The three levels of AI adoption

The short answer

Most companies say they use AI. Very few have changed how an operation runs because of it. The gap between those two statements is where most of the value sits.

A simple way to see it is three levels: people using tools, tasks being automated, and operations being redesigned. Each one has a different return, a different owner and a different risk.

Level 1: people using tools

Employees use assistants to write, summarise and search. The return is real but diffuse: individual productivity that rarely appears on any financial line.

The risk is mostly data exposure and inconsistent quality. The owner is usually IT or HR, through policy and licences. This level is necessary and it is not a strategy.

Level 2: tasks automated

A specific task inside a process is handled by a system: extracting data from documents, classifying tickets, drafting a first answer. The return is measurable in hours or cost per unit.

The limit is that the process around the task does not change. The extracted data still waits in the same queue, the classified ticket still goes to the same team. Savings exist, and they are capped by everything the system does not touch.

Level 3: operations redesigned

The process is rebuilt around what the system can now do. Decisions that used to wait for a person are made with the evidence assembled, and people review the exceptions instead of every case. Handoffs between areas disappear.

This is where losses and operating costs move at the scale a CFO notices. It is also where the risk is highest, because the system now acts, and where ownership has to sit with the executive accountable for the outcome.

How to tell where you are

Ask which business number has changed because of AI. If the answer is none, you are at level 1. If it is the cost of a task, level 2. If it is a number like loss ratio, days to close or cost to serve, you are at level 3 in at least one process.

It is normal to be at different levels in different processes. What matters is whether you have at least one process where you are deliberately moving to level 3.

What moving up requires

From 1 to 2 you need a well chosen use case and integration with the systems where the work happens. From 2 to 3 you need a process owner with authority across areas, a measured baseline, human review where the decision carries risk, and a plan for who runs the system after it ships.

The technology is roughly the same at levels 2 and 3. The organisation around it is not.