Skip to main content

Automations

Automations follow up on their own when something happens on your site. An automation (a rule) watches for a moment — a visit's lead score crossing a number, a first visit, a view of your pricing page, a form, a funnel stage change — and runs a workflow of steps for that person: place an AI call, text them if they don't answer, tag them in your CRM, tell your team.

Automations are available on every account, with or without Agent K. Open Automations in the main sidebar (it's shown to anyone who can manage your AI agents). Old links to Agent K's automations land there.

Who an automation runs on. On a visit, it runs on the Agent K leads behind that visit — or, when there are none, on the visitor themselves, using the name, email and phone they gave you and their CRM contact if they're linked to one. Bots are never acted on, and neither is a CRM contact that enrichment only guessed. A funnel stage change runs on that CRM contact.

With Agent K on you also get its triggers (email clicks, new leads, emails without a visit), the Came from filter, the Find phone number / Find LinkedIn steps, Test on a lead, and bulk actions. Without Agent K those triggers aren't offered, and a lookup step is skipped ("lookups are only for Agent K leads").

The Automations page​

Two tabs: Automations (your rules) and Logs (what they did — see Logs). New automation opens the workflow editor.

Four tiles sit along the top:

  • Automations — Running or All paused. Its switch is a kill switch: Pause all automations? stops every automation until you resume. Anything already booked or waiting for approval is untouched.
  • Sending hours and Message limit per contact — the limits every automation runs under. Change opens When messages may go out.
  • Waiting for you — calls and texts automations have lined up, with Review to open the Waiting to go out list: what needs your approval first, then what's going out on its own.

Automation cards​

Each automation is a card:

  • A switch to Turn on / Turn off, its name, and a status — Live, Dry run, Off, or Paused.
  • The recipe in one line — When (what starts it) and Then (each step in order; a step marked · asks you first waits for your approval).
  • Edit, Test on a lead (with Agent K on — see Test on a lead) and delete.
  • Today — how many people it acted on against its daily limit, and the credits used against its budget — plus its Last 7 days.

An automation that stopped itself (for example, its daily credit budget ran out) says Paused itself: and why, with Resume. You also get a notification — "Automation … paused itself".

No automations yet? A good first one: when a visitor's lead score reaches 50, tell your team — or book an AI call that waits for your approval.

When messages may go out​

These limits apply to every automation — and to Nox and Agent K too:

  • Quiet hours — Send from / Until. Calls, texts and emails go out only inside this window, in each contact's own time zone. Anything due outside it waits until the window opens.
  • Limit how often one contact is messaged — At most a number of messages per channel In the last a number of days. Off means no limit at all.
  • Space out messages to one contact — Wait at least a number of minutes between Nox's follow-ups to the same person. Automations don't wait for this.

The workflow editor​

New automation (or Edit on a card) opens the workflow editor full-screen. It's laid out like a LeadConnector workflow, with three tabs — Builder, Settings, and Execution logs — and a status switch at the top: Draft (doesn't run), Test (logs what it would do — nothing sent, nothing charged), or Published (acts on people who match, within your quiet hours, send cap and do-not-contact list). On the card these read Off, Dry run and Live. A new workflow starts as a Draft. Name it and click Save.

Builder — the workflow runs top-down from its trigger cards:

  • Click the + on any line to add a step there, picked from a searchable list:
    • Find — Find phone number, Find LinkedIn (Agent K leads only). The steps after a lookup wait for its answer.
    • Communicate — AI call, Text, Email, LinkedIn invite, Greet on your site.
    • Send to — Add to ManyReach campaign.
    • Your team & CRM — Add tag (tags to add in your CRM), Notify your team (an in-app and push notification naming who it's about — for a visitor it opens their visit, for an Agent K lead the lead).
    • Flow — Wait (minutes, hours or days, up to 30 days) and If / Else.
  • If / Else splits into a Yes path and a No path, side by side; each path ends in END or joins another step (shown as Go to). The checks: They came back to your site, They replied, They clicked a link in Agent K's email, They booked a meeting, They answered the AI call (only after a call step), We have their phone number, We have their email, We have their LinkedIn, They're in LeadConnector, and Their lead score is at least…. "Came back", "replied", "clicked" and "booked" count since the last wait before the check.
  • Click any step or trigger to set it up in the side panel, or delete it there:
    • AI call / Text — the agent (Outbound agent / Sends from) and Extra instructions (optional). If the chosen agent has no phone number the step warns you, with a link to add one — otherwise that step would be skipped.
    • Email — AI-written for each lead or My template (subject and body). Template merge fields: {{first_name}}, {{company}}, {{why_now}}, and {{link}} — your tracked site link. Any link in the template that points to your own website gets the lead's tracking token too, so their click is recognized.
    • Add tag — Tags to add in your CRM, separated by commas. A tag can start one of your CRM's workflows.
    • Add to ManyReach campaign — pick the Campaign. The lead is added with their tracking token and your campaign links, and the campaign emails them. It needs ManyReach connected (the step links to Integrations if it isn't), skips anyone on the do-not-contact list or without an email, and — as there's no approval queue for a campaign — skips leads you don't allow to be emailed without asking.
    • Greet on your site — the greeting.

"Text them if they don't answer" is an AI call followed by an If / Else — They answered the AI call — with a Text on the No path. A workflow can't loop back on itself, can't end on a Wait, and every If / Else needs a step on at least one path; if a save fails, the editor opens the step or trigger to fix.

Settings — who can enter and the limits:

  • Came from (with Agent K on: New people Agent K found, Past contacts in your CRM, Your website visitors — or Any source), Funnel stage (or Any stage), Already in your CRM, Not contacted in the last (days), and Lead score at least (not used by the lead-score trigger, which has its own number).
  • Limits — each one has its own switch; off means no limit:
    • Limit how many people it acts on a day — people per day, counted in your time zone (20 by default).
    • Run once per person, then wait — days before someone who went through can start it again (30 by default). Off means it can run for them every time.
    • Daily credit budget — credits per day. Lookups that find something use credits; the automation pauses for the day once it's spent.

Execution logs — what this workflow did, person by person (the same entries as Logs).

What starts a rule​

Triggers sit at the top of the builder. An automation can have up to five — Add New Trigger adds another — and any one of them starts it:

  • A visit's lead score reaches a number — A visit's lead score reaches N, optionally Only while they're on your site.
  • They're highly engaged — a visit's engagement score reaches a number — A visit's engagement score reaches N (out of 100, scored a few minutes after each visit).
  • They visit your website for the first time (once per visitor).
  • They visit your website — every visit; a new visit starts after 30 minutes away.
  • They view a specific page — Page address contains (for example /pricing).
  • They submit a form on your site.
  • Their funnel stage changes — From (or Any stage) and To: the moment a contact moves between those funnel stages, up or down, automatically or by hand.

With Agent K on, also:

  • They click a link in Agent K's email.
  • Agent K finds a new lead — including website visitors as they become leads.
  • They got emails but haven't visited — After this many emails and And this many days: checked hourly, for a lead who got that many emails (from Agent K, or from your LeadConnector workflows) and hasn't been on your site since the first. Days count from the last email; 0 = as soon as it goes out.

How an automation's messages go out​

An automation is your deliberate choice, so its steps send the way you built them — Agent K's Sending grid doesn't sit on top of it. Quiet hours, do-not-contact, the message limit per contact, and the daily send cap still apply; the Space out messages to one contact gap doesn't. Calls and texts an automation schedules show in Waiting to go out and on the Calls page, which says when a call was moved and why.

With Agent K on, an automation that can reach people who haven't visited your site — new people Agent K found, or past contacts in your CRM — asks once for I confirm we have consent to contact these people when you save it. It's recorded once for your account.

Credits a run might spend are reserved up front and given back for any branch it never reached. A run where every step was skipped (no phone number, say) doesn't use up a place in the daily limit, and the person can come back through the next day rather than after the full wait.

Test on a lead​

With Agent K on, Test on a lead on a card opens a searchable picker (search by name, email or company). Test walks the automation against that lead without sending anything: whether their lead score clears the threshold (event-triggered automations act as if the event just happened), whether it would act, and each step with its outcome and reason — Done, Queued, Waiting for approval, Skipped, Failed, or Would run.

Logs​

The Logs tab is the History of every decision your automations made — ran, didn't run and why, step by step, with what each queued call or text is doing now ("Waits for the call — only if they don't answer", No answer / Answered / Booked, or why a text never went), and the credits it cost. Filter by automation and by result (Ran, Didn't run, Problems, Test runs). Entries link to Open the lead → or Watch the visit →, and a step waiting for approval opens Waiting to go out. The same decisions show up in Nox Activity.

Bulk actions​

With Agent K on, tick leads in its Feed (or select every lead matching your filters, up to 500) and a bulk bar appears: Text, Call, Email, and under More — Find phone numbers, Find LinkedIn profiles, LinkedIn invite, Push to LeadConnector, Approve & send drafts, and Skip. Each action shows how many of the selected leads it applies to; one that fits none is greyed out with the reason ("Not for these leads — no phone number").

The bulk dialog sets the action up — the agent and extra instructions for calls and texts (or the exact Text to send, with merge fields like {{first_name}}), an AI-written or template email (template merge fields include {{link}}, your tracked site link) — and previews the run before anything happens: who's eligible, who's skipped and why, and what the lookups will cost. Like automations, a bulk run sends the way you set it up and asks once for consent when it includes people who haven't visited your site. Start it and watch the progress, or Keep running in background — a progress chip follows the run to the end. One bulk run at a time per account. Bulk runs appear in Logs under Bulk runs.

Test bench​

info

The test bench is enabled by the Knock Knock team — contact support. Its sends are real: calls, texts and emails go to the details you type, and are charged like any other.

With it on, + Test lead in the Agent K Feed's rail creates a lead from your own details — name, email, phone, LinkedIn, company and title — shaped for the case you want to try: where it Came from, its Why-now signal, its State, and a LeadConnector contact id (blank = not in your CRM yet). Test leads carry a TEST chip, and TEST LEAD · edit on the lead reopens the form. From the dialog, set a visit score and run it to see what each automation did.

Automation cards also get Run for real beside Test: after a confirmation (Run this automation for real?), every automatic step actually happens to the chosen lead — calls dial, texts and emails send, and your wallet is charged; steps set to wait for approval queue instead — and the result says what was done.

Test leads never count in Reports.