AI & Workflow Automation

Custom automation built in n8n, Make.com, Zapier and the Claude API. The repetitive work in your business, running without you touching it, with alerts when something breaks.

Agencies quote automation projects at $3,000 to $15,000 and staff them with a project manager, a strategist and a junior building the actual workflow. You get the same workflow either way. Here you get it from the person who builds it, at $10 per hour.

What I automate

Lead capture and follow-up

Form or ad to CRM to tagged sequence to notification, with nothing retyped by hand. Speed-to-lead matters more than almost anything else in local services: the first business to respond usually wins the job. I have built lead-capture pipelines in GoHighLevel for Linkup Advanced Scaffolding in Australia, where the owner is on site rather than watching an inbox.

Order, fulfilment and support workflows

Order to fulfilment to tracking to customer notification. Refunds and returns routed to the right person. Support tickets classified and triaged before a human opens them. For ecommerce this is usually where the hours actually go, not in marketing.

Data movement between systems

The spreadsheet somebody exports every Monday and pastes into another tool. That is a workflow, and it is costing you a morning a week. Sheets, CRMs, accounting (Xero, MYOB, QuickBooks), ecommerce platforms and internal databases, kept in sync on a schedule or a webhook.

Document and content processing

Invoices read into structured rows. Long documents reduced to the three fields you actually need. Drafting, classifying and summarising with the Claude or OpenAI API, wired into the place the output has to land.

Reporting that builds itself

The weekly numbers pulled, calculated and delivered to Slack or email on a schedule, instead of someone rebuilding the same spreadsheet every Monday.

n8n, Make.com or Zapier?

This is the question most automation pages avoid, because agencies have a platform they resell. I do not.

n8n when you want to self-host, keep data inside your own infrastructure, or run high volume without per-operation pricing. It handles branching logic and custom code nodes properly, and I can host it on your server so the running cost is a VPS rather than a subscription that scales with your success.

Make.com when you want it managed, the volume is moderate, and your team needs to see the scenario visually without learning anything new. The visual builder is genuinely the best of the three.

Zapier when the integration you need exists there and nowhere else, or when your team already lives in it. It is the most expensive per operation and the least capable at branching, but the integration library is unmatched.

Plain code on a cron when the workflow is simple and stable. Not everything needs an automation platform and a monthly fee. I will tell you when that is the case, which is more often than an agency selling retainers will admit.

Where AI actually earns its place

Most “AI automation” being sold right now is a normal workflow with a language model bolted on so it can be priced higher. Being straight about it: if the logic is deterministic, plain rules are cheaper, faster and more reliable. A model that is right 95% of the time is worse than an if-statement that is right always.

Models earn their place on messy input. Classifying a support email by intent. Extracting fields from an invoice that arrives in a different layout every time. Drafting a first-pass reply a human then edits. Summarising a long thread into the three things that matter. Deciding which of 40 leads to call first.

So a workflow I build usually has a small model call inside a mostly deterministic pipeline, not the other way round. That keeps it cheap to run and predictable to debug.

Why automation projects fail

They break silently. A credential expires, an API deprecates a field, a webhook stops firing, and nobody notices for three weeks because the automation never announced it was dead. By the time it surfaces, you have a month of missing data to reconcile.

So every workflow I build gets four things that rarely appear in a quote:

Error handling on every external call. Anything that touches another system can fail, and the workflow has to know what to do when it does.

Retry with backoff. Most API failures are transient. Retrying three times over a minute fixes the majority without anyone being told.

A failure notification that reaches a person. Not a log entry nobody reads. An email or a Slack message saying which workflow failed and on what.

Run logging. So when you ask what happened to Tuesday’s orders, there is an answer rather than a guess.

This is the boring half, it is roughly a third of the build time, and it is the reason the automation is still working next year.

What this costs

My rate is $10 per hour. Most single workflows are a small fixed-price build. Larger multi-system automation gets scoped and quoted up front so the number does not move unless the scope does.

For comparison, agencies typically quote $3,000 to $15,000 for a project of this shape, plus a monthly retainer. You pay for a project manager, a strategist and the junior who builds it. Here there is one person, and it is the one building it.

Platform costs are yours and paid directly: an n8n VPS runs a few dollars a month, Make.com and Zapier are priced per operation.

How a build runs

Map the process first. I want the current steps written down, including the exceptions people handle by memory. That is usually where the real complexity is hiding.

Pick the platform after, not before. The process decides the tool.

Build, then break it on purpose. I test the failure paths, not just the happy path. What happens when the API is down, the field is empty, the same webhook fires twice.

Hand over documented. A diagram of what runs when, where the credentials live, and what to do when an alert arrives. Your automation should not be hostage to me still answering messages.

Questions people ask

Do I need AI in my automation?
Often no, and I will say so. Rules are cheaper and more reliable for anything with a definite answer.

Can you self-host n8n for me?
Yes, on DigitalOcean or your existing server, with SSL, backups and monitoring. See hosting and server management.

Will it connect to my existing tools?
If it has an API or webhooks, yes. If it does not, I can usually build a bridge. See API and integrations.

What happens when it breaks?
You get a notification, not a silence. Fixes fall under maintenance or are billed hourly.

Can you fix an automation someone else built?
Yes. I audit what is there and tell you honestly whether it is worth repairing or rebuilding.

How is this cheaper than an agency?
Because you are not paying for account management, a sales process, or an office. The workflow is the same. See the GoHighLevel pipeline work for what that looks like in practice.

Start a project

Tell me the process you repeat most often. That is almost always the first thing worth automating.