Six lessons from designing AI tools people actually use

Too often the question is ​“can we do something with AI?” instead of ​“what problem are we solving for users?” That sidelines everything we know about design thinking and user-centric design. AI should be an answer, not the starting point. Begin with the use case, then explore how AI can contribute.

1. Wait time as part of the experience

AI needs time to think. That sounds trivial, but it often determines whether someone feels a tool works well or not. When nothing seems to happen for a moment, it can create uncertainty for users. We’ve learned to design precisely those moments.

By providing feedback, showing small visual cues or briefly explaining what’s happening, users stay engaged rather than anxious. The experience feels calmer, more understandable and more reliable. People don’t mind waiting a bit, as long as they understand why they’re waiting and what’s happening behind the scenes.

In projects where AI pulls insights from multiple fragmented data sources, we learned that the problem isn’t the delay, it’s the silence. Instead of generic loading spinners, we experimented with:

• Micro-explanations (“Searching across employment history and training data…”)
• Progressive results instead of one final reveal
• Clear fallbacks when information couldn’t be retrieved

AI feels unreliable when it behaves like magic. It feels trustworthy when it behaves like a colleague thinking out loud.

2. Privacy starts with design

As soon as data plays a role, the question arises of what happens with that information. People want to know why their data is being used, how it’s being used and what say they have in it.

We don’t consider that concern only at the end, but from the very beginning. Privacy isn’t a technical requirement, it’s part of the experience itself. That’s why we make visible which data is being processed, how we secure it and when something deliberately remains anonymous.

By being open about this, the feeling that AI makes invisible decisions disappears. Transparency builds trust. And trust determines whether people actually use what you build. In other words people don’t just want secure systems, they want understandable logic.

That means designing for:

• Explainability: Why did I receive this recommendation?
• Control: Can I correct or influence the input?
• Boundaries: What does AI decide, and what remains human?

AI should not replace responsibility. It should clarify it.

3. Start from the data that already exists

The real value of AI isn’t in the model, but in the knowledge an organization already possesses. So every project begins with understanding the existing data and what it can mean for the user.

When you start from those existing strengths, you build solutions that feel familiar and align with the organization’s culture. A customer service AI that speaks like your team. A recommendation engine that understands your catalog’s quirks. That makes an AI application not just smarter, but more relevant. People recognize their own expertise in it, and that strengthens the sense of ownership.

AI works best when it amplifies existing expertise instead of overriding it. When the system reflects institutional knowledge and tone, it feels supportive. When it ignores that context, it feels threatening. Adoption isn’t a technical challenge. It’s a cultural one.

4. Think small, improve purposefully

Integrating AI works better in small steps than in big leaps. Take one part of a process, improve it with AI and observe the impact. That not only delivers faster results, it also makes the technology more understandable for the teams working with it.

Think of AI as a specialist, not a generalist, exceptional at specific tasks, not everything at once. By working step by step, knowledge and confidence within the team naturally grow along. Small improvements immediately show where the value lies, and that motivates further experimentation.

5. Build AI in where people already work

AI adoption only succeeds if people encounter it naturally. That means the technology is best built into the tools and systems that are already part of their daily work.

When an AI feature simply appears in a familiar environment, the threshold to try it is much lower. An auto-complete suggestion in the CRM they already use. A smart filter in their existing dashboard. At the same time, that creates space to pay extra attention to security and privacy. You can, for example, automatically filter sensitive information or make clear when a piece of text has been generated by AI.

This way, awareness doesn’t arise through separate trainings or communication campaigns, but during the work itself. People discover what AI does, what it means and where the boundaries are, simply by using it.

6. AI as a layer, not a destination

In strategic workshops, we’ve seen how easily AI becomes the centerpiece. A chatbot. A predictive dashboard. A bold AI roadmap. But the strongest concepts were never standalone AI features. They were layers. AI as a skin on top of complex systems. AI as a facilitator between fragmented data sources. AI as a translator between policy logic and human language. Instead of building something ​“AI-powered,” we asked: Where does complexity currently live? And can AI reduce cognitive load there? That’s where AI becomes meaningful. Not as spectacle, but as simplification.

The core

Designing for AI isn’t just about technology, but about the people who work with it. It requires insight into behavior, trust and context. By consciously designing for the experience, the data and the adoption, we make AI more understandable, more usable and more human.

And that’s when AI stops feeling like technology and starts feeling like a natural part of how work gets done.