The Unlearning Method.

We do not start with tools, and we do not start with training. We start with a map of the work: we sit with the team, look at a normal working week, and write down what repeats, what drags, what is done out of habit and what blocks everything else. Out of the conversations with people and their managers comes a concrete list of tasks, not a general impression.

By the end of the phase we know three things: which tasks are worth changing first, where AI genuinely helps and where it would only add noise, and what the starting level is, so that later we can prove something actually changed. This is also where the Measured AI Adoption Hub captures the baseline, directly from the people who will go through the programme.

We take the tasks selected in Assess and design their new version, role by role: what AI prepares, what a person verifies, which decisions are never delegated. This is not a theoretical exercise. It happens together with the people who do the work, on their real documents and real situations, so the new design survives contact with reality.

This is also where the rules of the game get written: what data may enter a tool, in what form, who approves exceptions. And the team's decision logic, what "good" means, when to escalate, gets documented once, so complex tasks stop being judged by individual habit and start being judged by a shared standard.

This is where training happens, but not as a one-day event. People practise on their own reports, emails and analyses, with the trainer next to them, until the new reflex replaces the old one. The unlearning part is taken seriously: we name the old habit that has to go, explicitly, because an old habit that is not named comes back within two weeks.

Between sessions, AI clinics and office hours keep the practice alive: people bring real cases, and every case ends as a prompt or template saved in the team's shared playbook. Copilot, Gemini or ChatGPT, whatever the company already has, enters the method here, as an instrument, not as the star of the show.

We measure business impact, not platform usage statistics. The thresholds are set in advance, back in Redesign: what accuracy is mandatory, how much time must be recovered, what cost per task is acceptable. Thirty days in, the Hub calculates the delta against the baseline: how many hours came back, how many tasks run the new way, how many people genuinely work differently.

The result is a report management can act on: what worked, what did not stick and why, what deserves to go further. If something failed, it shows up here, after a month, not after a year of pretending.

Practices validated with one team become company standards: playbooks move into an internal library any team can draw from, the AI champions formed during the programme carry the practice into their own teams, and the rules for data, access and cost become written policy instead of tribal knowledge.

Only here do we talk about automation and AI agents. Not because we dislike them, but because an agent built on a badly redesigned process scales the problem, not the solution. Where the evidence from Measure justifies the step, stable flows move from AI-assisted to automated.