Making sense of the EU AI Act with smart tools

The upcoming EU AI Act might seem like another regulatory hurdle for your business, but it's really an opportunity. By understanding how the Act classifies AI tools and using practical AI features, you can streamline compliance and improve how your business runs.

Why this matters for SMEs

This new regulation impacts more than just your IT department; your operations teams, legal counsel, and even marketing will need to understand its implications. Getting ahead means avoiding potential fines and building customer trust, both of which are critical for staying competitive.

What it looks like in practice

Take Orla, the Operations Manager at 'Celtic Crafts', a growing Irish online retailer with 60 staff. Celtic Crafts uses an AI-powered chatbot on its website to handle customer service queries and recommend products based on browsing history.

Before: Orla was worried about the EU AI Act's implications for their chatbot. She'd heard about 'high-risk' categories and feared a big, costly audit. She wasn't sure if their chatbot, which often collected customer preferences, fell into a category that needed extensive oversight. The legal team had flagged concerns, but concrete steps felt overwhelming.

After: Artellis helped Orla understand that their customer service chatbot likely falls into a 'limited risk' category under the Act, meaning it primarily requires clear transparency. They implemented a simple AI-powered tracking tool that monitors the chatbot's performance and logs how it responds to different types of queries. This tool automatically flags any unusual behaviour or responses that might suggest bias or incorrect information. Now, when a customer asks a question, the chatbot explicitly states it's an AI and provides a human contact option. The tracking tool also generates a weekly summary, showing Orla how often the chatbot suggests products and if it's consistently offering a balanced range, ensuring fairness. This not only meets transparency requirements but also helps Orla spot areas for improving customer service, such as frequently asked questions the AI struggles with.

What could go wrong

• Misinterpreting the Act: Incorrectly categorising your AI system could lead to insufficient or excessive compliance efforts, wasting resources or leaving you exposed.
• Over-relying on automated tools: While AI can help, human oversight is still essential to interpret results and make informed decisions, especially for ethical considerations.
• Ignoring data quality: AI systems are only as good as the data they're trained on. Poor or biased data will lead to compliance issues, even with the best tools.
• Viewing it as a burden: Approaching compliance only as a tick-box exercise rather than an opportunity for improvement can hinder innovation and trust.

What to try this week

1. Identify your AI tools: List all the AI tools your business currently uses or plans to use, from chatbots to predictive analytics. Don't forget off-the-shelf software with AI features.
2. Estimate their risk category: For each tool, based on whether it impacts fundamental rights or safety, make a rough guess if it's 'minimal', 'limited', or potentially 'high' risk. The Artellis website has a simpler guide to help with this.
3. Check vendor compliance: If you use third-party AI tools, ask your providers about their plans for EU AI Act compliance. It's important to understand their roadmap.

Approaching the EU AI Act with a practical mindset can turn a regulatory challenge into a strategic advantage for your business.