Before automating anything with AI, I ask one annoying question: which number needs to change? It sounds obvious. Almost nobody does it.
Most people start with the tool. They buy the AI, build the pilot, and only afterward go looking for where it fits. Then they spend a fortune automating a process that was crooked from the start.
In Lean Six Sigma we learn it the other way around. First you map out how the process actually works. You find the bottleneck. You define the metric that matters. You fix the process. And only then do you automate what's left. That's how it went at Marksell (27 to 500+ leads) and on the growth teams (103% of target): the leap didn't come from the tool, it came from fixing the process first.
The first person to nail this down was Bill Gates, back in 1995, in his book The Road Ahead: automation applied to an efficient operation will amplify the efficiency; automation applied to an inefficient operation will amplify the inefficiency. Swap "automation" for "AI" and the sentence is still exactly true thirty years later.
New technology doesn't fix a bad process. It just speeds up the mistake. Next time someone pushes an AI tool on you, ask: what's the number, and where's the bottleneck? If nobody can answer, it's not time to automate yet.
