the middle of the night.

Hand the conversation to another published workflow, from its first step or a chosen one, so a large design can stay several small workflows instead of one long flow.

Ask Instagram whether this person follows your account, so a later Condition step can route follower-only offers.

Send each contact down exactly one of several weighted lanes at random, using a fixed per-contact draw so the same contact keeps landing in the same lane as long as the step and its weights stay put.

Compute a price quote from a fixed, built-in product-and-shipping catalog collected by earlier Ask a Question steps, then optionally record the order and send the customer a receipt.

Cancel this conversation's pending waits — the step that stops a reminder chase the moment the contact replies or pays, before the next one is due.

Stop the run right here — including whatever sits further out than the lane the End step is nested in, such as a Condition branch or a Date & Time branch.

Hand the conversation to one named teammate, or spread it across the workspace by round robin or fewest open contacts, with an optional timeout message if nobody answers.

Route the conversation into a team's queue, and optionally also hand it straight to one member of the workspace by round robin or fewest open contacts.

Pin the conversation to one named AI agent for its next replies, overriding whatever would otherwise decide which agent answers.

Move the conversation to Open, Pending, Snoozed or Closed. A close can carry a closing-note category and summary that only this step sets, and can send the same CSAT survey the Inbox's close button sends.

Leave a note on the conversation's timeline for agents — never sent to the customer — with an optional @mention to notify a teammate or the assignee.

Pause the AI on this thread and page a human right now — the safety net at the end of a path the AI could not, or should not, finish alone.

Silence the AI on this conversation and clear its agent binding, with no page, no note and no message — a quiet, workflow-internal hold meant to be lifted by a later Assign To (AI) step.

Add one or more tags to this conversation — read by saved views as a fallback when a contact tag doesn't match, and gone once the thread is.

Remove one or more tags from this conversation — the other half of Tag Conversation, so a flow can clear a state it no longer applies.

Put an item with the thread attached onto the workspace's Tasks list — the thing that survives the end of a shift, which an internal note does not.

Write a value onto one of this workspace's custom contact fields, so an answer collected in the conversation lives on the record instead of only in the transcript.

Add one or more tags to the person behind this conversation — what the contact sidebar, Contacts filters, segments and broadcasts all read, and what a pipeline stage can watch for to move a deal.

Remove one or more tags from the person behind this conversation — clears the old state when someone moves forward, so a segment or broadcast stops targeting them for it.

Move the contact to another lifecycle stage, or clear the stage entirely — what the dashboard's lifecycle counts read, and what a pipeline stage can watch for to move a deal forward.

Finalise this conversation's appointment as completed, a no-show or canceled — the same move the /bookings page's own buttons make, so a webhook or a question the customer answers can record it automatically.

Add this contact to an event's list, which is what makes them eligible for its reminders — a named event, or the next published one coming up.

Write the outcome — attended, no-show or canceled — onto this contact's event registration, with an optional minutes-attended figure for engagement grading.

Open this contact's deal in a pipeline stage, or move it if one is already open — one fact from the funnel's point of view, with an optional amount and a guard against re-opening a deal someone has since moved.

Report a conversion — a lead, a purchase, a cart — to Meta's Conversions API from this conversation's channel, so ad optimisation can see what happened after the click.

Meant to report a lower-funnel event — Place an Order, Complete Payment or one of four others — to TikTok for ad attribution. Not wired up yet: every attempt fails and nothing reaches TikTok.

Call an outside API over HTTPS and keep its response for later steps and conditions. Two things to know before using it today: placeholders in the URL and body are not resolved, and production had no host on its outbound allow-list at the time of writing — see Limits.

Append one row to a connected Google Sheet — the cheapest way to hand data from a conversation to a team that lives in a spreadsheet.