In 1964, Japan unveiled the Shinkansen, the world's first high-speed train. The engineers asked themselves a question nobody else was asking: if we were to build the railway from scratch today, what would it look like?
The answer was an entirely new system. Different track, different rolling stock, a different timetable. Not a faster version of the old one. Something else entirely.
Most companies do the opposite with AI automation. They take the existing track and put a faster train on it.
Faster mess is still a mess
A retailer we spoke with had "automated its returns process with AI." That sounded impressive in the steering group. In practice it meant: the same six approval layers, the same manual checks, the same forms. They just moved faster. The process had become quicker. Faster at being bad.
This is the pattern we encounter in almost every organisation. Someone says "we're going to automate with AI" and the project starts with the existing process diagram. Step by step it gets translated into a digital version. Same route, same stops, same detours. Just with an AI label on it.
There's a name for this: the Cow Path Trap. You pave the path instead of designing a road. The result is a sealed route that goes nowhere, just faster now. This is Solutioneering in its purest form: starting with what already exists instead of what should exist.
The real problem runs deeper. Many of those processes exist not because they make sense, but because they were once a workaround for a constraint that no longer applies. Three approval layers because there was no digital overview back then. Manual data entry because two systems couldn't talk to each other. Copy-pasting between spreadsheets because nobody ever took the time to set things up differently.
AI puts a turbocharger on that. Process debt becomes accelerated process debt.
Start with the pain, not the process
What actually works is precisely the opposite. Start with the people doing the work. Sit down next to someone. Watch for twenty minutes. Just watch.
People describe their work the way it should be. They do their work the way it actually is. The gap between those two is where the opportunity lives.
Ask: "What's frustrating about this?" Ask: "Why do you have to enter this into three different systems?" Ask: "What would you do if this step didn't exist?"
The answers are almost always surprising. People are so used to the friction that they no longer see it. They navigate around it the way you avoid a pothole in the road you cycle past every day. You stop noticing it. It's just there.
That friction is the starting point. We call this principle Follow the Friction: the best AI opportunities become visible by looking at where work gets stuck.
Start small, learn fast
The temptation is to think big. Redesign the entire process, connect all the systems, automate the complete chain. That sounds ambitious. It is mostly slow.
What works: the thinnest possible slice. The smallest part of the process where the most pain lives. Not the wedding cake, the cupcake. Build that in two weeks. Let real people use it during their normal working day, not in a demo environment.
Then measure. Not just speed. Speed without quality is a dangerous metric. Always measure in pairs: if something gets faster, is it also better? If it gets cheaper, do people still trust the result? Faster and worse is not progress.
There is a simple litmus test. Threaten to switch the tool off. If people complain, you've built something valuable. If nobody notices, it was a solution to a problem that didn't exist.
That sounds harsh. But it protects you against the most common AI problem in organisations: pilots that get stuck. Too alive to bury, too weak to scale. We call them Zombie Pilots. They linger in a twilight zone of "we're keeping an eye on it." That is not a strategy, that is procrastination.
What's left when it works
The beautiful thing about good AI automation is that nobody talks about it anymore. No steering group update needed, no dashboard with AI metrics, no quarterly presentation on progress. The work is simply better. Quieter. Less hassle.
That is the point you're working towards. The moment a team member says: "Oh right, the system does that automatically now. Works great."
Boring is success.
The Shinkansen engineers were laughed at initially. Too expensive, too radical, who needs this. Sixty years later, the Japanese railway system is the reference point for the entire world. They had asked the right question.
The right question for AI automation: what would this process look like if we were to design it from scratch today?