AI AGENT AUTOMATION · N8N
A multi-tool AI agent that answers verified business questions, manages CRM leads, prevents duplicate records, escalates sensitive requests to a human team, and books appointments in Google Calendar.
PROJECT OVERVIEW
Customer-facing teams often repeat the same answers, manually copy lead details into a CRM, check whether records already exist, escalate sensitive issues by hand, and switch between tools to schedule calls.
That creates slow response times and makes it easy for information to become duplicated, incomplete, or lost between systems.
I built an AI agent in n8n that can decide which business tool to use based on the conversation.
The agent retrieves approved knowledge, checks and updates CRM records, creates new leads when needed, escalates requests to Slack, checks calendar availability, and creates confirmed appointments.
STAGE 01
The first part of the system handles conversation context, verified answers, lead capture, duplicate prevention, and CRM updates.
The workflow begins with n8n chat and routes every message through an AI Agent connected to an OpenAI model and memory.
The agent can choose between CRM tools, a business knowledge base, human handoff, Slack notifications, calendar availability, and appointment creation depending on what the customer needs.
The agent uses a dedicated knowledge table for company-specific information such as services, pricing, support, implementation timelines, and consultation details.
This keeps business answers grounded in approved information instead of allowing the model to invent policies or pricing.
When a prospect provides qualified sales information, the agent checks the CRM first and then creates a structured lead record only if the email does not already exist.
Details such as name, company, need, budget, timeline, intent, lead status, and AI summary are captured automatically from the conversation.
Returning prospects are matched by email before any new record is created. If the lead already exists, the agent updates the current record instead of inserting a duplicate.
In this test, the existing lead's budget and launch timeline were changed while the original CRM identity remained intact.
STAGE 02
When the AI should not resolve an issue on its own, the workflow creates a structured handoff and alerts a real support channel.
Requests involving refunds, disputes, sensitive issues, or an explicit request for a human are escalated instead of being handled entirely by the model.
The Human_Handoff tool creates a structured escalation record and the agent only confirms escalation after the tool succeeds.
After the escalation is created, n8n sends a formatted message to the support-alerts Slack channel.
The notification includes priority, reason, customer message, a concise summary, and the chat session ID so a human can quickly understand what requires attention.
STAGE 03
The sales side of the agent can check availability first and create a real calendar event only after the customer confirms the appointment.
Before booking anything, the agent sends the requested start and end time to Google Calendar and checks whether the slot is free.
If the time is available, the agent repeats the exact date, time, timezone, and duration to the customer and asks for confirmation.
Once the customer confirms, the Create_Appointment tool creates the Google Calendar event using the approved time and customer details.
The event includes the customer as an attendee and carries useful meeting context from the sales conversation.
The final test shows the agent coordinating multiple tools in one system: conversation handling, CRM lookup and update, knowledge retrieval, human escalation, Slack, availability checking, and appointment booking.
The result is a single conversational interface connected to real operational actions rather than a chatbot that only generates text.
BUSINESS IMPACT
Company-specific responses come from an approved knowledge source instead of unsupported model guesses.
Email-based lookup prevents duplicate records while existing leads can be updated automatically.
Sensitive requests can move from AI to a human support channel with structured context attached.
The same conversation can move from support to sales to calendar booking without manual copying between systems.
TECH STACK
LET'S WORK TOGETHER
I'm available for AI automation projects, n8n workflow builds, support automation, CRM systems, and scheduling integrations.
Send me an emailOr contact me directly at consuegradenrey [at] gmail [dot] com