The 5-Question AI Autopsy

What it is

When an AI initiative fails, the reflex is to blame the model and buy a better one. The evidence says otherwise: 95% of corporate AI transformations returned zero measurable P&L impact, and the cause was a learning gap, not a weak model. In nearly every famous failure the model did exactly what it was built to do: what broke was a layer that was rotten before the technology arrived.

This is a printable, fillable post-mortem you run on one initiative at a time. Five questions, one per layer of the Stack (Vision, Data, Process, Human Experience, then AI), each with write-in prompts and a SOLID / CRACKED / BROKEN verdict. A grid at the end identifies the red layer, the cause of death, and the one fix that brings it back. The order is the diagnosis.

What’s inside

• 01Name the caseA short intake: the initiative, the owner, the one P&L line it was meant to move.
• 02Q1 Vision · Q2 DataWas there one owned outcome, and could the system reach how the company really works?
• 03Q3 Process · Q4 Human ExperienceDid it absorb the real workflow, and did people trust it, brief it, and own the verdict?
• 04Q5 the exponentDid you multiply something solid, or scale chaos faster? The question you answer last.
• 05The verdict gridMark each layer, find the red one (the cause of death), and write the one fix, owner, and date.

How to use it

1. Print it, take one failed or failing initiative, and answer all five questions honestly in about fifteen minutes.
2. Found the red layer? Run The Stack Self-Audit on the whole house, or The Index for a measured read.