The Survival Test

Most of the AI cases I am asked to look at are true and useless at the same time.

True, because the technology really can do the thing. Useless, because "it can do the thing" was never the question. The question is whether doing the thing is worth what it costs, to you, at this company, after the excitement wears off. And almost nobody writes the case that way, because while the technology is new and everyone is talking about it, you do not have to.

So here is a test you can run this week, on whatever you are being sold, including the thing you are selling yourself.

The one question #

Does this still make sense if the hype disappears completely, and the price of the technology moves by a factor of ten in either direction?

That is the whole test. The trick is that it has two directions, and most people only check one.

Ten times more expensive #

Start here, because it is the uncomfortable one.

If the model you are building on got ten times more expensive tomorrow, would you still do this?

If the answer is no, you do not have a business case. You have an arbitrage on a price you do not control. That can still be worth doing, the way any temporary edge is worth doing, but you should know that is what it is, and you should not build a five-year plan on it. Vendors lose money to win a market and then raise prices once you have rebuilt your operation around them. If your case only works while the technology is subsidized, you are betting the operation on someone else's pricing decision.

Ten times cheaper #

Now the one that feels like good news and often is not.

If the technology got ten times cheaper, does your advantage grow, or does it quietly evaporate?

It evaporates when the thing you were proud of was simply that you did it first. Cheap means everybody can do it. If your edge was access to the capability, cheap hands that capability to your competitors too, and the head start closes. It grows when your advantage lives in something the cheap technology cannot copy: your data, your process, the particular way your operation is put together, the judgment of the people running it. Ten times cheaper makes those worth more, because now you can apply them everywhere.

A case that only works at today's price is not a case. It is a bet on the price.

Strike every sentence that needs the technology to be new #

There is a faster version of the same test you can do on any pitch, including your own memo.

Go through it and cross out every sentence whose persuasive force depends on the technology being new, exciting, urgent, or something a competitor is doing. "Everyone is adopting this." Gone. "We cannot afford to fall behind." Gone. "This is the biggest shift since the internet." Gone.

Read what is left standing. That is your actual case, and it is usually a lot shorter than the pitch. Sometimes there is a real, boring, durable business underneath, a workflow that got shorter, a decision that got faster, knowledge that stopped living in one person's head. Sometimes, when you cross out the excitement, there is nothing left. Better to find that out now than eighteen months into a build.

Why you cannot run this on yourself #

Here is the part that makes the test hard, and it is not a matter of intelligence or discipline.

You cannot be the advocate for a case and the honest check on it at the same time. Everyone looks defensible when they are the one who wrote the memo. I have talked myself into things this way, more than once, with good slides. It is not a flaw in you. It is just that the person who fell in love with the idea is the worst-positioned person in the building to find the hole in it. The case needs an opponent, and the opponent cannot be you.

Most people do not have that opponent standing by. So we built one.

A small tool that argues back #

We put a free version of this test online. No signup, no email, and nothing you type is sent to us or stored anywhere. You paste in your case, it walks you through the questions above, and then it hands you a prompt that makes an AI argue against your own case as hard as it can.

You can run it here: The Survival Test.

That last part is worth saying plainly, because it is the whole discipline. The same model will write you a beautiful justification for your project, and then, asked the other way, a devastating rebuttal, with equal confidence both times. It is not on your side. It is not against you. It is a mirror with an excellent vocabulary, and confidence from a model is a formatting choice, not a signal of truth. So the answer you are looking for is not in either argument. It is in whatever is still standing after both.

Build on the part that survives the pop #

None of this is an argument against AI. We use it every day, and most of the work we do for clients depends on it. It is an argument against building on the part that only holds while the room is excited.

There is a correction coming. There always is. I will not pretend to know when, because I have been wrong about timing more than once. But the businesses that come out of it ahead will be the ones that spent the froth building on what survives the test, while their competitors spent it standing on the part that does not.

Run the test on your biggest AI bet this week. If it survives, build it with confidence. If it does not, you just saved yourself a very expensive year.