--- title: "Slack Integration: AI Agent in 10 Seconds | JustAutomate" description: "An AI agent in your channels and DMs: a question about company data gets an answer in 10 seconds, the action runs from the message. Live in 2 to 6 weeks." url: "https://justautomate.ai/en/integrations/slack/" lang: en --- Tasks, communication and customer service # Slack integration an AI agent where the team already works Your team sits in Slack all day, while company data sits in the CRM, in the task system and in the data warehouse. We put an AI agent in there: you ask it on a channel or by direct message and you get an answer built from your own data in 5 to 10 seconds. The action runs from the message, with a button or a command, without opening one more tool. Book a free consultation [See how the workshop works](https://justautomate.ai/en/services/process-workshop/) - 3 secondsis all Slack gives an app to acknowledge an event with an HTTP 2xx response, before it treats the delivery as failed and retries three times. Slack developer documentation - 5 to 10 secondsfrom a question asked on a channel to an answer the agent builds out of your data - 2 to 6 weeksfrom workshop to an agent answering in your Slack, depending on how many systems the answer has to come from The problem ## How many times a day does your team leave Slack to fetch one number? Somebody asks on a channel how many orders are waiting for payment. A second person logs into the system, filters a view, copies the number and pastes it into a reply. A question about the stage of a client case travels through three people before it reaches the one with CRM access. A rep messages a manager privately, because he does not know which of five tools holds the answer. Slack has something none of those systems has: the whole history of what was agreed, in one thread, with the files and the decisions. What it does not have is data. So the team works in two modes at once, talking in one place and searching in several others. Every one of those trips costs a few minutes and ends with a number pasted in by hand, which nobody will check afterwards. A bot built in-house over a weekend usually breaks on one thing. The [Slack developer documentation](https://docs.slack.dev/apis/events-api) requires an app to acknowledge an event with an HTTP 2xx response within three seconds, and after a failed attempt Slack retries the call three times. A question about company data takes longer than three seconds, so the channel first shows an error and then gets the same answer several times over. That is why we separate the acknowledgement from the answer. - Answering a simple question about data means logging into the CRM - What you agreed with the client is in a thread, the numbers are in a system the asker never opens - The same question goes round the channel, because nobody knows who holds the answer - A bot put together in a hurry shows an error and then answers the same thing three times Why now ## What changed, so that Slack can be the interface to company data? Slack stopped being only a chat tool and became a platform you can build an app on. A bot listens for events, replies with a message carrying fields and buttons, handles slash commands and opens a form in a separate window. What it does not do matters more: it is a member only of the channels you add it to, and outside them it sees nothing. The second change is on the AI agent side. A question asked in a plain sentence can now be turned into a query against the CRM, the data warehouse and the task system, with the result handed back as one paragraph. BOND, our company AI agent, does that in 5 to 10 seconds and asks nothing of the person about which system holds the answer. Nobody builds a report just to learn one number. The third change is the direction of travel, and it is the one that sorts the whole subject out. Notifications run FROM the system TO the person, and we cover those separately: one scenario in [automatic Slack notifications](https://justautomate.ai/en/automations/automatic-slack-notifications/), the whole area with thresholds and escalation in [notification automation](https://justautomate.ai/en/services/notification-automation/). This page is about traffic the other way: a person asks, the agent answers, a person clicks, the system executes. What changes ## What does a team that stays in Slack get? - ### A question about company data without leaving the channel You write to the agent on a channel or by direct message: how many orders are waiting for payment, what stage this case is at, who owns this client. The answer is built from your systems, not from the model's general knowledge. An answer from company data in 5 to 10 seconds, in English or in Polish - ### An action run from the message, with no system opened Under the answer there is a button or a command that does something in the source system: moves a stage, assigns a task, approves a document. Longer input we collect with a form in a separate window, so the channel does not fill up with it. A button, a slash command or a form window, triggered straight from the message - ### One channel per client or project, with one history Decisions, files and numbers from the systems land in a single channel, so a new person reads one thread instead of asking three people. The agent in that channel works in the context of that case and answers about it, not about the whole company. One channel per client instead of an email thread, a file on a drive and a note in the CRM - ### A direct message for what is not for the channel Questions about margin on a client, about pay or about personal data go to the agent in a direct message, and only the person asking sees the answer. The range of data the agent reaches for on behalf of a given person follows from their role. Data scope per role, set in the workshop and written into the agent configuration - ### You can see who asked what and what came of it Every question and every action is recorded: who, when, about what, with what result. When somebody disputes a number shown on a channel, you can reconstruct which system it came from and at what time it was pulled. A log of questions and actions, the same mechanism our own agent runs on - ### The privacy boundary set before you start The agent is a member only of the channels you name, and has no view of the team's private conversations or of channels it is not on. You approve the channel list in the workshop and change it yourself, with no ticket raised with us. Access limited to channels on an approved list, the team's private conversations out of reach Proof ## Do you work this way in Slack yourselves? Yes. The AI agent in our founders' agency is available where the team already talks, and that is the only proof we can show without asking a client for permission to publish their numbers. Below is the scale of use, together with what the number does NOT prove. Built for ourselves. These are not client deployments but implementations inside the founders' own companies. We say so plainly, because proof you cannot verify is not proof. - ![JustIdea Agency](https://justautomate.ai/klienci/justidea.png)![JustBot](https://justautomate.ai/agenci/justbot-80.webp)JustBottheir agent JustIdea Agency, a performance marketing agency of around 30 people 23 people from the company work with the agent 38 systems wired in permanently 112 processes run without a human 475 thousand lines of the agent’s code The agent is wired into the CRM, the task system, the advertising tools, Analytics, Search Console and the data warehouse, and is available to the team in the place where it works. First commit on 1 March 2026. 8 of 38 systems it works in - Google Ads - Meta Ads - HubSpot - ActiveCollab - Ahrefs - GA4 - Search Console - BigQuery Before Answering a question about data meant logging into one of several systems and building a view. In March 2026 four people used the agent, because asking a colleague was faster than asking the tool. After The question now goes to the agent where the team talks, and the answer comes back as a sentence. In August 2026 twenty three people used it, which is most of the agency. 64,903 operations in 30 days, 29,854 of them with no human involved, across 23 people using the agent And in your company ## What if our Slack looks different? Four sentences we hear most often, and how we answer them. - We already have the Slack app from our CRM vendor. If it is enough for you, we will say so in the workshop and there is no reason to build anything else. A native app shows records from one system and answers the questions its vendor anticipated. An agent earns its place when a single answer is assembled from two or three sources at once, for example from the CRM and the data warehouse. - We do not want anything reading our Slack conversations. The app sees only the channels you add it to, and has no access to the team's private conversations. You approve the channel list in the workshop and change it yourself. What goes to the model is the question put to the agent plus the data needed to answer it, not the contents of the channel, and we write that boundary down before the rollout, with the personal data in scope agreed with you under GDPR 2016/679. - A public channel is visible to the whole company, and some of our data is not for everyone. That is why for every kind of question we settle two things: the answer itself, and the place where it may be given. Sensitive numbers come out only in a direct message or on a private channel, and the public channel gets the answer with those fields removed. That is a decision from the workshop, not a default setting in a tool. - We do not use Slack, we are on Microsoft Teams. The mechanism is the same: the agent as an app in the place where the team talks, plus actions run from the message. We roll it out on Teams in exactly the same way. The choice of tool is something we settle in the workshop, not an assumption we arrive with. Team work Slack, Microsoft Teams, Google Workspace, Microsoft 365, Notion, ClickUp, Jira, Asana CRM HubSpot, Pipedrive, Salesforce, Zoho Sales and shop Shopify, WooCommerce, Magento, BigCommerce, Stripe Finance and accounting Xero, QuickBooks, Sage, sevDesk, Chargebee Data and analytics BigQuery, Google Sheets, GA4, SQL databases Automation layer Make, n8n, Zapier, custom API integrations ## What the team asks, where the answer comes from and where it may be given Each row shows one question put to the agent in Slack, together with its two consequences: where the data has to come from, and where the answer is allowed to appear. The third column matters most, because it decides whether an agent on a public channel is a convenience or a leak. | What the team asks | Where the agent gets the answer | Where it may be answered | | ------------------------------------------- | -------------------------------------------------- | ------------------------------------------------------------ | | What stage is this client case at | CRM: stage, owner, date of last contact | the client channel and the sales channel | | How many orders are waiting for payment | the order system and accounting | the customer service channel | | Has this invoice been paid | accounting and the bank account history | a direct message to a person with the permission | | Who on the team is available today | the calendar and the absence register | the team channel | | How is the campaign doing week on week | the data warehouse and the advertising systems | the marketing channel, once a day in a digest | | What is the margin on this client | the data warehouse and the profit and loss account | direct message only, to people on the list from the workshop | | What did we agree with the client in August | the client channel history and notes in the CRM | the client channel | A first rollout usually covers four to six kinds of question. The rest goes in once it turns out somebody actually asks them, rather than up front. How we work ## Five steps from a question going round the channel to an answer from company data Every step ends with something you can judge before you decide on the next one. 1. ### Workshop We list the questions the team asks each other in Slack, and check which system holds the answer and how long getting it takes today. We settle the other half of the same job right away: who may see what, and where it is allowed to be answered. 2. ### Access and action design The list of channels the agent is a member of, the data scope per role, the list of actions run from a message and a note of the ones that need a human to confirm. We pick the tools: a Slack app plus n8n, Make or an API integration with your systems. 3. ### Pilot on real data The agent joins one test channel with a few people and answers your real questions. We check every answer against the source system, because an agent that gets one question in ten wrong is worse than no agent: the team stops believing the answers that are right as well. 4. ### Production and training We move onto the real channels, and we separate the event acknowledgement from the answer, so Slack does not retry the call after three seconds. Somebody on your side learns to add the agent to a channel, take its access away and change the range of questions with no ticket raised with us. 5. ### Measurement at 30 and 90 days We look at three numbers: how many questions were asked, how many answers somebody disputed and which actions the team actually clicks. A kind of question nobody asks drops out of scope, and an action nobody uses disappears from the message. Before and after ## How a day on the channel changes ### Before - A question about a number goes round the channel until somebody logs into the system - A new person on the project asks three colleagues for the client history - Approving a document means opening a tool and hunting for the record - Sensitive numbers end up on a public channel, because that was easier ### After - A question about a number gets an answer from the system in seconds, in the same thread - A new person reads one client channel and asks the agent for the rest - Approval goes through a button under the message, with an entry in the log - A sensitive number arrives in a direct message, because that is what you agreed The first answers are visible on pilot day. We judge the range of questions and the list of actions after 30 days against what the team actually asked, and again after 90 days. For the person signing this off ## What the person accountable for data access gets out of it Access to data stops being a verbal arrangement. You see which channels the agent is a member of, what data scope it reaches for on behalf of which role, and who asked what. When somebody disputes a number shown on a channel, the log says which system it came from and who saw it. The second thing matters more in front of the board. You know how many questions the team put to the agent in a month, and how many of them were about data somebody used to walk into a system for. That is a number you can base a decision on, whether to run the next rollout or hold it, and today you do not have it. - A list of channels with agent access, changed on your side with no ticket raised with us - Data scope per role written down in the workshop, not buried in a tool's settings - A log of questions and actions: who, when, about what, with what result - A pilot on one channel, so backing out costs a single click What it costs ## The workshop has a published price, the rollout is quoted after it The process workshop costs **EUR 3,500 net per week**, and runs from one to seven weeks. Slack work usually needs one week, because the scope is narrow: the list of questions, the data scope per role and the list of actions. We quote the rollout after the workshop, once it is clear how many systems the answer has to come from and what state the data in them is in. You settle Slack directly with its vendor, on the price list of your own plan, and you pay the language model cost to the model provider, where a single query costs roughly EUR 0.005 to 0.07\. No margin from us. The same goes for n8n, Make or Zapier if they end up in the solution, or you pay nothing at all when n8n sits on your own infrastructure. - ### Process workshop EUR 3,500 net per week, and Slack work is usually one week A self-contained piece of work. It ends with a list of questions, a data scope per role and a channel map you could execute without us. - The list of questions the team asks on the channel today, with the source system for each one - Data scope per role, and a note of where each thing may be answered - A tool recommendation: the native Slack app, n8n, Make or an API integration - The metrics to measure at 30 days: number of questions and number of disputed answers Start with the workshop - Most popular ### Workshop and Slack rollout Quoted after the workshop scope follows from the number of systems the answer comes from The option most companies choose. A working agent in your Slack 2 to 6 weeks after the workshop, tested on your own real questions. - Everything in the workshop - A Slack app wired into the CRM, the task system and the warehouse through n8n, Make or the API - Answers on channels and in direct messages, with a data scope per role - Actions from the message: a button, a slash command, a form in a separate window - A pilot on a test channel, error monitoring, training for the person managing access Talk about a rollout - ### Slack and BOND, the company AI agent Quoted after the workshop a BOND rollout takes around 2 months For companies where questions about data run all day and cross every department, rather than sitting in one team. - Everything in the Slack rollout - BOND wired into the CRM, accounting, sales and the data warehouse - An answer from company data in 5 to 10 seconds, in English or in Polish - A Personal AI Agent for people who work on their own set of data - Daily and weekly digests written by the agent, not by a person Ask about the agent If the workshop shows that the native Slack app from your CRM vendor plus two commands is enough, we say so and show you how to set it up. The workshop ends with a document that stands on its own. And if it turns out there is nothing to automate, because the questions are rare or the data is incomplete, we say that too, instead of hunting for a scope to roll out. ## Frequently asked questions What is a Slack integration in practice? It is a Slack app wired into your systems, doing two things. It answers questions about company data asked on a channel or in a direct message, in 5 to 10 seconds and from your own sources. It also runs actions triggered from a message: a button, a slash command or a form in a separate window. That way the team is not switching between Slack and five other tools. Does an AI agent in Slack read the team's private conversations? No. The Slack app is a member only of the channels you add it to, and outside them it sees no messages. Private conversations between people stay out of its reach, because the agent is not a participant in them. You approve the list of channels with access in the workshop and change it yourself, with no ticket raised with us. Who sees the agent's answer on a public channel? Every member of that channel, which is why for each kind of question we settle up front where it may be answered. Sensitive numbers, for example margin on a client or personal data, come out only in a direct message or on a private channel. The public channel gets the answer without those fields, and the detail sits behind a link to the record, which needs permissions in the system anyway. How long does a Slack integration take? One week of workshop and 2 to 6 weeks of rollout, depending on how many systems the answer has to come from. One system and a few kinds of question is the lower end. A full company AI agent, meaning BOND wired into the whole business, is a different scale and around 2 months of work. How is this different from automatic Slack notifications? The direction of travel. A notification runs from the system to the person: something happened, so the channel gets a message, and that is what our pages on Slack notifications and on notification automation cover. Here the traffic runs the other way: a person asks or clicks, and the system answers and executes. The two work well together, but they are rolled out separately and quoted separately. Which actions can be run straight from a Slack message? The ones you can run through the source system's own interface: moving a case stage in the CRM, assigning a task, approving a document, adding a note, starting a workflow. Short actions go through a button, longer input we collect with a form in a separate window. Actions that cannot be undone, or that cost money, get an extra confirmation and a list of people allowed to trigger them. What if the agent gives a wrong number? That is why the pilot stage is compulsory: the agent answers on a test channel and we check every answer against the source system before it goes onto the team's channels. Each answer carries its source and the moment the data was pulled, so it can be verified in half a minute. After 30 days we look at how many answers somebody disputed, and narrow the scope where the data in the system is incomplete. Can the same thing be done on Microsoft Teams? Yes, the mechanism is identical: an app in the place where the team talks, access only to the channels you name, and actions run from a message. We roll it out on Teams exactly as we do on Slack. We settle the choice of tool in the workshop, because in some companies teams sit in both at once, and then what decides it is where the conversation about work actually happens. ## Let us find out what your team keeps asking about on the channel The free consultation takes 30 minutes and you write directly to Paweł, our founder. You tell us which channels you work on and what the team asks most often, even when the answer sits in a spreadsheet. We show you which questions the agent can answer from your systems straight away, and which ones need the data tidied up first. You do not have to share any systems at this stage. We run at most two rollouts at a time, so that someone is genuinely sitting on each of them. [Book a free consultation](https://justautomate.ai/en/contact/) **PS.** If anyone in your company wrote "can someone check how many of these we have?" on a channel this week, that is exactly the problem. The answer sits in a system the person asking has no habit of logging into, so the question goes round the channel for half a day and ends with a number typed in from memory. An agent on that same channel cuts it to a few seconds, and listing those questions is where the workshop starts. ## Related - [BOND, the company AI agent](https://justautomate.ai/en/services/bond-business-ai-agent/) - [Automatic Slack notifications](https://justautomate.ai/en/automations/automatic-slack-notifications/) - [Notification automation](https://justautomate.ai/en/services/notification-automation/) - [API integration](https://justautomate.ai/en/services/api-integration/) - [HubSpot integration](https://justautomate.ai/en/integrations/hubspot/) - [Process workshop](https://justautomate.ai/en/services/process-workshop/) --- **JustAutomate** · contact: contact@justautomate.ai · form: https://justautomate.ai/en/contact/