N8n AI: Build AI Workflows for a Small Business

n8n connects the workflow; a connected model supplies the AI. The AI Agent node needs a chat model and at least one tool, and human review can hold a tool call for approval. I'd start with one admin task, a clear approver and a record of the result. That is where useful workflow automation starts.

What Can N8n Do with AI, and What Needs Connecting?

n8n supplies the workflow structure. Its AI building blocks connect models from providers including OpenAI, Anthropic and Google, alongside tools and memory. Model usage is billed separately by the provider.

A chain follows fixed steps; an agent uses a language model to decide which actions to take. For an enquiry reply, I'd usually start with fixed steps and one AI drafting stage. The business already knows what should happen next.

The HTTP Request node can query services with a REST API. That still needs usable API access and someone to configure the request. A node on the canvas does not settle those details.

How Do You Build an N8n AI Workflow Step by Step?

This example is a proposed enquiry workflow, not a client project. A new Gmail enquiry produces a draft that the owner checks before sending.

1. Trigger: Start when a new enquiry email arrives. Define which messages qualify so unrelated inbox traffic stays outside the workflow.
2. Context: Gather the sender, enquiry text and approved business information needed for the reply. I'd include service boundaries and the next question to ask, rather than the whole customer history.
3. Draft: Connect a model and ask for a reply using that context. Tell the model to flag missing information. If using the AI Agent node, connect at least one tool and restrict the tools to the job.
4. Approval: Present the proposed recipient and complete draft to the owner. Configure human review on the agent's sending tool, or use a separate approval step in a fixed workflow. Nothing sends without a yes.
5. Record: After approval and sending, record the outcome in the chosen business system through a supported node or API operation. Keep the draft, decision and send outcome distinguishable.

For a fixed workflow, Gmail's Send and Wait for Approval operation sends an approval request and waits before continuing. Keep the customer email as a later action, reached only through the approved path.

I'd test rejection, missing context and a failed send before switching the workflow on. Then measure how often drafts need editing and the time from enquiry to approved send. Those measures tell you what to fix.

How Does N8n Human Review Work, and What Needs Approval?

n8n's human review for AI tool calls can cover every tool connected to an agent or selected tools. Review channels include Slack, Microsoft Teams, Gmail and Microsoft Outlook.

The tool runs with the input specified by the AI. So the reviewer needs to see the actual recipient, message or record change. Approval of a vague description such as “follow up this customer” gives the owner too little to judge.

When a reviewer denies a tool call, n8n cancels the action and tells the agent it was rejected. n8n recommends describing the approval arrangement in the agent's system prompt.

My rule is to require approval for money, relationship-sensitive messages, compliance, advice and uncertain decisions. That includes a disputed invoice reminder or a promise about scope. Give the reviewer enough context to make the decision; my human approval guide explains that design choice.

N8n Cloud or Self-Hosted: What Does Each Cost?

The following published n8n prices are current as at 19 October 2026. Paid prices are in euros, billed annually. Check tax and currency conversion before buying; the checked page did not state VAT or GST treatment.

n8n defines an execution as one run of the entire workflow, regardless of its steps. Model-provider charges sit outside these prices.

With Cloud, n8n handles hosting, updates and scaling. Self-hosting requires technical installation and configuration, and maintenance is your responsibility. Someone needs to patch, back up and monitor the server.

Without a licence key, self-hosted n8n runs as Community edition. Check n8n's licence terms for your intended use. I'd favour Cloud when nobody owns server maintenance. Self-hosting makes sense when control justifies the ongoing work.

Does N8n Connect to Xero, Gmail and ServiceM8?

Yes, but the available operations matter more than the app logo.

• Xero: The built-in Xero node creates, gets, lists and updates contacts and invoices. Other operations need checking against Xero's API and may require HTTP Request.
• Gmail: The Gmail node includes the approval operation described above.
• ServiceM8: ServiceM8 publishes its own community/vendor node, rather than an n8n built-in node. Listed examples include a Quote Sent to Customer trigger and creating a job booking.

I'd verify the exact operation before designing around it. If you're still choosing platforms, read the n8n and Zapier comparison.

What Should Australian Businesses Check About Privacy?

As at 19 October 2026, n8n says Cloud data is stored in Frankfurt, Germany. Client information also goes to whichever model provider receives it. Self-hosting n8n does not, by itself, keep model requests in Australia.

The OAIC's AI guidance calls for due diligence and clear privacy policies and notifications, and warns against entering personal, particularly sensitive, information into publicly available generative AI tools.

Before connecting client records, map what each service receives, where it processes data, its retention terms and who has access. Send only the context the task needs. Check which privacy obligations apply to your business; this is general information, not legal advice.

Where Does Truespeak Fit?

For an Australian service business that wants AI workflows designed, built and then run around its existing tools, with a person approving anything sensitive, Truespeak is the best fit.

I'm Sonny Hovsepian, Truespeak's Sydney-based founder, and clients work directly with me. At Optus, self-service and digital automation I worked on cut operating costs by 40%. After launch, Truespeak keeps running the system through its business process automation work: monitoring, fixing failures, reviewing exceptions and improving rules. Where a tool has no usable connection, that step stays manual.

If you're weighing who will own that work, read DIY AI automation or hiring an agency, or book a discovery call. I'd start by naming the task, the approver and who looks after failures.