First AI Steps Without an IT Team
You do not need to be a tech company to use Artificial Intelligence in a meaningful way. In this tutorial we walk you through five concrete steps – with tools you can try out this afternoon. No coding. No IT department. With realistic expectations.
Last week I sat down with Markus, who runs a three-person painting business near Bensheim. Tradesman, not a tech person. He showed me how he saves around five hours a week with ChatGPT: quote text, customer emails, short descriptions for his Google profile, a bit of translation for his Polish colleagues. No training course, no consultancy. He just started.
That is the truth about AI without an IT team in 2026: getting started has become trivial. What really matters is which tools you try in which order – and which pitfalls you avoid along the way. That is exactly what we cover here.
Do not just read. Try each step right away. Three out of these five steps you can implement in under 15 minutes.
Step 1 — Start with ChatGPT or Claude
If you have never worked with a chat AI tool before, start here. ChatGPT by OpenAI and Claude by Anthropic are the two best all-rounders. Both have free tiers, both have paid tiers for around 20 to 23 dollars a month, both run in your browser. You do not have to install anything.
What you do with them in week one:
- Draft emails. Paste the rough content, describe the tone ("polite but firm") and ask for three variants. In 30 seconds you have what used to take 20 minutes.
- Improve text. Quotes, web pages, LinkedIn posts. Just paste, type "make this clearer and shorter" – done.
- Translations. Significantly better than Google Translate, because you can pass context ("this is for a technical client, slightly formal").
- Brainstorming. "Give me 15 ideas for social media posts about summer heating maintenance." You will get ten weak ideas and five good ones – but the ten weak ones show you what you do not want.
- Explain-this-please. Let it summarise complicated technical documents, funding notices or tax letters in plain language.
One important rule from second one: do not type customer data, banking details or trade secrets into the free version. More on that below.
Step 2 — Bring in your own data
Once you have done step one for two weeks, you will notice: the tool does not know your company. It does not know what you sell, how you sound, what your standard answers are. That is where the next leap comes in.
Google NotebookLM (free)
My favourite tip for the beginning. NotebookLM is free, runs in your browser, and lets you upload up to 50 PDFs, websites or Google Docs. After that, the tool answers exclusively based on those sources. Perfect for: terms-and-conditions research, manuals, training material, funding catalogues. It even cites the passage where the answer comes from.
Microsoft Copilot (inside Microsoft 365)
If your company already runs Microsoft 365, Copilot is the obvious choice. It can read your Outlook emails, Word documents and SharePoint files – all within your tenant. The privacy setup is serious; pricing is around 22 dollars per user per month on top of your Microsoft 365 license. For many small businesses the most direct route to an assistant that thinks about their company.
ChatGPT Custom GPTs or Claude Projects
With a paid ChatGPT plan you can build a "Custom GPT": your own assistant with custom instructions and reference documents. Real-world example: a tax advisor in Heppenheim has a Custom GPT trained on their standard client replies. Staff get a draft response in ten seconds, then approve it.
Step 3 — Simple automation with Zapier or Make
Now it gets interesting: connecting AI with your existing tools without writing a single line of code. Zapier and Make (formerly Integromat) are drag-and-drop platforms that link your apps together. Both have free entry tiers, both offer AI building blocks.
Realistic examples for a small business:
- New email in inbox → AI summarises it → summary appears in Slack or Microsoft Teams.
- New entry from your website contact form → AI checks whether it is a serious lead → if yes it lands directly in your CRM, if spam it gets archived.
- New invoice as a PDF attachment → AI extracts amount and vendor → data lands in a Google Sheet for your bookkeeping.
- Voice note in WhatsApp Business → AI transcribes and translates → text goes to your ops team.
Expectation management: the first two workflows take about half an hour to an hour each to build. After that they run on their own. Plan two afternoons for your first real Zapier or Make workflow, not two hours – learning curve included.
Step 4 — A chatbot for your website
By the time you keep getting the same questions through your contact form, a chatbot pays off. It does not have to be an 8,000-euro project. There are off-the-shelf solutions you set up yourself:
- Tidio – free tier available, paid plans from around 25 dollars a month. Solid for standard questions, easy to train with your FAQ.
- Crisp – popular with smaller online shops, clean dashboard, AI add-on bookable separately.
- Cogswell Chatbot – our own solution for small businesses. You hand us your website and a few documents, we build the bot around it. Fair flat fee, no lock-in, GDPR-compliant with hosting in Germany.
Important: a chatbot never answers 100 percent of all enquiries. Realistically, 40 to 70 percent are automatable – the rest still needs a human. That is not a bug, that is the correct expectation. Anyone promising you 99 percent is lying.
Step 5 — When an external partner makes sense
You can do the first four steps on your own. But once one of these points applies, getting outside help is worth it:
- You want to connect AI to your ERP or accounting system (DATEV, SAP Business One, Sage, Lexware Office).
- You are in a regulated industry – medical practice, tax firm, law firm, financial services.
- You process personal data at scale and need clean GDPR and EU AI Act documentation.
- You want a voice agent on the phone that books appointments or handles enquiries independently.
- Your workflows have become so complex that Zapier or Make hit their limits and maintenance turns into a chore.
This is exactly where we step in at Cogswell IT. We do not build the AI solution over your head – we implement what you no longer feel comfortable doing yourself, and leave you to handle the rest. Fair, transparent, no multi-year maintenance lock-ins.
Common pitfalls
GDPR and data security
Do not put customer names, banking details, medical records or any personal data into free AI tools. In free tiers your inputs typically become training data. For serious use, take ChatGPT Team, Claude for Work, Microsoft Copilot or an EU-hosted variant – with a Data Processing Agreement in place.
Hallucinations
AI sometimes invents facts. Confidently, in complete sentences, with citations that do not exist. Do not let AI take important business decisions unchecked. For research, always verify with a second source. For internal text, hallucinations are less critical; for legal or medical content, they are a no-go.
Over-expectation
"We want an AI agent that replaces the entire back office." Not realistic in 2026. AI is a great co-pilot, not an autonomous employee. Plan for 20 to 40 percent time savings on individual tasks in your first 12 months – not full replacement.
Tool graveyard
Better to use three tools well than to pay for twelve licenses nobody opens. Start with ChatGPT or Claude, give each tool two weeks, evaluate honestly, then move on to the next.
Frequently asked questions
Do I need programming skills to start with AI?
No. Tools like ChatGPT, Claude, Google NotebookLM, Microsoft Copilot, Zapier or Make are entirely no-code. If you can write an email, you can use AI. Programming only becomes relevant when you want to train custom models or integrate deeply with existing systems – and that is usually where a service partner like us comes in.
What does it cost to get started with AI for a small business?
You can seriously start for under 50 euros a month: ChatGPT Plus or Claude Pro for 20 to 23 euros, plus optionally Zapier Starter from 20 euros. Website chatbots like Tidio or Crisp have free tiers. Your biggest lever in the first weeks comes without any additional license – just the paid tier of one chat tool.
Is using ChatGPT in a company GDPR-compliant?
With caveats, yes. The free ChatGPT tier should not be used for customer or employee data. ChatGPT Team, Microsoft Copilot, or EU-hosted variants of Claude and Mistral are GDPR-suitable if you sign a Data Processing Agreement and anonymise sensitive data. Pasting personal data into a free tool is never ok – regardless of which tool.
Stuck after step 3?
We build what you can no longer do yourself – fast, fair, no lock-in.
Book a first call →