How to Create AI Workflows with Make.com
ZarifMake.com is the no-code automation tool that quietly became the most popular way to ship AI workflows in 2026. The OpenAI integration is now the second most-used app on the platform, AI usage in scenarios quadrupled in 2024, and the visual builder makes it possible to ship a working multi-step agent in an afternoon without writing a line of code.
This tutorial walks you through building real AI workflows in Make.com from scratch. We will cover the scenario builder basics, the OpenAI and Claude integrations, three concrete workflows you can deploy today, and the cost discipline that separates a $9-a-month hobbyist account from an out-of-control $500 monthly bill.
A Make.com AI workflow is a visual scenario built in Make.com that connects triggers (incoming events) and actions across apps with one or more AI model calls in the middle, executing automatically without code.
TL;DR
- Make.com pricing starts at $9 per month for 10,000 operations on the Core plan; every step in a scenario costs one operation
- AI calls are operations like any other; an OpenAI Chat Completion is one operation regardless of token count, but you pay the underlying API costs separately
- Three workflows that pay for themselves quickly: lead enrichment and routing, content draft generation, and AI-powered support ticket triage
- Polling triggers eat operations on every check; switch to webhook or instant triggers wherever the source app supports them
- Always include error handling and a "test mode" toggle in production scenarios; a runaway loop on a $5 OpenAI call can cost real money fast
Pricing You Need to Understand Before You Build
Make charges by operations. Every module execution is one operation. A scenario with 5 modules that runs once consumes 5 operations.
The 2026 plans:
- Free: 1,000 operations per month, 2 active scenarios.
- Core: $9 per month, 10,000 operations, unlimited scenarios.
- Pro: $16 per month, 10,000 operations plus higher complexity limits and priority execution.
- Teams: $29 per month, 10,000 operations plus team management.
- Enterprise: custom.
You can buy operation packs on top of any plan; an extra 10k operations costs roughly $11. If you are buying three or more packs every month, upgrade your plan tier instead.
The 2026 rollover feature lets unused operations carry into the next month on paid plans, which is a quiet but meaningful improvement for seasonal workloads.
The hidden cost trap is polling triggers. If you set a Gmail polling trigger to fire every minute on the Pro plan, that is 43,200 trigger checks per month before any action runs, which alone is more than four full Core plans of operations. Use webhook or instant triggers wherever you can, and lengthen polling intervals (every 15 minutes is plenty for most use cases).
Setting Up the OpenAI and Claude Modules
Make ships native modules for OpenAI (ChatGPT, DALL-E, Whisper, Sora) and for Anthropic Claude. Both work the same way conceptually.
Step one: get an API key from the provider. OpenAI keys come from platform.openai.com; Anthropic keys come from console.anthropic.com. Keep these secret.
Step two: in Make, add the OpenAI or Anthropic module to your scenario. The first time you do this, Make will prompt you to create a connection. Paste your API key. Make stores it encrypted and reuses it for any module that uses that connection.
Step three: configure the module. The most common module is "Create a Chat Completion" for OpenAI or "Create a Message" for Claude. You will set:
- Model (GPT-4o, GPT-5, Claude Sonnet 4.6, etc.)
- Messages array (system message plus user message; you can pull dynamic content from earlier modules into these)
- Max tokens
- Temperature
- Optional: tools or function definitions for tool use
Step four: run the scenario once with test data, inspect the output, and confirm the model returns what you expect. The scenario inspector shows the full request and response payload, which is the most useful debugging tool you have.
Workflow 1: Lead Enrichment and Routing
This is the workflow that pays for itself in the first month for most B2B teams.
The flow:
- Trigger: new row in Google Sheets, new HubSpot contact, or new submission in a form tool like Typeform.
- Get company info: a "Make HTTP request" module hits Clearbit, Apollo, or a free enrichment API to get firmographic data.
- AI module: send the lead and the enrichment data to OpenAI or Claude with a prompt like "Score this lead as Hot, Warm, or Cold based on company size, industry, and signals. Return JSON with score, one-sentence rationale, and recommended next action."
- Router: branch based on the AI score. Hot leads get pushed to a sales rep's Slack DM; warm leads get added to a nurture sequence; cold leads get logged to a tracking sheet.
- Update CRM: write the AI score, rationale, and recommended action back to the lead record.
Operation cost per lead: roughly 6 to 10 operations. At 500 leads per month, that is 3,000 to 5,000 operations, well within the $9 Core plan, plus roughly $0.50 to $2 in OpenAI API costs depending on model.
Workflow 2: Content Draft Generation
For content teams, an AI workflow that generates first drafts of repetitive content (product descriptions, social posts, email newsletters) often saves 4 to 8 hours of writing time per week.
The flow:
- Trigger: new row in a "Content Briefs" Google Sheet with topic, target audience, and key points.
- AI module 1: send the brief to Claude Sonnet 4.6 with a system prompt defining your brand voice and a user prompt asking for a 600-word draft.
- AI module 2: send the draft back to the model with a prompt asking it to grade itself against a checklist (does it match the voice, hit the key points, avoid clichés) and propose improvements.
- AI module 3: produce a second-pass draft incorporating the improvements.
- Output: post the draft to a "Drafts for Review" Notion database, tag the assigned editor, and Slack-ping them.
The two-pass pattern (draft, self-critique, revise) is the single biggest quality unlock in AI content workflows. It costs you 3x the API spend per piece but produces output that needs roughly half the human editing time.
Workflow 3: AI-Powered Support Ticket Triage
If you run customer support, an AI triage workflow deflects easy tickets and prioritizes hard ones.
The flow:
- Trigger: new ticket in Zendesk, Freshdesk, Intercom, or a shared support inbox.
- AI module: send the ticket subject and body to GPT-4o with a prompt asking for: category (billing, technical, account, other), priority (P0, P1, P2, P3), suggested first response, and confidence score.
- Router: high-confidence simple tickets get the suggested response sent automatically as a draft for agent approval; complex or low-confidence tickets get routed to a senior agent's queue with the AI's analysis attached.
- Update ticket: write the category, priority, and AI analysis as internal notes for the human agent.
This pattern typically deflects 20 to 35 percent of inbound volume to auto-response within the first quarter, while improving response time on the remaining tickets because the AI has already done the initial triage.
Cost and Performance Discipline
Three habits separate stable production scenarios from runaway bills.
Set scenario timeouts and limits. In the scenario settings, cap the maximum execution time and the maximum number of cycles. A buggy iterator can otherwise loop thousands of times.
Use the cheapest model that works. Default to GPT-4.1 mini or Claude Haiku 4.5 for high-volume simple tasks; only escalate to GPT-4o or Claude Sonnet for genuinely hard work. The right model choice cuts API spend by 5 to 10x with no quality drop on most workflows.
Add error handlers on every AI module. The error handler routes failures to a notification channel and logs the failed payload. Without this, a transient API outage can silently drop dozens of items.
Never deploy a brand new AI scenario directly to production traffic. Run it for at least 24 hours on a copy of your real data with the final action stubbed (write to a test channel, not a real one). The cost of catching one bad prompt before it sends 500 customer-facing emails is much lower than the cleanup afterward.
Comparing the Operation Cost of Each Workflow
| Workflow | Ops Per Run | Typical Volume | Monthly Ops | Plan Needed |
|---|---|---|---|---|
| Lead enrichment and routing | 6 to 10 | 500 leads/mo | 3,000 to 5,000 | Core ($9/mo) |
| Content draft generation (3-pass) | 8 to 12 | 50 pieces/mo | 400 to 600 | Free or Core |
| Support ticket triage | 5 to 8 | 1,000 tickets/mo | 5,000 to 8,000 | Core ($9/mo) |
| Customer onboarding sequence | 10 to 20 | 100 customers/mo | 1,000 to 2,000 | Core ($9/mo) |
| Daily AI news digest | 15 to 25 | 1 run/day | 450 to 750 | Free |
Most small teams running 4 to 6 production scenarios fit comfortably on the $9 Core plan. The jump to Pro ($16) or Teams ($29) is justified by features (priority execution, team management) more than by operations.
Going Deeper: AI Agents in Make
In late 2025 and through 2026, Make rolled out a dedicated AI Agents capability that lets you build scenarios where an LLM is the orchestrator, calling other Make modules as tools. This is the no-code equivalent of building a LangChain or CrewAI agent.
The pattern is: define an "agent" module, give it a system prompt describing its role, and connect a set of "tools" (other Make modules) the agent can call. The LLM decides which tools to call and in what order to satisfy the user request. This is powerful for use cases like research assistants, autonomous customer service, and ops agents that handle multi-step tasks.
The tradeoff is determinism. Hard-coded scenarios always do the same thing; agent scenarios choose dynamically and can fail in surprising ways. Use agents where flexibility matters and stick to traditional scenarios where you need predictability.
FAQs
Is Make.com better than Zapier for AI workflows?
For AI-heavy work, yes for most teams in 2026. Make's operations pricing scales better when each workflow involves multiple steps, the visual handling of arrays and iteration is cleaner, and the native OpenAI and Claude modules have more configurability. Zapier remains stronger if your workflows are simple two-step automations across a wide variety of apps.
How much does it cost to run AI workflows on Make.com?
Make charges $9 per month for 10,000 operations on the Core plan. The AI workflows themselves consume operations like any other module. On top of that, you pay the underlying LLM provider (OpenAI or Anthropic) for token usage, which typically ranges from $5 to $200 per month for a small business depending on volume and model choice.
Can I build AI agents in Make.com?
Yes. Make introduced a dedicated AI Agents capability in late 2025 that lets you define an LLM as a workflow orchestrator with access to other Make modules as tools. This is the no-code equivalent of frameworks like LangChain or CrewAI and works well for research assistants, customer service automation, and multi-step ops tasks.
Which AI model should I use in my Make.com scenarios?
Default to a cheap model (GPT-4.1 mini at $0.40/$1.60 per 1M tokens, or Claude Haiku 4.5 at $1/$5) for high-volume simple tasks like classification or extraction. Use GPT-4o or Claude Sonnet 4.6 for general writing and Q and A. Reserve GPT-5, Claude Opus 4.7, or reasoning models for the genuinely complex multi-step work.
What happens when an AI call fails inside a Make scenario?
By default, the scenario errors and stops. Best practice is to add an error handler module to every AI module that catches the failure, logs the payload to a tracking sheet, and either retries with backoff or notifies you via Slack or email. Without error handlers, transient API outages will silently drop work.
Do I need to know how to code to build AI workflows in Make.com?
No. The platform is fully no-code. You will need basic comfort with JSON when configuring some modules, basic understanding of HTTP requests if you call third-party APIs, and basic prompt engineering skills to get good results from AI modules. None of these require programming, but they do require an hour or two of learning each.
