The Problem with AI Use Cases

Bombardment from everywhere. I don’t know where up and down is, let alone what to do next. No manual exists for such a novel situation. No checklist. Nada.

They just keep detonating. Round after round, with no end in sight. For the first time in my life, I feel outgunned, and so does my team.

It’s May of 2023, the ChatGPT-craze just hit our offices, and I’m working for the AI team. My employer is an insurer, not the most digital native industry. You might be able to imagine what this means: artillery fire from every department inside the organization – the ammunition being the most dreaded of all working in innovation and IT: proposals for “AI use cases”.

Digging myself deeper into my foxhole, I knew we had to get a grip of the situation somehow. My team lead was a great guy and was already working on a solution. We knew we couldn’t win the war in a day, but maybe stopping them in their tracks would give us time to reposition. For that, their attention needed to be somewhere else. We decided on using a tactic as old humankind (and now the most profitable business model apparently): distraction.

A colleague and I were tasked to create an explanation and instruction manual about large language models. Our goal was to make it accessible for the whole organization, which were more than 5000 people.

One problem, though: I had a completely different field of study at university. My master’s degree isn’t in computer science, let alone artificial intelligence. It’s in business innovation. But here is where I got lucky: I was a huge geek growing up. I knew the maps of my favorite video games better than the street names of the town I lived. It might not’ve been good for my social life, but it was amazing for technical understanding. When you’re in awe of digital worlds that much, then you’re bound to become interested in the technologies behind them.

So when GPT 3.5 hit the mainstream, my tribe of fellow geeks were all over it.

Even though I might’ve played video games twice a year at most during this time, my love for science and technology was already imprinted so deeply my study of large language models was almost a given. Who said video games couldn’t be beneficial? Eat that, Mrs. Signer! (…apart from the math test I failed, but a legendary two-handed sword isn’t going to farm itself).

A couple of night shifts later, my colleague and I completed the instruction manual. The manual wasn’t just used by our business unit, but parts of the whole corporation! What a great feeling that was. My team and I were convinced the worst was behind us.

How naive we were.

The next day, the head of finance scheduled an urgent meeting with our team. If you’ve ever worked in corporate-land you know that when the C-suite comes knocking, you better answer. Quickly.

Mr. Finance made his request clearly unclear: find me an AI use case for my department.

The problem is that this isn’t how this works.

Let’s imagine you want to build a house. There are some necessary steps you must take to ensure you’re building a house and don’t end up with a hut.

Now think of the architect planning this house. He or she might start with listening to what the client wants. Next is the foundation. Then, the different areas are designed their distinct functions, such as the kitchen, the bedroom, and so on. When this is clear, then you decide on the materials you’re going to use to construct the building.

Software is almost the same: first, you must understand the problem you’re trying to solve – in other words, what the user wants. Then you create a backlog with all the features necessary to solve the problem defined in the step before. Afterward you decide what systems (software architecture, programming language, etc.) to use.

The core must be about value creation. This can only happen when problems are solved. Otherwise, your value-creating feature may cause more problems than it solves. Especially in the beginning of starting an innovation project, it’s seductive to implement new technical possibilities because “everybody does it”.

Which brings us back to Mr. Finance. I understand the inclination of wanting to use a new tool that seems like magic. Artificial intelligence looks like you just entered Hogwarts. But it’s further away from Hogwarts and closer to a math and statistics faculty. These models are based on cold mathematics, nothing more, nothing less.

That doesn’t mean they’re not useful. On the contrary! But artificial intelligence is, in its essence, a tool. If an AI application doesn’t solve your problem, then you don’t have the wrong problem, but the wrong tool.

“If the only tool you have is a hammer, you tend to see every problem as a nail”

“But there are already AI-native products!”. Yes, but that’s not the foundation. It’s always about the problem you want to solve. The problem is the peak of the mountain you want to conquer, the solution is all the steps you take on the way up. AI-native only makes sense if the solution to the problem you’re facing is an AI-native stack. Otherwise, your non-solution will create more problems, throwing money and resources down the drain.

A lot of the time it already breaks down with the definition of AI. When most people say “AI”, they mean large language models (LLMs). But AI as a field is much more diverse with many more possible applications.

LLMs are phenomenal helpers for creating prototypes of software products fast. But because they approximate to some mean, these software products are going to look similar. You can already see it with some website designs. Vibe-coded websites all give off, well, the same vibe. The good news is that many people can now create digital products. The bad news is that many people can now create digital products that are going to be similar to each other or just plain bad. But – and that’s where it gets exciting – the very good news is that putting in extra effort will propel you to a whole different level than before.

Fast-forward two years. I’m working on a different project at a different company. After going back and forth on the problem definition for a week, we finalize it. After all this work of drilling down to the core, the next question baffles me:

“How about an AI implementation – maybe a chatbot?”

Different company, same fallacy.

Which makes the point clear: play more video games and join the geekdom.

Just kidding. Video game companies also fall prey to this line of thinking.

Instead, before you implement artificial intelligence into your product because of its current hype, take a step back and look at the problem definition, then think from there. That’s when you will delight people and provide real value instead of an artificial one.