Where should an SME start with AI?

When I start building an AI system, the first question isn’t whether to use Gemini, ChatGPT, or some other AI model; it’s which tasks are actually worth building a system for. Choosing the tool is really only the third step.

Plan first, review the workflows, and only then build the tool

Before introducing any software or service, it’s worth clarifying the basics:

• Which tasks take up most of your team’s time, and where do deadlines matter most?
• Which of these repeat weekly or daily, following roughly the same steps?
• Can we optimize the current manual process? AI can work from past patterns, but the more structured the process we hand over, the faster and more effective the result.

Once you’ve answered these questions, you’ll have a list of tasks worth automating. That’s the point where development can actually begin.

Why not start with the biggest challenge?

Because the first project is still about building trust. People need to get used to working with AI. Once they experience how much an AI agent can help with their day-to-day work, they’ll start asking for more complex solutions on their own.

The best first project is one where the result is quick and tangible.

What can AI realistically do for an SME today?

What I use in my own business:

• company emails answered autonomously in trained scenarios,
• personal assistance: organizing daily tasks, managing the calendar,
• the full payroll process,
• structured data extracted from documents (invoices, forms),
• internal reporting agents that answer questions in plain language,
• website development and maintenance,
• content production.

The list could really go on forever (the only real limits are imagination and budget), but what matters most is that these areas can be freely combined. If your business offers a service, one obvious task is tracking competitors, another is adjusting prices to the market. How would I set this up in an AI system? For example: an agent checks competitor price ranges at set intervals, and if it judges it worthwhile, proposes a price increase with concrete figures, sent to you in a report. If you approve the proposal, that agent passes the new prices to the developer agent, who automatically publishes them on the company website. So there’s no need for long back-and-forth with people in between.

The order I’d recommend

1. Assessment: a conversation about your processes, ranked by return.
2. Introduction: automating a single, well-measurable process within a fixed budget.
3. Extension: whatever proves itself in the introduction phase gets expanded further.

If you are curious which process would be worth starting with in your company, request the free assessment.