Have you seen this scene? A company gets excited about AI, builds a nice pilot and shows it on the big screen. Six months later, nobody uses it.
I see it every week. And it's not just my impression: in 2025, MIT measured that around 95% of AI pilots don't move the company's results.
So is technology what's missing? I think process is. Here's why.
What engineering taught me
I come from industrial engineering and Lean Six Sigma. The first thing I learned was to ask before doing. Which number needs to change? Where is the bottleneck?
Most companies do the opposite. They buy the tool and then go looking for the problem. That's when AI becomes an expensive ornament.
The order I follow
First I define the metric. Then I fix the process. Only then do I put AI to do the heavy lifting.
When you follow that order, the number shows up. At a market-leading manufacturer, leads per month grew more than 3,600%. In the growth teams I led, we hit 103% of target with an average ROI of 11.6x. None of those started with the question "which model should we use?".
The money goes to the wrong place
Do you know where companies usually go wrong? The target.
MIT itself found that more than half of AI budgets go to sales and marketing. But the biggest return was in the back office, automating processes. IBM says the same: the gain almost never comes from the model, it comes from embedding AI in the workflow that already exists.
The money goes to what shines, not to what pays the bills.
And where do I fit in?
I don't write code as an identity. I direct the AI, lead the team and answer for the result. That's how what I build lands in the 5% that survive contact with reality.
And your last AI pilot: did you know, before starting, which number it needed to change?
