How to analyze Slack sentiment with OpenAI using Pipedream

Pipedream > Workflows > New

Create new Pipedream workflow

Head to pipedream.com and click New Workflow. You'll land on the workflow builder with an empty canvas. The trigger step will be pre-selected at the top. Click on it to configure your Slack connection.

1. 1Click 'New Workflow' from your dashboard
2. 2Click the trigger step labeled 'Select a trigger'
3. 3Search for 'Slack' in the app list
4. 4Choose 'New Message Posted to Channel'

Trigger Configuration > Connect Account

Connect your Slack workspace

Click Connect Account to authorize Pipedream with your Slack workspace. You'll be redirected to Slack's OAuth page. Make sure you're logged into the correct workspace before authorizing. Pipedream needs read access to channels and message content.

1. 1Click 'Connect Account' button
2. 2Select your Slack workspace from the dropdown
3. 3Click 'Allow' on the permission screen
4. 4Wait for the green 'Connected' confirmation

Trigger Configuration > Channel Selection

Configure channel monitoring

Select your feedback channel from the Channel dropdown. This list pulls all channels where the Pipedream bot has access. If you don't see your target channel, go to Slack and type /invite @Pipedream in that channel first. Leave the other filters empty to capture all messages.

1. 1Click the Channel dropdown menu
2. 2Search for your feedback channel name
3. 3Select the channel from the list
4. 4Leave 'Username' and 'Bot' filters empty

Trigger Configuration > Deploy & Test

Test the Slack trigger

Click Deploy at the bottom of the trigger step. Then go to your Slack channel and post a test message. Return to Pipedream and click the refresh icon next to Test Event. You should see your test message data populate in the preview panel.

1. 1Click 'Deploy' to activate the trigger
2. 2Go to Slack and post a message in your monitored channel
3. 3Return to Pipedream and click the refresh icon
4. 4Click 'Use This Event' if the message appears correctly

Workflow Builder > Add Step > OpenAI

Add OpenAI step

Click the + button below your trigger to add a new step. Search for OpenAI in the app directory and select 'Chat' as the action type. This gives you access to GPT-3.5 and GPT-4 models for text analysis.

1. 1Click the '+' button below the trigger step
2. 2Type 'OpenAI' in the search box
3. 3Select 'OpenAI' from the apps list
4. 4Choose 'Chat' from the actions dropdown

OpenAI Step > Connect Account

Connect OpenAI API

Click Connect Account and paste your OpenAI API key. Get this from platform.openai.com under API Keys. Create a new secret key if you don't have one. The key starts with 'sk-' and is about 50 characters long.

1. 1Click 'Connect Account' in the OpenAI step
2. 2Open platform.openai.com/api-keys in a new tab
3. 3Click 'Create new secret key' and copy the key
4. 4Paste the API key into Pipedream's field

OpenAI Step > Model & Messages

Configure GPT model and prompt

Set the model to 'gpt-3.5-turbo' for speed and cost efficiency. In the Messages array, add a system message with your sentiment analysis instructions and a user message containing the Slack message text. Use the dynamic data selector to insert the message content from step 1.

1. 1Set Model to 'gpt-3.5-turbo'
2. 2Click '+ Add Message' and set role to 'system'
3. 3Add your system prompt in the content field
4. 4Click '+ Add Message' again, set role to 'user'
5. 5Insert Slack message text using the data selector

OpenAI Step > System Message Content

Write sentiment analysis prompt

Your system message should give GPT clear instructions for sentiment classification. Be specific about the output format you want. Ask for a simple response like 'Positive', 'Negative', or 'Neutral' plus a brief reason. Keep the prompt under 100 words to minimize token usage.

1. 1Click in the system message content field
2. 2Type: 'Analyze the sentiment of this customer feedback message'
3. 3Add: 'Respond with only: Positive, Negative, or Neutral'
4. 4Add: 'Then add a dash and brief 3-word explanation'

Workflow Builder > Add Step > Slack

Add Slack response step

Add another step and select Slack again. Choose 'Send Message to Channel' or 'Add Reaction to Message' depending on how you want to display results. For this workflow, we'll post the sentiment analysis as a threaded reply to the original message.

1. 1Click '+' below the OpenAI step
2. 2Search for and select Slack
3. 3Choose 'Send Message to Channel' action
4. 4Set Channel to the same feedback channel

Slack Response > Message Configuration

Configure sentiment response

In the Message Text field, build your response using data from previous steps. Include the original message author, the sentiment classification from OpenAI, and maybe add an emoji for quick visual scanning. Use the thread_ts field to make this a reply rather than a new message.

1. 1Click in the Message Text field
2. 2Add static text like 'Sentiment:'
3. 3Insert OpenAI response using the data picker
4. 4Set Thread Timestamp to the original message's ts value
5. 5Test the message format in the preview

Workflow > Deploy & Test

Test full workflow

Deploy the entire workflow by clicking Deploy at the top. Go to your Slack channel and post a clearly positive or negative message. Watch for the sentiment analysis to appear as a threaded reply within 5-10 seconds. Check the workflow execution log if nothing appears.

1. 1Click 'Deploy' at the top right
2. 2Go to Slack and post 'This new feature is amazing!'
3. 3Wait 10 seconds for the analysis reply
4. 4Check the workflow Logs tab if no response appears
5. 5Test with negative feedback too

Add this Node.js code step between OpenAI and Slack to track sentiment trends and only alert on concerning patterns instead of every message.

export default defineComponent({
  async run({ steps, $ }) {
    const sentiment = steps.openai.choices[0].message.content;

export default defineComponent({
  async run({ steps, $ }) {
    const sentiment = steps.openai.choices[0].message.content;
    const messageText = steps.trigger.event.text;
    const userId = steps.trigger.event.user;
    
    // Parse GPT response
    const isNegative = sentiment.toLowerCase().includes('negative');
    const isPositive = sentiment.toLowerCase().includes('positive');
    
    // Get recent sentiment history from workflow state
    const recentSentiments = $.service.db.get('recent_sentiments') || [];
    
    // Add current sentiment with timestamp
    const newEntry = {
      sentiment: isNegative ? -1 : (isPositive ? 1 : 0),
      timestamp: Date.now(),
      user: userId,
      text: messageText.substring(0, 100)
    };
    
    recentSentiments.push(newEntry);
    
    // Keep only last 20 messages
    const recent = recentSentiments.slice(-20);
    await $.service.db.set('recent_sentiments', recent);
    
    // Calculate trend - percentage negative in last 10 messages
    const lastTen = recent.slice(-10);
    const negativeCount = lastTen.filter(item => item.sentiment === -1).length;
    const negativePercentage = (negativeCount / lastTen.length) * 100;
    
    // Only proceed to Slack if concerning trend (>40% negative)
    if (negativePercentage > 40) {
      return {
        shouldAlert: true,
        trendAlert: `🚨 ${negativePercentage}% of recent feedback is negative`,
        currentSentiment: sentiment,
        recentNegative: negativeCount
      };
    }
    
    // Stop workflow execution for normal sentiment
    $.flow.exit('Normal sentiment trend - no alert needed');
  }
});