Why customer support automation is the #1 AI opportunity for UK SMBs
If you run a small or medium business in the UK, there's a good chance your team spends a significant chunk of every working day answering the same questions. Password resets. Pricing queries. Order updates. Refund requests.
These aren't complex problems. They're repetitive ones — and repetitive problems are exactly what AI is built to solve.
In this guide, we'll walk through exactly how to build an AI customer support automation system using Claude AI and n8n — the same stack we use for our clients at MS IT Solutions, based in St Albans, Hertfordshire.
What you'll build
By the end of this guide, you'll have a working system that:
- Reads every inbound support email automatically
- Classifies it by type (billing, technical, refund, general)
- Generates an intelligent, personalised reply using your knowledge base
- Auto-sends high-confidence replies
- Routes low-confidence or sensitive messages to a human
- Logs everything to a dashboard
The tools you'll need
Claude AI (Anthropic) — the AI model that reads, classifies and writes the responses. We use Claude Sonnet for the best balance of performance and cost.
n8n — the workflow automation platform that connects everything together. Think of it as the plumbing between your email, Claude, your CRM and your logging system.
Airtable or Google Sheets — for storing your knowledge base and logging tickets.
Gmail or SMTP — for sending and receiving emails.
Step 1: Build your knowledge base
Before you write a single workflow, you need to document your business's knowledge. This is what Claude will use to answer questions accurately.
Create a Google Sheet or Airtable base with columns: Category, Question, Answer. Add your top 20–30 FAQs. Include your refund policy, pricing, product descriptions and escalation rules.
The more detailed your knowledge base, the more accurate Claude's responses will be.
Step 2: Set up n8n and connect Claude
Sign up for n8n cloud (around £20/month) or self-host via Docker. Add your Anthropic API key as a credential. Create a new workflow and start with a Gmail trigger node set to poll your inbox every minute.
Step 3: Add the classification node
Add a Claude node and configure a system prompt that asks it to classify the incoming message into one of your categories (BILLING, TECHNICAL, REFUND, GENERAL, ESCALATE) and return a confidence score from 1–10 in JSON format.
Step 4: Add the response generation node
Add a second Claude node that takes the classification, the original message and your knowledge base, and writes a complete reply email — using your company's tone of voice.
Step 5: Add confidence routing
Use an IF node to split: confidence ≥ 7 sends the reply automatically. Confidence < 7 or category = ESCALATE sends a Slack alert to your team for manual review.
Results you can expect
Our clients typically see 60–80% of tickets resolved automatically within the first month, with response times dropping from hours to under 60 seconds.
If you'd rather have us build this for you, book a free strategy call and we'll have you live within two weeks.