A company decides to use AI. It buys access, organizes a demo, and appoints a point person "on top of their current job." Six months later, two people use it now and then and the topic has vanished from meetings. This story repeats itself, almost identically, in companies of every size.

The common thread is never the technology. It's the organization.

Cause #1: nobody owns it

Handing AI to someone "on top of their job" amounts to handing it to no one. The work takes sustained time: describing tasks, preparing the material, tuning instructions, correcting, measuring. That time doesn't exist in a schedule that's already full, and the urgent always wins.

A project that nobody owns isn't a project. It's an intention.

Cause #2: the scope is too broad

"Use AI to boost productivity" can't be implemented. A workable scope looks more like this: "every incoming request is sorted, summarized and matched with a template reply in under ten minutes." You know what to build, you know when it's done, and you know whether it's better than before.

A broad ambition isn't a lack of vision: it's a lack of breakdown. It makes measurement impossible, and therefore correction, and therefore progress.

Cause #3: the material isn't ready

AI works on what you give it. If your procedures aren't written down anywhere, if your documents are scattered across five different spaces, if every salesperson has their own version of the price list, the result will look the same. That's not a flaw in the model, it's a reflection of the state of your information.

The good news: this clean-up work has value in itself. Many companies discover, while preparing an AI project, inconsistencies that were already costing them money.

Cause #4: no success criteria

Without criteria, assessment becomes a matter of mood: "it's amazing" in week one, "it's not much use" in week six. Three numbers are enough: how long the task used to take, how long it takes now, and the share of outputs that need correcting. They settle in ten minutes what a debate of opinions never will.

What an orchestrator actually does

  • They choose which tasks to handle, and turn down those that aren't a good fit.
  • They prepare the material: documents, rules, examples of good results.
  • They connect the tools to each other so information flows without re-entry.
  • They monitor quality, costs and drift: a setup left unmonitored degrades.
  • They keep the decision in human hands wherever a mistake is costly.

Who owns it, and on what scope

Before relaunching an AI project, answer three questions: who owns it, on what precise scope, and how will you know in two months that it has worked? If any one of the three goes unanswered, the project will fail for the same reason as the last one.

AI doesn't lack power. It lacks someone whose job it is to land it.