AI on top of an unclear decision makes the unclear decision faster.
Carsten Christiani
If nobody in your company can say who really decides something, putting AI into that process will make it faster without making it any clearer. It will also make it much harder to see. And as far as I can tell, it is how a good share of what people call AI chaos starts, the kind that creates more mess than it removes.
Think of it as hiring a hundred very fast, very literal new starters and training them on the process document. They will follow it exactly. If the document describes how the decision really gets made, that is wonderful, and you get the same good decision more cheaply every time. If it describes how the decision was supposed to be made when somebody wrote it down three reorganisations ago, you get that version at speed, and now it comes with tidy paperwork that makes it look decided.
The test for which kind you have is one question, and it's the one our diagnostic leans on hardest. The last time something went wrong, who actually decided what happened next? I mean who did, which isn't always who should have. In a lot of companies the org chart gives one answer and the corridor gives another. Some meetings produce decisions and some produce the feeling of a decision, after which the real one gets made somewhere else, by whoever was going to make it anyway.
Now put AI on top of that. What gets automated is the version you can describe, which means the org chart and the process document, because those are the parts that are written down. The real decision carries on where it always happened. The difference is that there is now a machine producing confident summaries of a process that isn't the one being used, and the gap between the two has become very well formatted.
That is how a company manages to get more sophisticated and less alive at the same time. It has more dashboards and more output, and less and less contact with what is actually going on inside it. Nobody chose this; each tool went in for a good local reason. But each went in before anyone had looked at how the company actually decides, so every tool sped up a different piece of a process nobody could draw.
There is a quieter cost too. AI getting stronger does not make human intelligence less important. It makes it more exposed, because the judgment in a process is exactly the part that has to stay with a person, and you can't protect that part if nobody knows where in the process it sits. Automate the corridor decision by accident and you have handed a judgment call to something that was only ever shown the paperwork.
So the order matters, and in our own work it doesn't change. First we find out who actually decides and which parts of the decision people still have to hold themselves. Only then does AI go in, where it earns its place, with a clear line where it has to be blocked.
On Monday, pick one decision that came back to the table this month, the kind people describe as "we keep revisiting this." Don't ask the person who chaired the meeting who decided it. Ask the person closest to the work that the decision affects what actually happened, because their account will often differ from the official one, and theirs is usually the one the work follows. Then look for the place where the reason for the decision is written down. If you can't find it, ask whoever really decided to write the reason in two sentences, and hold off automating any step that relies on that decision until those two sentences exist. If it turns out the person who really decided is you, that's fine. You can write them tonight.
Not the most exciting AI strategy, I know.