AI does not respect your org chart

The short answer

The processes where AI pays back are almost never inside one department. A claim touches intake, analysis, finance and customer service. A credit decision touches commercial, risk, operations and legal. The budget, however, sits with one of them.

So the project is owned by one area and depends on three others that have no reason to prioritise it. That is the most common way a good AI initiative dies, and it has nothing to do with the technology.

Why the org chart is the wrong unit

Org charts are built around functions. Losses happen at the handoffs between functions: the document that goes back for correction, the exception that waits for someone in another team, the decision that needs data owned by a third.

An AI system that only works inside one box can at best speed up a task. The money is in removing the handoff, and the handoff by definition belongs to nobody.

What it looks like when it goes wrong

Operations sponsors an agent to triage exceptions. It works in the pilot. To go to production it needs read access to finance data, a change to a form owned by commercial, and a sign-off from compliance.

Each of those is a reasonable request and each sits in a queue behind that area's own priorities. Six months later the pilot is still a pilot, and the conclusion in the steering meeting is that AI is not ready. It was ready. The organisation was not arranged to let it through.

Who should own it

Own the process, not the tool. The owner should be the executive accountable for the outcome the process produces: loss ratio, cost to serve, days to close, default rate. That person can resolve conflicts between the areas the process crosses, or can escalate to someone who can.

IT and data teams are essential partners, but when they own the initiative it tends to be measured by delivery, not by the business number. When a single business area owns it, it tends to be optimised for that area and stall at the border.

Set it up before the project starts

Name the process end to end, including every area it touches. Name one accountable executive. Get each touched area to commit, in writing, to what it will provide and by when: access, a person for review, a decision on a policy.

Agree the number that defines success and who reports it. Agree what happens if an area does not deliver its part. This takes a week of conversations and saves months of stalled delivery.

What we ask for in discovery

When we map a process in discovery, the map of who owns which step is as important as the map of systems. If the process crosses areas that have not agreed to change together, we say so before a pilot is built, because a pilot that cannot reach production is the most expensive outcome there is.