Stuck in pilot - Part 1: no foundations, no future
“We just need to test the AI. We’ll figure out the data later.”
That sentence has quietly killed more AI pilots than any model failure ever could. Most AI pilots don’t fail because the tech didn’t work. They fail because no one did the groundwork. No clean data. No integration path. No clear problem. No business owner. Just a well-intentioned scramble to “try AI” and show something in a demo, preferably fast. And then everyone wonders why it didn’t go anywhere.
The myth of the fast pilot
Pilots are often sold as a quick win. Low commitment, high visibility. A chance to show progress without the red tape of production work. But if you skip the prep: if you start before you’ve validated the problem, the data, and the path to value, then you’re not testing AI. You’re testing how far a half-baked prototype can limp before someone shelves it with a polite “let’s revisit this later”. Big sigh!
You think you’re being agile. You’re actually being avoidant.
What’s almost always missing
Let’s get specific. Here are four foundational gaps that quietly ruin most AI pilots before they even start, and unfortunately, I see all of them when my customers show me their pilots.
1. Clean, reliable data
Everyone assumes the data is “somewhere”. No one checks if it’s usable, consistent, labeled, or accessible. So when the model performs poorly, people blame the algorithm instead of the dataset. Garbage in, garbage model. We should know by now, that this sentence correctly says: garbage in - garbage out.
2. An integration plan
The pilot lives in a beautiful little bubble. It doesn’t connect to the CRM. It doesn’t update any real records. It doesn’t push anything to production systems. So even if it works, it can’t scale, because no one thought about how it would fit into the real-world process.
3. A clear problem statement
You get goals like “reduce time spent” or “optimize the workflow” or “improve customer experience”. Vague enough to be exciting. But totally unhelpful when you try to define success. What, exactly, are we fixing? How do we measure this?
4. Business ownership
When a pilot is purely IT- or innovation-driven, it often lacks a business owner who actually cares about the outcome. No one to champion it. No one to fight for its rollout. No one whose KPIs depend on it. So when priorities shift, the pilot just… dies in corporate.
Every pilot needs a contract (even if it’s just a slide)
Before you build anything, you should be able to answer five basic questions:
-
What problem are we solving? Use plain language. No jargon. No “digital transformation” fluff
-
How bad is it right now? What’s the baseline? What’s the pain? What are the symptoms?
-
What data are we using? Where does it live? Who owns it? How clean is it? Can we access it?
-
How will we test it? What’s in scope? What’s off limits? How will we simulate usage?
-
What does success look like? Define the win condition now, not retroactively after the pilot’s already derailed
This doesn’t have to be a 10-page deck. But it has to exist, before the pilot starts.
Skipping the foundation isn’t agile—it’s wishful thinking
You can’t move fast if you don’t know what you’re building, or for whom. And no amount of generative flair or AI hype will rescue a pilot that’s missing the basics. If your pilot is slow, buggy, or underwhelming, it may not be the tech. It may be the homework you skipped to feel fast.
Closing thought
You can’t build a second floor when the foundation is still guesswork.
If your pilot’s in trouble, don’t blame the model or the team. Look backward. Because chances are, the cracks were already there before you wrote a single line of code (or clicked yourself through a portal).
More parts of this series:
- Stuck in pilot - Part 0: The comfort of the sandbox
- Stuck in pilot - Part 1: No foundations, no future [📍 You are here!]
- Stuck in pilot - Part 2: Fear of success - to be published soon™️
- Stuck in pilot - Part 3: your chatbot isn’t useful - to be published soon™️
- Stuck in pilot - Part 4: from stuck to scaled - to be published soon™️
Published on:
Learn more