Skip to content
A model knows what usually happens. You know what is true here.
The most dangerous output a model gives you is not the one that is obviously wrong.
It is the one that is well structured, confident, and almost right.
You can catch obviously wrong. Everybody catches obviously wrong. What slips through is the recommendation that reads as if it came from a good consultant who has never met your company.
The dividing line
After a year of watching executives work with these tools, the line has gotten pretty clear to me.
AI is strong on what usually happens. Benchmarks, patterns, comparable situations, the base rate, the way this normally goes.
AI is weak on what is true here.
It does not know that this customer has been quietly unhappy for eight months and has not said so. It does not know your head of operations is retiring in April and has not told anyone but you. It does not know that the last time your company tried this exact move, the market punished you for it, and half your leadership team still carries that.
That is not a gap in the technology. That is a category of knowledge that has never been written down anywhere, because it lives in one person’s head.
Where it should augment, and where you own the call
Let AI do the work it does well. Drafting. Summarizing. Finding the comparable. Pressure-testing an argument. Give you the base rate before you decide to be an exception.
Keep four things for yourself.
Consequence. What actually happens downstream, to your people, your customers, and your reputation, not what happens in a model of a company like yours.
Values. What you will and will not do, which is not a data question and never was.
Tradeoffs. What you are weighing against what, which only you can see because only you carry it all at once.
History. What happened last time, and what it cost.
Two questions worth stealing
When the dashboards disagree, ask: what would have to be true for this data to change our decision?
And for anything a model hands you, ask: what would AI need to know about us before I would trust this recommendation?
I like the second one for a specific reason. It is not a question about AI. It is a question about you.
Answering it forces the context out of your head and onto the table, in words, in front of your team. Which is useful immediately, and useful for a reason I want to end on.
The part that should bother you a little
Everything I just described- the consequence and the values and the tradeoffs and the history- is the most valuable asset in your building.
Twenty years of pattern recognition. The thing that tells you something is off before the data says so.
And right now it exists in exactly one place. Inside one person’s head.
That is not a strength. That is a single point of failure, and it happens to be the most expensive one you have.
That is what I want to talk about next week.
Lisa
If this made you notice something you haven’t said out loud yet, message me. Tell me what you noticed.