Insights

AI adoption is a process change before it is a training need

Staff are already using AI. What is usually missing is a process that has been changed to match.

  • Ravi R. Singh
  • 4 min read

Your people did not wait

In most companies AI adoption has already happened, and management was not involved. The 2024 Work Trend Index from Microsoft and LinkedIn reported that 75% of knowledge workers use AI at work. It also reported that 78% of those users bring their own tools.

That second number deserves attention. Staff are pasting company information into personal accounts on tools nobody approved, to do tasks nobody redesigned. Each person has a private method. None of it is shared, checked or measured.

The usual response is a training programme and a policy. Both are needed, and neither changes how the work gets done.

Why training alone does not stick

Generic AI training teaches people what the tool can do. It does not tell an accounts clerk what to do differently on Tuesday morning. People leave the session interested, go back to a process that has not changed and carry on as before.

I have watched the same thing happen with earlier technology in teams across this region. A new platform arrived, everyone was trained, and six months later most of the team used a fraction of it. The teams that got the full benefit had changed the workflow first and then trained people on the new workflow.

Adoption belongs to a process more than to a person. Someone adopts a tool when the process in front of them assumes it.

Redesign the work, then name the roles

Start with one process and draw it as it runs today, step by step, including the waiting and the rework. Mark the steps where the tool will act. Then redraw the process with those steps changed.

Three decisions fall out of that drawing. The first is where a human checks the output and what they are checking for. A reviewer told to look it over will approve everything. A reviewer told to confirm the invoice total and the supplier name will catch errors.

The second decision is who owns the process once it changes. The third is what happens to the time released, which the business case should already have answered.

Write the changed roles down. If a coordinator used to draft and now reviews, their job has changed. Their manager and their targets should reflect it.

Guardrails people can remember

Policies fail when they run to twelve pages. Staff need a short set of rules they can hold in their heads. Say which tools are approved. Say what information must never be entered, such as customer personal data or unpublished financial results. Say which outputs need a human check before they leave the building. Say who to tell when something goes wrong.

Give people an approved tool that is at least as good as the one they were using privately. A ban without an alternative moves the behaviour out of sight, and the organisation loses the little visibility it had.

What good looks like after 90 days

Three months in, a process that has been adopted properly shows visible signs. The team uses one approved tool and one shared set of instructions, kept where everyone can find and improve them. New joiners learn the redesigned process as the normal way of working, and nobody calls it the AI project any more. The reviewer can say what they check and how often they find a problem. The manager can state the current volume and the time taken without looking it up.

Shared instructions deserve a note. When each person keeps a private prompt, quality depends on who happens to be at the desk. A common set that is updated in one place makes the output consistent and lets one improvement benefit everyone.

When those signs are missing, the cause is rarely the tool. Look for a step that was never redesigned, a role that was never rewritten or a manager who was never asked to own the result. Fix that one thing before adding anything new.

Managers decide whether it holds

The last piece is the weekly routine of the line manager. If the manager asks how the new process is running, looks at the numbers and removes obstacles, the change holds. If the manager never mentions it, the team reads that as permission to drift back.

Plan for a dip. For the first few weeks the new way is slower than the old one, and this is the moment most changes are quietly reversed. Tell the team to expect it, and set the first review after the dip has passed.

Involve the people who do the work in the redesign. They know where the process breaks, which exceptions matter and which review steps are for show. A process designed without them will be worked around within a month.

This is what I mean by the Work stage. I run it as the AI Adoption and Operating Model Programme, a fixed scope at US$7,500 planned at 40 founder hours. The principle holds at any scale. Change one process properly before you train a hundred people on a tool.

Sources

AI Adoption and Operating Model ProgrammeAll articles