Believe it or not, space exploration is a good analogy for AI developments. And like for space exploration, data separates hype from results.

What kind of emotions does AI evoke in you? A sense of wonder or scepticism?

This isn’t a trick question. I am not trying to pull you into one or the other direction. As always, I’m after that tightrope walk all entrepreneurs have to walk.

Recently, we held a session on leveraging AI tools in Yonder, the B2B SaaS company I co-founded. Before the session, one of our developers shared the keynote from The Pragmatic Summit given by Laura Tacho, titled “Data vs. Hype: How Organizations Actually Win with AI”.

In my humble opinion, this keynote nails the AI tightrope walk. Let’s look into this in detail.

Analogy: Space Exploration and AI

The keynote uses space exploration and AI as an analogy. Both technologies invoked a sense of wonder, but also scepticism.

Landing a man on the moon was about redefining what was possible. At the same time, people were asking why spend all this money to go to the moon when we had lots of problems to solve here on Earth?

As with space exploration, there is a lot of awe around AI: A promise of huge productivity increases. The prospect of all our software code being written by AI. And that, finally, the first single-person unicorn will emerge thanks to powerful AI tools. Similarly, there is a lot of scepticism towards AI in the corporate world: AI tools are costly, they consume a vast amount of energy, and nobody has yet figured out how they can add value to the bottom line.

What Should Organisations Do?

Organizations need to learn to balance the sense of wonder with pragmatism.

And what’s the best tool available to move from hype to pragmatism? Data.

Organizations that are winning with AI are defining clear AI goals and measuring against them. And that’s a clearly distinct approach from the “spray and pray” approach we see in most organizations — give AI licenses to all your employees, and tell them to “adopt AI” in all departments and processes. This will not work.

Let’s look at a specific example: Developer experience, or DevEx. That’s the tools and processes software engineers use to produce and deploy production code — documentation, debugging tools, CI/CD pipelines, etc. Ironically, human software engineers were begging for those tools literally for decades, and management always said no. Now, suggesting to spend money on such tools for robot engineers, everybody screams “yes! yes! yes!”.

Let’s not fuss over this irony; let’s use it to make the organization more productive. Good DevEx tools and processes avoid urgent issues and outages, thereby not distracting developers all the time. They also help to ship products faster and with higher quality.

You see the pattern? In this example, we’re not using AI to fuel the hype; we use AI to fix a system-level problem. And if you fix system-level problems with AI, you’re set to achieve higher revenues, better profitability, and faster time-to-market. But only if you think of AI as an organizational problem rather than a fancy new technology.

What Does It Mean for Your Product?

So far, we’ve been talking about using AI to fix system-level problems in your organization. But what about your product?

Just like space exploration, AI is great, but it costs too much money to use it for everything and everyone. If you can use AI to solve a real problem for your customers, go for it. If you just want to introduce AI into your product because everybody else does, be careful: AI features carry significant token costs whenever they are used.

As a SaaS entrepreneur, I follow the rule that all features that can be solved algorithmically should be solved algorithmically instead of with AI. This helps maintain testability and predictability — remember that AI works statistically, and that results given by AI can slightly vary between executions. In our business serving regulated customers, that’s not what our customers want. Furthermore, running an algorithmic feature on a backend server is much cheaper than paying for AI tokens each time a feature is used.

So what features should you use AI for? I would argue for two cases:

• AI is much better than humans at identifying and visualizing patterns in large data sets. Whenever context between large data sets is relevant for your product, AI can be a real game-changer. Humans will not be able to connect those dots, and even if some bright minds do, they will do so too late.
• Looking back, we had many very specific feature requests along the lines of “can you include some more data columns in this report?”. Whenever a feature needs to be customized for individual needs, AI is way more effective than an algorithmic feature. And if your customers can chat with your product like they know it from Claude and ChatGPT, you will have way fewer customer support tickets.

See that tightrope again? Resist the sense of wonder and follow the pragmatism in product management, too.

Conclusion

Always see AI as a business tool to achieve business outcomes, not just a technology. The core question should always be: Who will pay for an AI feature?

The entrepreneur’s role is to find out which AI features customers are willing to pay for. And if you build AI features that no customer wants to pay for, you end up paying for them yourself.