Intermediate

Make.com AI Automation

Make.com (formerly Integromat) provides a visual scenario builder with native AI modules. Its strength lies in visual data flow, powerful data transformation, and a massive library of 1,800+ app integrations.

Why Make.com for AI?

• Visual data flow: See exactly how data moves through your automation with a clear, visual interface
• Native AI modules: Built-in modules for OpenAI, Anthropic, Google AI, and other LLM providers
• Data transformation: Powerful functions for manipulating data between AI steps
• Error handling: Built-in error routes, retry logic, and break/resume capabilities
• Scheduling: Flexible scheduling from every minute to custom cron expressions

Key AI Modules

Module	Capabilities	Best For
OpenAI	Chat, completions, embeddings, image generation, speech	GPT-powered text and image tasks
Anthropic	Messages API, tool use, vision	Complex reasoning, long documents
Google AI	Gemini models, multimodal	Vision tasks, multi-turn conversations
HTTP + JSON	Call any AI API directly	Custom models, self-hosted LLMs

Building an AI Scenario

Let us build a customer feedback analyzer that processes reviews and generates reports:

1. Watch for New Reviews

   Use a Google Sheets or Airtable module to watch for new rows containing customer feedback.

2. Analyze Sentiment

   Pass each review to the Anthropic module with a prompt that extracts sentiment (positive/negative/neutral), key themes, and urgency level.

3. Route by Sentiment

   Use a Router module to handle different sentiments: negative reviews go to a support queue, positive reviews go to marketing for testimonials.

4. Generate Weekly Report

   Use an Aggregator module to collect a week of analyzed reviews, then pass to an LLM to generate an executive summary with trends and recommendations.

Advanced Patterns

• Iterators + AI: Process arrays of items through AI individually, then aggregate results
• Webhooks + AI: Receive data via webhook, process with AI, and return the result synchronously
• Multi-model pipelines: Use a fast model for classification, then a powerful model for generation only when needed
• Data stores: Use Make.com data stores to cache AI results and build up context over time
• Error routes: When an AI call fails (rate limit, timeout), route to a retry queue with exponential backoff

Cost Optimization

Make.com charges per operation. AI API calls add additional costs. Optimize both:

• Filter before AI: Use filters to skip items that do not need AI processing
• Batch processing: Aggregate multiple items and process in a single AI call when possible
• Model selection: Use smaller, cheaper models for simple tasks (classification, extraction)
• Caching: Store AI results in a data store to avoid reprocessing identical inputs