> ## Documentation Index
> Fetch the complete documentation index at: https://docs.dmly.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Automation not triggering

> Work out why a bot never started, in the order that finds the cause fastest — from the channel it is built on to the guards that stop a flow without saying so.

The flow is built, the trigger looks right, a customer messaged — and nothing happened. Work
through this page in order. It follows the same order DMLY itself checks things in, so the first
check that fails is your answer.

The uncomfortable part: of the reasons a flow refuses to start, **only two write anything to
[Logs](/troubleshooting/logs)**. Silence is not proof that nothing ran. That is why the first four
steps are things you check in the product rather than in the log.

## What has to be true

An automation starts for a contact only when **every** one of these holds. Any single one false
and the flow never runs:

* It is built on the **same channel** the contact messaged on.
* Its status is **Active** — not **Paused**, not **Draft**.
* It has a **published version**.
* Its trigger matches the event, and the contact's message gets past the keyword filter.

## Work through it in this order

<Steps>
  <Step title="Select the channel the contact actually messaged on">
    An automation belongs to one [channel](/channels/overview) and only answers messages arriving
    on that channel. A flow built on your WhatsApp number will never answer a Messenger DM, however
    correct the trigger looks. There is no way to share one automation across channels — build a
    second automation on the second channel.

    Do this before anything else, because [Logs](/troubleshooting/logs) only ever shows you the
    channel you currently have selected. If the bug *is* a channel mismatch and you go looking for
    evidence on the wrong channel, you will find nothing and draw the wrong conclusion.
  </Step>

  <Step title="Check the status, then check it is published">
    On the Automation list, the row must read **Active**. If the **On/off** switch springs back when
    you turn it on, you are at your plan's cap on active automations for that channel — pause one
    you no longer need, or see [Plans](/billing/plans).

    Then confirm it is published. Editing the canvas writes to the **draft**: **Save draft** and
    autosave both change the draft only, and the running bot is still the previous published version
    — or nothing at all, if you have never published. Select **Publish**. See
    [Publishing bots](/automation/publishing-bots).

    An unpublished flow is one of the two causes that leaves no trace anywhere. Rule it out here
    rather than hunting for it later.
  </Step>

  <Step title="Compare the trigger with what the customer actually sent">
    First the event: a **WhatsApp message** trigger answers a message, not a call and not a cart
    order. See [Triggers](/automation/triggers) for what raises each one.

    Then the keywords, which are the quiet one. With any match mode other than **Any message (no
    keyword filter)**, a message that does not match is dropped and **nothing is logged** — the flow
    is simply skipped. Matching ignores capitals and surrounding spaces, and the mode decides the
    rest:

    * **Message is an exact match** — the whole message has to equal the keyword. `hi there` does
      not match a keyword of `hi`.
    * **Message starts with a keyword** — anything after the keyword is ignored.
    * **Message contains a keyword** — the keyword can appear anywhere.

    Paste the customer's real message next to your keyword and read it against the mode you chose.
    An exact-match keyword is the most common silent filter there is.
  </Step>

  <Step title="Check the bot is not paused on that conversation">
    Open the conversation in the [Inbox](/inbox/overview). If the bot is paused for that contact,
    every automation is frozen for them — flows, sequences, AI, auto-replies — and again, nothing is
    logged. Select **Resume bot**.

    Some things pause it for you: an AI agent handing over to a human, a flow with a handover step,
    and blocking a contact. Unblocking does not resume the bot; you have to select **Resume bot**
    yourself.

    <Note>
      A flow containing a **Resume bot replies** step is allowed to run while the bot is paused — but
      only when something other than an inbound message starts it (a scheduled or appointment
      trigger, or another flow). An incoming message from that contact is dropped before any flow is
      considered, so you cannot build a "customer types *resume*" escape hatch. Select **Resume bot**
      in the Inbox instead.
    </Note>
  </Step>

  <Step title="Read Logs for a suppressed start">
    With the right channel selected, open **Configurations → Logs**, select **Filters**, and set
    **Event** to **Automations** and **Status** to **Warning**. Read the result against the table
    below.
  </Step>
</Steps>

## What Logs will and will not tell you

Four things stop a flow from starting. Two of them announce themselves; two say nothing at all.

| What you see                                                         | What it means                                                                                                                                                                                   |
| -------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `Flow not started — an activation is already in progress (<status>)` | A run of this automation is already live for this contact. Reason code `already_active`. See below.                                                                                             |
| `Flow not started — "only respond once per contact" already fired`   | The trigger has **Only respond once per contact** set and this contact has already had their one run — including a run that failed or was cancelled. Reason code `once_per_contact`.            |
| **Nothing at all**                                                   | Not evidence that the engine never looked. A paused bot and a flow with no published version are both silent, and a keyword filter that fails is silent too. Steps 1–4 are what rule those out. |

<Tip>
  Read the silence as a signal rather than a dead end. A bot that fires for nobody, with a
  completely empty automation log, is almost always an unpublished draft. A bot that fires for
  everyone *except* one person is almost always a paused bot on that one conversation.
</Tip>

## A run that is already in progress

Only one activation of the same automation can be live for the same contact at a time. While a step
is waiting for that contact's reply, a fresh trigger is dropped and logged as `already_active`. That
is deliberate — it is what stops one person triggering ten copies of the same conversation.

Dead runs do not block that contact forever. Before the guard is applied, DMLY clears out runs that
can no longer go anywhere, and each clean-up writes its own warning so you can see it happened:

* A run that got stuck part-way through a step for more than **5 minutes** is marked failed with the
  reason `stuck_running`, logged as `Flow run reaped — was stuck Running`.
* A run waiting on a reply or an input that has been idle for **24 hours** is cancelled, logged as
  `Flow run expired — session idle`.
* A step that is *meant* to be waiting — a delay you built into the flow — is left alone, however
  long it has to wait. It resumes on its own schedule.

Either way the contact is released and the next trigger starts a fresh run. **If a contact seems
stuck, wait rather than rebuilding the flow.**

<Note>
  `Flow run expired — session idle` is worth reading as feedback on the flow, not just as
  housekeeping. It means you asked a question the contact never answered. A pile of them on one
  automation points at the question, not at the engine.
</Note>

## Chains stop after three hops

An automation that starts another automation is capped at three hops. Flow A can start B, B can
start C, and C can start D — but a further hop out of D goes nowhere. The cap exists so a flow that
eventually points back at itself cannot loop forever.

This one is not silent. The hop that exceeds the cap fails the run with `start_flow_depth_exceeded`
and writes an **Error** log — `Flow run failed` — so you will see it in [Logs](/troubleshooting/logs).
Restructure so the work happens within three hops.

## It started, but the customer got nothing

That is a different problem with a different page. If the flow clearly ran and the message simply
never arrived, go to [Messages not sending](/troubleshooting/messages-not-sending).

One thing worth knowing before you go: a flow's send step always leaves a message bubble, marked
failed if the send did not work, plus a `send_failed` entry in [Logs](/troubleshooting/logs). So for
a flow, an empty conversation *is* evidence the step never sent.

## Still stuck

If you have been through all five steps and the automation log is still empty, the cause is
upstream: the message may never have reached DMLY at all. Check whether inbound messages are
arriving on that channel — see [No inbound messages](/channels/no-inbound-messages) — before you
touch the flow again.

{/* TODO(verify): "Only respond once per contact" is taken from the log line's own wording. Confirm the literal label on the trigger step before shipping. */}

## Next

<Columns cols={2}>
  <Card title="Reading the Logs page" icon="clipboard-list" href="/troubleshooting/logs">
    Filters, statuses, and what the page cannot show you.
  </Card>

  <Card title="Common mistakes" icon="triangle-exclamation" href="/automation/common-mistakes">
    Duplicate replies, blocked publishes, and the WhatsApp traps.
  </Card>

  <Card title="Triggers" icon="bolt" href="/automation/triggers">
    What starts a flow, and how keyword matching works.
  </Card>

  <Card title="Messages not sending" icon="paper-plane" href="/troubleshooting/messages-not-sending">
    The flow ran, the message did not arrive.
  </Card>
</Columns>