Where’s the customer in this story?

Ten minutes into a mock quarterly business review, I had to stop the candidate.

The deck was polished. Adoption charts, usage trends, ticket volume, Net Promoter Score. Data everywhere. And completely soulless.

“Where’s the customer in this story?” I asked. “What were they trying to achieve when they bought us? Where is that in this narrative?”

She couldn’t answer. Nothing in the deck connected to the customer’s original goals. It was an internal update we happened to be sharing externally.

I told that story on LinkedIn a while ago, and it was about customer success. But I’ve been running AI teams since, and here’s what I didn’t expect: it’s the same test.

Ask an AI to summarise a customer escalation, and it will give you something confident, fluent and data-rich. Adoption-chart energy. The question that separates a useful answer from noise is exactly the quarterly-review question: where’s the task in this story?

You asked for a summary so someone could act. Does the output tell them what to do next? You asked for an invoice’s details so a payment decision could happen. Does the output say what’s missing before anyone should pay?

Fluency is the deck. The task is the customer. Most AI evaluation goes wrong the same way most quarterly reviews go wrong — grading the polish instead of the progress.

This is why the Lab’s Compare experiment doesn’t score answers for you. It puts two outputs beside the same task and asks which one you’d actually use. That judgement — output against task, not output against vibes — is the entire skill. If you spent years asking “where’s the customer in this story?”, you already have it.

The slides are a prop. The story is the value. That was true in every quarterly review I’ve ever sat in, and it’s true of every AI answer I’ve ever shipped.

Try this

Try it: run Compare with the same task and two models, and grade the answers against the task before you look at anything else.