The biggest barrier to AI automation in small and medium-sized businesses is not technology. It is the belief that "we need an IT team for this". This notion is outdated. In 2026, there are tools that anyone who can use Excel and send emails can set up.
Here is a specific guide on how to do it.
5 most common myths about AI implementation
No-code AI tools: What to use
Here are the key tools that require no programming:
4-week implementation plan
Here is a concrete plan for the first AI automation in a company without an IT department:
Where do you spend time on routine tasks?
- Review an average working day and estimate how many hours are spent on repetitive tasks.
- Select one process with the highest time burden and lowest creative value.
- Document the process: what is the input, what is the output, and what are the exceptions.
- Estimate current time costs (hours/week × rate)
Select the right stack and prepare the environment
- Set up accounts in Make.com (or Zapier/n8n) and OpenAI API
- Prepare sample data, 10-20 real examples of inputs and desired outputs
- Write a basic AI prompt and test it manually in ChatGPT.
- Identify exceptions and decide how to process them (manually/skip).
Build your automation and test it
- Build a workflow in Make.com or n8n (visually, no coding required).
- Test on sample data and adjust the prompt and workflow based on the results.
- Set up notifications for exceptions and errors
- Launch "shadow mode" - automation runs in parallel with the manual process; compare results.
Switch to live operation and start measuring.
- Launch automation in live operation and switch off or significantly reduce manual processes.
- Monitor: number of processed items, number of exceptions, time saved
- Collect feedback from your team
- Plan your second automation based on experience from the first one.
What to measure: Key success metrics
Without measurement, you cannot tell if automation is working. Track these metrics from week one:
- Number of items processed automatically vs. Total count (target: 80%+)
- Number of exceptions/errors requiring manual intervention (target: under 15%)
- Hours saved per week - measure for 2 weeks before and after launch
- Team satisfaction - short questionnaire after 4 weeks (1 question: "Is it working well?")
Team training: Before launching automation, we recommend investing in training for key personnel. BeeSkills.cz offers practical AI workshops for companies, your team will understand the basics of AI and be ready to collaborate with automations from day one.
When to call in an expert
You cannot handle everything yourself, and that is fine. Here is the decision table for when it makes sense to call an AI specialist:
| Situation | Can you manage on your own? | When to call an expert |
|---|---|---|
| Email sorting and replying | Yes (Make.com) | If you require CRM integration or multi-level approval workflows |
| Invoice processing (PDF) | Basic version yes | Integration with Pohoda / Money S3 requires API knowledge. |
| AI chatbot for websites | Yes with Voiceflow / Botpress | Connection to your own order database = expert |
| Automatic reports | Yes (Make + GPT) | If data are in non-standard formats or across multiple systems |
| Comprehensive workflow with conditions | Recommended expert | From the start, save time and money |
Practical rule: If you spend more than 2 days setting up a single process and it still does not work, choose an expert. The cost of professional implementation is typically significantly lower than the value of your time lost on debugging.
Let us take the first step for you.
flylight.ai will carry out a free analysis of your processes and show where to start. Specific recommendations with no obligation, in 5 minutes.
Start free analysis →