Business process automation with AI hands repetitive tasks that always follow the same steps to a system: reading emails and documents, extracting data, drafting replies, updating your management software, alerting the right person. AI adds the part that used to need a person: understanding freely written text. Automations that work start from a clear process, leave approval of sensitive steps to a person and send an alert when something gets stuck.
I’m Toure Tia Kevin, a developer in Bassano del Grappa, Italy. I build automations by writing custom scripts with Claude, without subscription automation platforms.
Which processes to automate
| Process | What the automation does | Who checks |
|---|---|---|
| Customer enquiries | reads emails and forms, sorts them, drafts the reply | a person approves sending |
| Orders and documents | extracts data from PDFs and emails and enters it in your management software | automatic checks plus spot checks |
| Quotes | fills in the template from the customer’s details | the owner reviews prices and terms |
| Content | turns articles and products into social posts | approval before publishing |
| Reports | gathers the numbers and sends a weekly summary | whoever reads it |
| Monitoring | checks sites and servers and alerts you if something’s wrong | action only when there’s a problem |
A good candidate is a task you do often, always the same way, that today means copying information from one program to another.
Examples from my projects
- BitcoinLive24: articles published automatically and news turned into posts on X.
- Arreda & Risparmia: a workflow publishes pins to Pinterest every day from the site’s content.
- Servers: scripts monitor servers and send an alert when a service stops responding.
- Company documents: I built a platform that answers questions about a company’s documents.
The social media case is covered in the guide to automating social media posts with AI.
How to build an automation you can rely on
- Map the process: who does what, with which data, and which exceptions come up.
- Choose where AI helps: only in the steps that need text to be understood. The rest is done with ordinary rules, which are cheaper and more predictable.
- Add approval where a mistake is costly: messages to customers, prices, payments.
- Plan alerts: if something gets stuck, someone needs to know straight away.
- Measure: time saved, errors, API costs.
Custom scripts or platforms like Make and Zapier?
Automation platforms are handy for connecting two services in a few minutes, but they charge per operation and the fee grows with use. As the workflow gets more complex, it also gets harder to see where it broke.
A script written for your process does only what’s needed, runs on your server or mine, and has no fee that keeps growing. Writing it with Claude keeps development quick, and the code is yours.
What it costs
In the AI and automation service, an initial consultation or a first automation starts at €500. On top of that come the AI API costs, which stay low with the right models: prices are in how much AI costs for a business.
Data and rules
- I use providers’ APIs and business versions, which don’t use customer data to train the models.
- I choose tools and settings based on the data involved, GDPR included.
- If the automation talks to the public, such as a chatbot, the EU AI Act requires you to say so.
Frequently asked questions
How long until the first automation?
For a simple workflow, such as sorting emails, from a few days to a couple of weeks, testing included.
Does it work for a very small business?
Yes, if there’s a task that comes up often. Even saving an hour a day on a tedious job is worth the investment.
What happens if an automation makes a mistake?
That’s what approvals and alerts are for: a mistake stops before it reaches the customer, and the problem shows up straight away instead of quietly piling up.
If you want to find out which of your processes can be automated, get in touch.
