How to analyze Slack message sentiment with Power Automate

My flows > + New flow > Automated cloud flow

Create new automated flow

Go to make.powerautomate.com and sign in with your Microsoft account. Click 'My flows' in the left sidebar, then the blue '+ New flow' button. Select 'Automated cloud flow' from the dropdown menu. Name your flow 'Slack Sentiment Analysis' and search for 'Slack' in the trigger search box.

1. 1Click 'My flows' in the left sidebar
2. 2Click the blue '+ New flow' button
3. 3Select 'Automated cloud flow'
4. 4Enter 'Slack Sentiment Analysis' as the flow name
5. 5Search for 'Slack' in the trigger picker

Trigger > Slack > When a message is posted to a channel

Configure Slack message trigger

Select 'When a message is posted to a channel' from the Slack triggers. Power Automate will prompt you to create a connection to Slack. Click 'Sign in' and authorize the connection in the popup window. After connecting, you'll see a dropdown for Team and Channel. Select your workspace and choose the #feedback channel you want to monitor.

1. 1Click 'When a message is posted to a channel'
2. 2Click 'Sign in' to connect Slack
3. 3Authorize Power Automate in the Slack popup
4. 4Select your Slack workspace from Team dropdown
5. 5Choose '#feedback' from Channel dropdown

New step > Control > Condition

Add condition to filter bot messages

Click '+ New step' below the Slack trigger. Search for 'Condition' and select the Control connector's Condition action. In the left value box, click 'Add dynamic content' and select 'User ID' from the Slack trigger. Set the condition to 'is not equal to' and leave the right value empty. This filters out system messages that have null User IDs.

1. 1Click '+ New step' below the trigger
2. 2Search for 'Condition' and select it
3. 3Click the left value box and select 'User ID' from dynamic content
4. 4Set dropdown to 'is not equal to'
5. 5Leave the right value box empty

Condition Yes > Add an action > HTTP

Add OpenAI GPT action

In the 'Yes' branch of the condition, click 'Add an action'. Search for 'OpenAI' and you'll see the HTTP connector since there's no native OpenAI connector. Select 'HTTP' and choose 'HTTP' action. Set the method to POST and the URI to 'https://api.openai.com/v1/chat/completions'. We'll configure headers and body in the next steps.

1. 1Click 'Add an action' in the Yes branch
2. 2Search for 'HTTP' and select the HTTP connector
3. 3Choose the 'HTTP' action
4. 4Set Method dropdown to 'POST'
5. 5Enter 'https://api.openai.com/v1/chat/completions' in URI field

HTTP action > Show advanced options > Headers

Configure OpenAI authentication headers

In the HTTP action, click 'Show advanced options'. In the Headers section, add a new header by typing 'Authorization' in the key field and 'Bearer YOUR_API_KEY' in the value field (replace with your actual OpenAI API key). Add another header with key 'Content-Type' and value 'application/json'. Your API key should start with 'sk-' if it's valid.

1. 1Click 'Show advanced options' in the HTTP action
2. 2Click 'Enter key' in Headers section
3. 3Type 'Authorization' as the key
4. 4Type 'Bearer sk-YOUR_API_KEY' as the value
5. 5Add another header with key 'Content-Type' and value 'application/json'

HTTP action > Body

Build OpenAI request body

In the Body field of the HTTP action, you need to construct a JSON payload for the OpenAI API. Click in the Body field and build this JSON structure. Set the model to 'gpt-3.5-turbo' for cost efficiency. Create a system message that instructs GPT to classify sentiment, then include the actual Slack message text as the user message. Use dynamic content to insert the 'Text' field from the Slack trigger.

1. 1Click in the Body field
2. 2Start typing the JSON structure with model and messages
3. 3Set system message with sentiment analysis instructions
4. 4Click 'Add dynamic content' for the user message
5. 5Select 'Text' from the Slack trigger dynamic content

Add action > Data Operation > Parse JSON

Parse OpenAI API response

Add another action below the HTTP step. Search for 'Parse JSON' and select it from the Data Operation connector. In the Content field, select 'body' from the HTTP action's dynamic content. You'll need to provide a schema that matches OpenAI's response structure. The key field you want is choices[0].message.content which contains the sentiment classification.

1. 1Click 'Add an action' below the HTTP step
2. 2Search for 'Parse JSON' and select it
3. 3Click Content field and select 'body' from HTTP dynamic content
4. 4Click 'Generate from sample' to create schema
5. 5Paste a sample OpenAI API response to generate the schema

Add action > Slack > Post message

Post sentiment result back to Slack

Add a final Slack action to post the sentiment analysis result. Search for Slack and select 'Post message' action. Choose the same channel or a different channel like #sentiment-alerts. In the message text, combine the original message text with the parsed sentiment result. Format it clearly so your team can quickly see both the original feedback and its classification.

1. 1Click 'Add an action' after Parse JSON
2. 2Search for 'Slack' and select 'Post message'
3. 3Select your target channel from the dropdown
4. 4Click in Message field and add dynamic content
5. 5Combine original Text and parsed sentiment result

Save > Test in Slack channel > Check run history

Test the complete flow

Save your flow by clicking the 'Save' button in the top toolbar. Post a test message in your monitored Slack channel - try something clearly positive like 'Great job on the new feature!' The flow should trigger automatically, call OpenAI's API, and post the sentiment result back to Slack. Check the run history to debug any failures.

1. 1Click 'Save' in the top toolbar
2. 2Go to your monitored Slack channel
3. 3Post a test message with clear sentiment
4. 4Wait 30-60 seconds for the flow to complete
5. 5Check run history in Power Automate for success/errors

Add this expression in the HTTP Body field to create a more robust JSON payload with proper escaping and token limits for the OpenAI API call.

{
  "model": "gpt-3.5-turbo",
  "messages": [

{
  "model": "gpt-3.5-turbo",
  "messages": [
    {
      "role": "system",
      "content": "Classify the sentiment of user feedback as exactly one word: positive, negative, or neutral. Only respond with that single word."
    },
    {
      "role": "user",
      "content": "@{replace(replace(triggerBody()?['text'], '"', '\\"'), char(10), ' ')}"
    }
  ],
  "max_tokens": 10,
  "temperature": 0.3
}