( Services )
AI Automation for Business
I move repetitive work onto n8n, Make and language models: invoices, email, reports, content publishing. My own business runs on more than a dozen workflows like that.
AI automation means moving repetitive work, such as retyping invoices, sorting email, building reports and publishing content, onto tools like n8n or Make, with a language model added where text has to be understood. Automating a single process starts at $1,400 with me. Start with the one process that takes the most time, not with choosing a tool.
What I have automated in my own business
Before I suggest an automation to anyone, I run it myself. My self-hosted n8n server runs more than a dozen workflows that support four apps and the business itself:
- Supplier invoices: they arrive by email, and a workflow pulls them out of Gmail and enters them into my accounting system (Infakt).
- E-book sales: Stripe payment, email with the file, signup to the mailing list. I am not involved at any step.
- Social media videos: a workflow generates short video ads for my apps and publishes them to YouTube and Instagram.
- Daily report and alerts: a summary waits in my inbox every morning, and if one of the apps starts throwing errors, I get notified.
- AI trends digest: every day at 6:45 a model reads Hacker News, Hugging Face and a set of RSS feeds and picks out what is worth reading.
- Blog drafts: an AI agent writes them and I get an email with a "Publish" button. Nothing reaches the site until I click it.
That last one is an AI agent rather than a plain automation; I explain the difference on a separate page. For n8n itself, including installing it on your own server, see n8n implementation.
When automation pays off and when it does not
It pays off when a task is repetitive, rule-based and frequent enough that the time saved outweighs the cost of the build. Routine enquiries, moving data between systems, processing documents, generating reports: these usually qualify.
It does not pay off when the process looks different every time, or needs judgement you would not hand to a new hire. Automating exceptions costs several times more than automating the rule.
A simple test: count the hours a month the task takes and multiply by the hourly cost of the person doing it. If the build pays for itself within a year, it is worth a conversation.
What automation vendors tend not to mention
Tool subscriptions are a separate cost. Make, Zapier, model access: all of it is billed outside the build. I give you those numbers up front so there is no surprise in month three.
Automation needs maintenance. A provider changes its API, a form gains a field, customers start phrasing things differently, and something breaks. That is why every workflow I build reports its own errors instead of silently stopping.
Not everything needs AI. Plenty of processes can be automated with a plain rule, more cheaply and more reliably. I add a language model only where the content genuinely has to be understood.
Data and GDPR
Before we start, we decide whether your data may go to an external model provider. If not, n8n can be self-hosted and nothing leaves your infrastructure. If it may, you need a data processing agreement. This choice shapes the architecture and the cost, so it is made at the beginning. I am based in the EU and work under GDPR by default, which matters if your customers are in Europe too.
What I automate most often
Documents and email
Invoices from your inbox go straight into the accounting system, enquiries get sorted and come with a draft reply, attachments land in the right folder.
Data flow
A Stripe payment adds the customer to your mailing list, a form submission goes to the CRM, an order goes to a spreadsheet. Nobody retypes anything.
Reports and alerts
A daily sales report in your inbox at 7 a.m., a phone alert when something stops working, a digest of trends pulled from several sources.
Content and social media
Short videos generated and published to YouTube and Instagram, blog drafts that a human approves before they go live, newsletters.
Handling enquiries
A language model classifies the message, pulls out the data and prepares a reply for you to approve or send.
Sales
A lead from your form goes straight into the CRM, scored, with a follow-up reminder. A digital product sale: payment, email with the file, mailing list signup.
How I build an automation
-
01
Pick the process
We find the one task that takes the most hours and follows repeatable rules.
-
02
Map the steps
I decide where a plain rule is enough, where an AI model is needed and where a person has to decide.
-
03
Build and test
I build the workflow and run it on real data alongside the manual process.
-
04
Monitoring
Every automation reports its own errors by email or to your phone instead of quietly stopping.
How much does AI automation cost?
Process automation
from $1,400
- Quoted per process, not as a flat package
- What drives the price: number of systems involved
- Whether off-the-shelf tools (n8n, Make) are enough
- Whether custom logic is required
- Tool subscriptions and AI model usage billed separately
Tools I use
- n8n
- Make (make.com)
- OpenAI
- Anthropic Claude
- Google Gemini
- Perplexity
- Webhooks and REST APIs
- Gmail, Google Sheets
- Stripe, accounting software
- WordPress
Frequently asked questions
A single process built on existing tools (n8n, Make) starts at $1,400 as a one-off fee. On top of that come tool subscriptions and AI model usage, which for a typical process runs from around $10 to a couple of hundred dollars a month. With n8n on your own server the tool subscription disappears. Larger builds with custom logic across several systems I quote after we have mapped the process.
The question is less about the tool and more about whether you can trust it with your data. For process automation I mostly use n8n, which can run on your own server so data stays in your infrastructure, and Make when fast delivery matters more. For language tasks I use models via API under a data processing agreement. The tool is chosen for the specific process and your GDPR requirements, not from a popularity list.
Start by naming one repetitive task that eats the most time, not by picking a tool. Write it out step by step, mark where a decision genuinely needs a human and where a rule or a model is enough, then automate the narrowest slice first. The most common mistake is trying to automate the whole process at once.
There is no single best one, only tools that fit a given scale. n8n works well when you want control over your data and to host it yourself. Make is quicker for simple integrations between popular services. Zapier is often the most convenient for straightforward flows, but the most expensive once you run a lot of tasks. The tool follows from the process, not the other way round.
RPA (robotic process automation) imitates a person clicking through a program's interface and only works when the data always has the same format. AI automation adds a language model that understands content: it can read an email written in someone's own words, pull data from an invoice in any layout, suggest a reply. In practice I combine both: rules where they are enough, a model where text has to be understood.