AI automation: a guide for companies and experts.

In this guide I answer the questions companies and experts ask most often about AI automation. My method is described on the Lean AI automation page; questions specific to factories are on AI agents in manufacturing.

What is AI automation, and how is it different from classic automation?

Classic automation follows fixed rules, such as "when a form arrives, send this email". AI automation uses a model that can read and understand text: it sorts an incoming inquiry, writes a fitting reply draft and summarizes a long email thread.

The difference is that it can also take on tasks that change a little every time and cannot be written as rules. That is why it needs to be set up with human approval.

Which tasks can an expert or boutique firm hand over to AI?

The best candidates repeat every week, cannot be fully written as rules, and have results that are easy to check:

• a first reply draft to an inquiry and a follow-up reminder
• a proposal or meeting summary draft from call notes
• a LinkedIn post draft from a case or an observation from the field
• updating the client tracking sheet
• preparing weekly reports and summaries

What is the difference between an AI agent and automation?

Automation follows a flow drawn in advance. An agent decides which step to take toward a goal: it first checks what information it needs, then prepares the draft. For experts and small firms, a mix usually works best: the flow is fixed, and the writing and summarizing inside it is done by the agent.

Who makes the decisions? What if the agent makes a mistake?

In every setup I build, the agent drafts and a person approves. No email, proposal or post goes to a client without your approval.

Agents can make mistakes, so in the first weeks we check the outputs together and adjust the instructions to your language and your rules.

Is my data safe? What about data protection rules?

Before setup, we write down which data goes to which tool. The agent only gets the information it needs for its task; passwords and credentials are never written into prompts. Accounts are opened in your name and you manage access.

If personal data is processed, you need a privacy notice and, where required, consent; transfers to service providers abroad are included. For a legal assessment, I recommend asking a lawyer.

Which industries and departments does it work for?

Anywhere with repetitive text, document and follow-up work. Examples from my own experience:

• automotive OEMs and suppliers: 8D, FMEA, APQP and PPAP documentation, supplier follow-up
• textiles and apparel: order and shipment tracking, quality reports
• advanced tech, IoT and medical devices: technical files, CE and GDPR documentation
• robotics and heavy industry: shift and downtime summaries, maintenance records
• services and consulting: inquiry follow-up, proposal and content drafts

Which tools are used: n8n, Make, Zapier or Power Automate?

I start with the tools you already have: email, calendar, Google Workspace or Microsoft 365, your ERP and your client tracking system if there is one. The goal is not to tie you to a new platform but to place an understandable helper inside the way you already work. The choice depends on the job:

• n8n: can run on your own server; a good option when data must stay in-house.
• Make and Zapier: many ready-made integrations; practical for a quick start and small teams.
• Microsoft Power Automate: fits companies already on Microsoft 365.
• A model API and custom scripts: when deeper integration with systems such as ERP or MES is needed.

Why someone with a lean manufacturing background?

Because an AI agent speeds up the flow you give it. If the flow contains an unnecessary approval or a double entry, it speeds that up too. So I simplify the process and measure the waste first, then automate. I describe this on the Lean AI automation page.

Having worked across production, quality, R&D, ERP and regulation, I can see from the start how a process that begins in one department touches others.

What do GDPR and the EU AI Act mean for automation?

GDPR applies to any automation that processes personal data in the EU (and KVKK in Türkiye): you need a written record of which data goes to which tool and, where required, consent.

The EU AI Act classifies AI systems by risk level and phases in obligations over time. Internal productivity agents that draft and leave the decision to a person are assessed very differently from systems used in hiring, credit scoring or product safety. With my experience preparing CE and GDPR technical files for the EU market, I take this distinction into account from the start; for a legal assessment, I recommend working with a lawyer.

Why do brand positioning and AI automation go together?

An agent multiplies the message you give it. If it is not clear who you help and with what problem, more content and faster replies just mean more confusion.

So positioning and evidence come first, then automation: once your one-sentence positioning, case studies and references are clear, the agent uses the same language consistently in every draft.

How long does it take, and what does it cost?

There are no fixed prices on the site; I send the scope, timeline and price in writing after the first conversation. The work scales to your budget.

We usually start with a small pilot on one repeating task. If the pilot works, we move on to the next one.

Where should I start?

Pick one task that takes the most of your time every week and whose result is easy to check. If your positioning or online presence is not clear yet, it is more effective to start there.

• see where you stand with the free Visibility Check
• choose the first task with the post AI agents for experts
• talk directly through the contact form

Does our data need to be perfect before we start?

No. But for each task we check four gates:

• Access: can the data be read from a system, or does it live in a file on someone's laptop?
• Permissions: who can see what? The agent only sees documents it is authorised for. Permissions are hard to add later, so they are set up from the start.
• Lineage: can you trace which document or record an output is based on?
• Quality: the accuracy you need depends on the task. A draft email and an invoice amount do not need the same threshold.
• For each task the result is one of three: ready, conditionally ready (a small fix first), or not ready yet.

How does my team learn to use the system?

As part of the setup, people get short, hands-on training based on their role. Your team works on its own tasks rather than listening to a generic presentation:

• A short session for managers: what has been automated, where the limits are, how results are tracked.
• A workshop for the team using it: checking, editing and approving drafts, and reporting problems.
• An internal owner: one of your staff who knows the system well and can make small adjustments. This is the step that really reduces dependency on me.
• What stays with you: a prompt library, usage rules and a short operating guide.
• Measuring the effect: a short knowledge check before and after the training, and after 30 to 60 days, whether the system is actually being used.

I set this up in my own work first.

At INVAMAR I set up a way of working with AI agents in finance, HR, operations and business development; I describe it, with its evidence level, on the Case studies page. I use the same approach for this site's content and follow-up work.