Hype or real? Ten AI claims, can you tell?

Ten claims you will hear in a board meeting or a vendor pitch. Pick hype or real. Then see how I read each one.

10 claims
3 minutes
Score on screen

Claim 1 of 10

Your score

You spotted 0 of 10.

See why each claim is hype or real, plus the Hype Filter: the seven questions I ask before I believe any AI claim. With a PDF to keep.

Photos from Unsplash by Blaz Erzetic, Austin Hervias, Jincong Zuo, chris panas, Dongsh, Mo Darasi, Remy Gieling, Matt Artz, Sergei Wing, Lilian Velet.

Our pilot cut handling time by 60%, so the business case is proven.
A pilot proves an output in good conditions. The business case is proven when the outcome still holds in production, at full volume, months later. Promised, piloted, production, permanent: most value leaks between the second and the fourth gate.
We track the cost per resolved case, and it went down after we scaled.
Cost per outcome, measured after scale, is the evidence most AI programmes never collect. If you have it and it falls as volume grows, the value is real.
We deployed 40 agents this quarter.
That is output, not outcome. Agents deployed say nothing about what changed for the business. Ask which metric moved, and by how much.
The vendor handles everything, so we don't need AI skills inside.
If nobody inside can change what runs in production, the vendor holds your Talent lock. Every change becomes a purchase order, and every renewal a negotiation you cannot walk away from.
We can switch our main model provider in a few weeks, without rewriting our applications.
That is the Model lock held, and it keeps your price and your roadmap in your hands. One follow-up question: when did you last test it?
Every funded AI initiative names, before the budget is approved, the outcome metric it must move.
This is how you stop paying for activity. Set the ladder (output, outcome, impact) before you set the budget.
Our AI is 95% accurate, so it is ready for production.
Accuracy on a test set says little about the cost of the other 5%. Ask what one error costs, who catches it, and how fast.
Our prompts, evaluation sets and embeddings live in systems we own.
That is the Data lock held. When you change vendor or model, your knowledge stays with you.
AI will replace 30% of our workforce next year.
A headline number with no list behind it is a guess. Ask for the processes, the controls and the owner of each result. Without that list, nobody can plan for it.
We priced what it would cost to leave our main AI vendor, in months and in money.
That is the Exit lock under control. A priced exit is your strongest card in every renewal.

You read AI like an operator.

Good instincts. A few claims still get past you.

Some vendors would love to meet you. Read the answers before your next pitch.

Right.

Not quite.

Hype

Real

It was:

Your answer:

See my score