Slack alerts & asking the Brain in Slack

Get alerts where your team already works, ask the Brain questions, and share files with it without opening the dashboard.

I'm using the platform as a
All help topics

Connecting Slack gives your team three things in the channel they already sit in all day:

  • Alerts when something needs attention - a spend spike, a broken tracking setup, a goal hit, the weekly performance write-up.
  • Answers - @mention the app with a question and it replies using your own data.
  • A place to share things - drop a document, a screenshot or a photo into the channel and the Brain reads it.

Connecting Slack

Slack is connected from Setup - Connections - Alerts on your dashboard. On a self-serve workspace only a workspace owner can connect it or change its settings; on an account managed by an agency, any member of your team can.

If you don't see an Alerts section there, your account team has Slack switched off for your workspace - ask them to enable it.

  1. Click Add to Slack. It opens in a new tab.
  2. Slack will show an "app not approved" style notice. That's expected for a private integration like this one and is safe to allow.
  3. Choose the channel alerts should go to. We recommend a private channel - see below.
  4. Back on the dashboard, invite the app into the channel so it can reply to you. In that Slack channel, type /invite @ and pick the Agentcroft app from the list Slack offers.

That last step is easy to skip and it's the one that matters for conversations. Alerts will arrive without it, but the app can't answer questions until it's a member of the channel.

Use Send test alert to confirm the connection end to end.

Choosing the right channel

Everyone in the channel gets the alerts and can ask the assistant. Channel membership is how you control access, so pick the channel accordingly.

A private channel is the safer default, since your performance data and anything the Brain knows about the business will be posted there. If you connect a public channel, the dashboard warns you.

If the card can't tell whether your channel is private, that usually just means the app hasn't been invited yet.

What you'll be notified about

Six categories, each an independent tick-box:

CategoryWhat it coversDefault
Critical alertsSpend spikes, budget breaches, conversion drops, broken tracking, integrations going downOn
WinsDeals won, goals hit, record monthsOn
Weekly insightsThe weekly performance narrative from the BrainOn
Monthly reportsA report has been published and is ready to readOn
RecommendationsBrain-suggested actions and tasks coming dueOff
Market & competitorsShare-of-voice shifts, AI-search visibility changes, competitor movesOff

When a connection stops

Critical alerts include your data connections going down. If we can't pull from one of your platforms, you get a message naming it - and a second one when it starts working again.

You'll hear about it once per outage, not once per day. A connection is retried on every refresh, so a long outage would otherwise repeat the same message daily. We send one alert when it stops, then stay quiet until either it recovers or it breaks again later.

While a connection is down, that tab keeps showing the last figures we successfully pulled, and every other channel carries on as normal. Nothing is lost or overwritten.

When you get one, open Connections and press Test connection on the platform named. It will tell you whether the fix is on your side - usually re-authorising an expired login, or re-sharing access - or whether it's something for us to sort out.

Turning down the noise

Alongside the categories there's a minimum severity setting: everything, warnings and critical only, or critical only. Raise it if the channel gets chatty.

Connection alerts are sent as warnings, so setting this to critical only switches off both the "stopped working" and "working again" messages. If you want to be told when a data source goes down, leave it at warnings and critical or lower.

It applies to alerts only. Your weekly insights and monthly report notifications are scheduled updates and always come through if their category is on.

Asking the Brain questions in Slack

@mention the app anywhere in the connected channel and ask in plain English - "how did last month go?", "which channel is my best performer?", "why did conversions drop?". Answers are grounded in your connected data and everything the Brain has learned, exactly like the chat on your dashboard.

Once it has replied, you can carry on in the thread without mentioning it again. It follows the conversation.

The Slack assistant is read and ask only. It will explain, analyse and recommend, but it won't change budgets, launch campaigns or run outreach on your behalf - those stay in the dashboard on purpose.

Sharing files in Slack

Attach a file to a message that @mentions the app and it will read it, the same way an upload on the dashboard does. It lands in the same place and teaches the Brain the same things.

That includes screenshots, photos and scanned documents, not just text files - a screenshot of a competitor's ad, a photo of a whiteboard, a scanned invoice. See Chatting with the Brain for the full list of what it can read.

The @mention matters: a file dropped in the channel without addressing the app is left alone deliberately, so nothing gets absorbed that you didn't mean to share.

If a file turns out to contain text written to manipulate an AI assistant rather than to inform it, the app will say so, ignore those parts, and hold anything it picked up for someone on your team to check.

Acting on recommendations

When a recommendation arrives, you can reply accept or dismiss in the thread to action it without leaving Slack. Anything more conversational is treated as a question, not a command, so you can safely discuss it first.

Disconnecting

Disconnecting stops both the alerts and the in-Slack assistant. You can reconnect at any time, and your category preferences are kept.

Allowing Slack for your clients

Every client can use Slack - self-serve and managed alike. They connect it themselves from their own Setup - Connections tab, into their own Slack workspace; you don't connect it on their behalf.

You control availability at two levels:

  • Workspace-wide: in your agency portal under Settings - Workspace, the Slack integration switch turns it on or off for every client at once. With it off, the Slack card doesn't appear on any client dashboard.
  • Per client: on that client's Client Settings page (Advanced), the Slack card carries an allow switch - on by default. Switching a client off takes effect immediately: alerts stop, the in-channel assistant stops answering (it explains itself once rather than going quiet), and the card disappears from their Setup tab. Their connection is kept, so switching back on resumes where they left off.

The same card shows you what a client has connected - the channel, who connected it and when, which categories they've ticked, and whether recent alerts actually landed - and lets you disconnect their Slack on their behalf if you ever need to.

Points worth knowing:

  • Everyone in a client's chosen channel receives their alerts and can ask the assistant, so the client controls access through channel membership.
  • On a managed account, any member of the client's team can connect Slack and change what gets sent.
  • When you publish a report for a managed client, the Monthly reports category tells their channel it's ready.
  • The assistant is read and ask only - it can explain and recommend, but it cannot make changes to campaigns or budgets.
  • The app that appears in the client's Slack is Agentcroft-branded. It does not currently carry your agency's branding, even on a white-label domain.

Connection alerts, and who gets them

Critical alerts include a client's data connections dropping: one message when a platform stops updating, one when it recovers, and nothing in between however long it lasts.

These ride on Slack, so they reach any client who has connected it - managed clients included. A client without Slack connected is not messaged when a connection drops.

They are not left in the dark, though: a paused connection is flagged on their Connections tab, naming the platform and pointing them at Test connection. The copy they see tells them it may be one for their account team rather than asking them to reconnect something that is not theirs to fix. You see a fuller version of the same note when you open that tab yourself.

For your own view across the whole portfolio, use More - Connection Health in your admin portal: every stopped feed across every client, worst first.