How I Automated Business Processes with AI Agents: A Step-by-Step Guide Using n8n and Make.com

## How I Automated Business Processes with AI Agents: A Step-by-Step Guide Using n8n and Make.com Two years ago, my business hit a ceiling. We processed 50–70 orders per day, and a team of three managers spent 70% of their time on routine tasks: transferring data from email to CRM, answering repetitive customer questions, and negotiating prices. I tried classic RPA (UiPath, Python scripts) — the entry barrier was high, and the logic broke with the slightest interface change. Everything changed when I implemented AI agents. Not just "smart bots," but a combination of n8n (self-hosted workflow automation), Make.com (formerly Integromat), and GPT/Claude APIs. Today, I'm sharing a real case: how we automated incoming order processing and cut routine time by 80%. ### The Problem: Manual Order Processing Killed Scaling Our process looked like this: 1. Customer submits a request via a website form or email. 2. Manager manually copies data into CRM (amoCRM). 3. Checks product availability in 1C. 4. Creates a commercial proposal (even for standard orders). 5. Sends the customer a response with an invoice. Average time: 45 minutes per order. Errors: mixed up SKUs, lost emails, forgot discounts for regular customers. I wanted AI to do 80% of this work, with humans only confirming. ### The Solution: AI Agents on n8n + Make.com I built a two-tier system: | Tier | Tool | Task | |------|------|------| | Orchestration | n8n (self-hosted) | Logic management, API calls, error handling | | Integration | Make.com | Connection to 1C, amoCRM, Telegram, email | | AI Agent | GPT-4o / Claude 3.5 | Text analysis, response generation, classification | | RPA (if needed) | Playwright (via n8n) | Browser clicks for legacy systems | Why not just one tool? Make.com excels at integrations with hundreds of services but is weak in complex logic and error handling. n8n is the brain: it handles branching, loops, external API calls, and restarting failed scenarios. ### Step 1: Collecting Requests and Initial Processing **Scenario in Make.com:** 1. **Trigger:** New email to support@company.com (IMAP module). 2. **Filter:** If subject contains "Order" or "Invoice" — proceed. Otherwise, route to manual manager. 3. **Parsing:** Extract from email: name, phone, email, product list (via regex). 4. **Send to n8n:** Pass JSON object with data via Webhook. ### Step 2: AI Agent Analyzes and Classifies **Scenario in n8n:** Upon receiving data, I call **GPT-4o API** with this prompt: > "You are an AI agent for order processing. Analyze the customer request. Determine: order type (new/repeat), urgency (normal/urgent), whether a discount is needed (if customer in database >6 months — 5%). Return JSON with fields: order_type, urgency, discount, product_list_clean." The output is structured data. Previously, a manager did this in 10 minutes; now AI does it in 5 seconds. ### Step 3: Stock Check and Response Generation 1. **n8n** sends a request to 1C (via Make.com HTTP module) for product availability. 2. If all in stock — **AI agent (Claude 3.5)** generates a commercial proposal based on a template, inserting prices and discount. 3. If out of stock — AI suggests an alternative (e.g., "Product A is out of stock, but there's an analog B at price X"). ### Step 4: Sending and CRM Recording Final scenario: - **Make.com** creates a deal in amoCRM with fields: amount, status, tags. - **Telegram bot** (via n8n) sends manager notification: "✅ Order #123: client Ivanov, amount 45,000 RUB. Requires confirmation." - **Email module** sends customer an invoice with a "Pay" button. ### Results After 3 Months | Metric | Before | After | |--------|--------|-------| | Time per order | 45 min | 8 min | | Data entry errors | 12% | <1% | | Max orders per day | 70 | 200+ | | AI API costs | 0 | $150/month | The most unexpected benefit — **AI agents stopped "forgetting" discounts**. Previously, managers saved time and ignored the loyalty system. Now Claude automatically checks customer history. ### Practical Takeaways (No Fluff) 1. **Don't try to automate everything at once.** Start with one process — e.g., incoming email processing. Build an MVP in 2 days with Make.com + GPT API. 2. **n8n is a must-have for complex logic.** If you plan branching, loops, retry mechanisms — Make.com won't suffice. n8n gives full control, especially if self-hosted. 3. **AI agent ≠ ChatGPT.** Use custom prompts with JSON output. This turns the language model into a structured "module" for business. 4. **RPA is a last resort.** If a system has an API (1C, CRM) — use it. Browser automation (Playwright, UiPath) only for legacy software without integrations. ### What's Next? We're now implementing **CrewAI** for a multi-agent system: one agent checks orders, another negotiates with suppliers, a third generates reports. This isn't science fiction — it's Python code running in the same n8n. If you want to learn how to build such a system from scratch — without deep programming, with ready-made templates and support — check out my AI automation course page. There, I provide step-by-step scenarios for Make.com, n8n, and AI agents (Autogen, CrewAI, LangChain) that can be adapted to any business. Automation isn't about firing people. It's about letting your team focus on strategy, not copying data from email to CRM. Start today.

How I Automated Business Processes with AI Agents: A Step-by-Step Guide Using n8n and Make.com

Comments

Recent articles

Building AI-Powered Automation Agents with n8n and Python: A 2026 Guide

Financial Data APIs in 2026: Moscow Exchange, Central Bank of Russia, and FRED — Which Source Is More Accurate and Faster for the Russian Market?

How to Build AI-Powered Micro-SaaS Products with Vibe Coding in 2026

AI Agents for Business Process Automation in 2026: A Step-by-Step Guide with n8n and Make.com

How to Set Up a CI/CD Pipeline with GitHub Actions and Docker: A Step-by-Step Guide 2026

Топ-10 AI-агентов для автоматизации бизнеса в 2026 году: обзор работающих сервисов (EN)

How to Build an AI-Powered Code Generator with Vibe Coding and Python: A 2026 Guide

How to Automate Business Processes with AI Agents in 2026: A Step-by-Step Guide

How to Build an AI Agent for Automated Customer Support in 2026: A No-Code Blueprint for Startups