I spent $100K on AI this month. Most of it worked. Some of it was the most expen
By Chiefdata · October 15, 2026 · Curated by George's Blog
I spent $100K on AI this month.
Most of it worked. Some of it was the most expensive lesson I’ve paid for this year.
Here are 4 rules I’m keeping:
1. No tool gets a budget before it gets an owner.
We had 23 AI subscriptions running. I asked who used them. Nobody could answer for 9. That’s $4,800 a month nobody was watching. Now if there’s no name next to it, it gets cancelled.
2. Automate the boring part, not the judgment part.
We let AI write and send 12,000 outbound emails in one week. Reply rate dropped from 4.1% to 0.9%. The copy was fine. The list was garbage, because we’d let the AI decide who to email too.
3. Pilot for 2 weeks, then kill or scale.
One “let’s see how it goes” tool ran for 5 months. $31K. Zero closed deals traced back to it. Now every test gets an end date and one number it has to hit.
4. Measure output, not usage.
Our dashboard said 94% of the team used AI daily. Great. Then I checked: proposals went out at the same speed as before. People were busy. Nothing got faster.
The expensive lesson wasn’t the tools. It was assuming spend equals progress.