top of page

AI keeps paying off when someone stays to own it

Writer: Sascha Cutura
Sascha Cutura
11 minutes ago
3 min read

Most AI work I see is shaped like a project. A vendor arrives, a tool gets chosen, a pilot goes live, and then the room empties.


That is usually when it should start earning its keep. In practice, it is often when the attention leaves with the project team.


I have been in this situation enough times to recognise it early. The licence is still being paid. A handful of people still open the tool when they remember. Nobody can tell you, without a pause, who is responsible for whether it is actually helping the business.


This problem is older than generative AI. Process work has always been at risk the day after implementation. AI makes the gap easier to miss, because the system can look busy. Drafts appear. Tickets get summarised. A report lands in a mailbox. Activity can look like progress while the underlying result stays flat.


The lesson I keep coming back to is this.

Decide what the work is, before you decide what to buy.


Take a process the business already measures, or should. Invoice handling. First-line customer replies. Qualifying inbound leads. Name the person who already owns that process, and agree the figure that would show the new approach is doing its job. Time to complete. Error rate. How much rework comes back. How often a person has to intervene. If you can't name the owner and the figure, you are not ready to roll anything out.


Then plan for the months after go-live with the same seriousness you planned the build.

A project is designed to end. The conditions around AI do not. Models get updated without a meeting in your calendar. Someone in the team adjusts an instruction and the output changes tone, or worse, changes a decision. A new feature appears inside a platform you already use, and a step that used to draft a message can now send it. The data the model sees this quarter is not quite the data it saw when you tested it.


Any of those can undo a careful launch. A closing workshop will not catch them.


So the controls have to sit in the process itself. When the system is unsure, the next step is a person, as part of how the work already runs. If more cases start needing a person to fix them, that should show up in the same operational view the team already uses, in time to act. The owner also needs the authority to slow it down or switch it off. A tool that is not earning its place should be stopped, and that should be treated as a normal management decision.


This is why I do the work as a Fractional Chief AI Officer, rather than handing over at go-live. You get a senior person inside the business, for part of the week, whose job continues after the implementation partner has gone. They keep an eye on what is actually in use, including the quiet features that vendors turn on by default. They sit with the process owner and decide whether to continue, adjust or withdraw. They make sure someone is still accountable a few months later, when the first wave of interest has passed and the real operating questions appear.


Most leadership teams don't need a new department to get this right. They need enough seniority in the room to be taken seriously, enough time on the ground to see what is happening, and a brief that doesn't expire at go-live.


If you can answer, midweek, who owns the AI in a given process, and whether the figure you agreed has moved, you are doing the work that actually pays off. If you cannot, you still have a project. You do not yet have an owner.


Sascha Cutura is a Fractional Chief AI Officer and founder of mws+. He has carried out more than 700 process optimisations in 90+ companies over the last 17 years.

bottom of page