Why Your AI Says Everything Looks Fine, Every Week

You set up an assistant to watch your numbers. Every week it comes back and tells you things look broadly fine. Slow start to the month, picking up, nothing urgent.

Then a month goes badly and you find out from your bank balance.

The usual explanation is that the model is agreeable, or that it is not smart enough yet. Neither is what is happening. The problem is almost always that you gave it targets, and a target is not something a month can fall below.

A target is a ceiling. A floor is a line.

A target is a number you are trying to reach. It describes an ambition. So when your assistant reports against it, the most it can ever say is a percentage: you are at 68% of target, you are at 91% of target. That is arithmetic. It is not a situation, and there is nothing in it to act on.

A floor is the other end. It is the minimum you would accept before you would say the month went badly. Not the aspiration, the line.

The difference matters because of what each one makes possible. With a target, every number is somewhere on the way to it, so nothing is ever wrong. With a floor, a number is either above the line or below it, and below the line is a headline.

Without floors, nothing can be off track, because there is no line to be under. Everything is fine because nothing is capable of being not fine.

What this looks like in practice

Take one number. Revenue, booked calls, active clients, whatever you would check first if you only got to check one thing.

Write down the lowest value you would honestly accept for the month. Not the number you want. The number below which you would change something.

Then tell your assistant to report two things about it: the distance to that floor, and whether it expects to clear it. Never percent to target.

The reports change immediately. Instead of "revenue is tracking at 68% of goal" you get "revenue is $1,400 under floor with nine days left, and the pipeline does not cover it." The first is a status. The second is a decision waiting to happen.

Why almost nobody does this

Floors feel pessimistic. Targets feel like ambition, and writing down the minimum you would accept feels like planning to fail.

That gets it backwards. The floor is not what you are aiming at. It is the tripwire that tells you to look up. You can hold an ambitious target and a conservative floor at the same time, and the floor is the one that does the work, because it is the only one that can be crossed in the direction that matters.

The other reason is that a target can be vague and still feel fine. A floor forces you to decide what bad actually looks like, in a number, before you are in it. That is uncomfortable to do in advance and enormously useful once it is written down.

The ten-minute version

You do not need a system for this. Open whatever file or doc your assistant reads, and for each goal replace the aspiration with the minimum. Add one line to your instructions: report distance to floor, not percent to target.

Then read the next report. It will say something different, and if it does not, you have learned that the number it is reading is not connected to anything real. That is worth finding out too, and it is a different problem: how to tell a useful run from a plausible one.

Where this fits

Floors are one of five steps in the method I use to get AI doing standing work rather than answering questions. The others are the charter that decides how it behaves, the schedule that makes it happen without you, the review that keeps it honest, and the ladder that decides what it is allowed to do alone.

The Board Method is a free seven-page guide that covers all five, with the real files from a board that is running.