The Question Amazon Forgot to Ask

In 2014, Amazon began building one of the most ambitious AI hiring tools in corporate history. It ingested a decade of CVs and trained itself to identify the patterns that led to successful hires. Within three years, the project was dead.

The problem? The model had been trained on CVs submitted over a decade when Amazon's workforce was predominantly male. It learned to penalise CVs that contained the word "women's" — as in "women's chess club" or "women's university". It downgraded graduates of all-female colleges. The team spotted the bias, tried to engineer it out — and concluded they could never guarantee the model wouldn't find new ways to discriminate.

The Real Cost of AI Failure

The Amazon story is one of the most famous AI failures in history — but it's not the most common type. Most AI projects don't fail because of hidden bias. They fail because nobody asked the most fundamental question before the project started: Should we actually be doing this?

The question isn't just "can we build it?" — it's "should we build it?", "are we ready to build it?", and "what happens if it goes wrong?" These questions sound obvious. They are rarely asked.

What This Section Will Do

By the end of Section 1, you'll have the Five-Dimension Framework: a structured evaluation model for any AI use case. You'll be able to assess a proposed deployment against five dimensions that determine whether it should proceed, be redesigned, or be declined outright. You'll also apply the framework to real scenarios so you can use it confidently in your own organisation.