June 10, 2026
How to kill your AI transformation with a spreadsheet
30% faster document processing. 25% less time in certain workflows. Numbers your CFO can put in a deck.
So the board nods. The AI program gets funded.
Six months later: stuck.
Every time. Every single time.
When you measure something new with old KPIs, the system optimizes for the old KPIs. And in companies, the old KPI is always cost.
So AI becomes a cost-cutting tool. The 25% efficiency gain becomes 25% fewer people. The people who survived watched it happen, and from that point on they stop showing you what else AI can do, stop experimenting, stop raising their hand.
Why would they?
Mollick put it bluntly: the measurement "always ends up falling to cost savings, always let's fire people, which undermines everything you're doing."
Always. His word.
You're not in the AI business yet. You're in the R&D stage, exploring what works inside your specific company. And in R&D, measuring productivity is like measuring a lab by how many test tubes it uses per hour. Technically measurable. Completely useless.
What did you learn this month about where AI works in your company? What failed? What surprised you?
If you can't answer those three questions, you don't have an AI program. You have a cost-cutting program with a fancier name.
Business fragile measures what it saves. Business solide measures what it learns.
You tell me the three KPIs your board tracks for the AI program. I tell you which ones are measuring R&D and which ones are quietly killing it.
PD: The spreadsheet is fine. What you chose to put in it is the problem. There's a version of that spreadsheet that actually tells you something useful. We build it in the first two weeks.
PD2: Tomorrow, something about the person in your company who's already solving this. You just don't know their name yet.