"We want to build something with AI” – what does that actually mean?

In an article from earlier this year, we explored the trend that the huge interest in AI is starting to settle into something more practical. The focus has begun to shift—from testing tools like ChatGPT to integrating AI into your own processes, systems, and products. And this is where we’re starting to see the real value of AI.

This is also where many start asking new questions. Because the idea “we want to do something with AI” often says very little. Behind it can be very different ambitions—from testing something on a small scale, to automating time-consuming processes, or building AI features into a product used by real customers.

Projects can vary widely—both in scope and in what they actually require. That’s also why the next question is often difficult to answer: “What would something like that cost?” The answer doesn’t depend on AI itself, but on the problem you’re actually trying to solve.

So what do people really mean?

Today, many have tried AI for writing text, structuring data, generating images, and much more.

But when you start talking about implementing AI for real, it almost always comes down to something more concrete:

• something that is currently done manually
• something that takes an unreasonable amount of time
• or something that costs more than it should
  

In reality, it’s often about processes that don’t scale. Work that requires manual handling. Or tasks that take time away from the right people. That’s where AI actually becomes interesting—as a tool to solve existing problems.

How to get started the right way

What determines whether an AI project succeeds or not is almost never the technology. It’s how well the problem is defined. Many start in the wrong place—in the possibilities, the tools, what AI can do. But it’s really about identifying the core of a problem. Making it concrete enough to build something around—and valuable enough to justify the investment.

• What exactly needs to be improved?
• Where does friction occur today?
• What is “good enough” in practice?

This is also where most projects fall apart—not because the AI is poor, but because the problem isn’t defined well enough. The projects that actually work almost always have one thing in common:

• de börjar smalt
• Ett tydligt use case
• En avgränsad del av en process
• Något som går att testa, mäta och förbättra

This is where AI becomes a tool you can actually use—not just an idea, but something that creates real value.