AI adoption starts with the work

When we started bringing AI into one client's operations department, some of the team already used it regularly, others had experimented with it, and some barely used it at all.

Our job was to help each person bring automation and AI into their own processes. We went team by team, starting with Planning and adding teams as the engagement went on.

Starting from each person's tasks

With each employee we mapped the recurring work: daily tasks, weekly routines, monthly and quarterly work, and the ad hoc jobs that mattered.

Some of it didn't need AI at all. We usually began with something simple, such as repetitive or technical work, or a job that took time and little human judgement.

We wanted each person to get an early result they could feel for themselves. Once AI had taken something annoying out of their week, they stopped thinking of it as an abstract tool they'd been told to learn.

The time savings became visible quickly

Some tasks that used to take half a day, or most of a working day, came down to around 10 minutes once the process was AI-supported. Other recurring activities saved about two to three hours per week.

These are examples from individual tasks in an ongoing engagement. They should not be read as an average saving across the team or a guarantee for other tasks.

The change in behaviour mattered more. Employees started coming to our weekly sessions having already tried things on their own.

The question they brought changed from "How do I use AI?" to "I tried this. It worked until here. How do I solve this part?"

Building independence from the consultant

Sometimes I developed part of a solution between sessions and then walked the employee through what I'd done and why.

Whenever we could, though, we worked on it together. The employee wrote the prompts and I challenged them, and then we changed the wording, tested the result, corrected mistakes and tried again.

It takes longer at the start than handing someone a finished solution. I accepted that because the aim was for the employee to solve the next task without me.

Validation remains part of the workflow

Every AI-supported process included validation before it became part of the employee's routine, so the sequence ran:

Task → AI-supported work → Human validation → Adoption

The amount of checking differs from task to task. In every case a person stays accountable for the output, even though AI speeds up the work.

Adoption is a change-management issue

Some people see the possibilities straight away. Others get interested only after they've felt a practical benefit, and some ask a reasonable question:

"If AI can do this part of my job, what does that mean for me?"

The answer can't be a promise that AI will never change roles, because it will.

A more useful conversation is about which work should stay with people: judgement, prioritisation, communication, handling exceptions and understanding context.

What I'd measure

For me, the number of employees who have completed AI training is a secondary measure of progress. The question I'd lead with is:

"How many recurring pieces of work are now being done differently, and can employees improve those workflows themselves?"

Danielle Angel · Opyflow