AI work redesign

• Copilot or another AI tool was switched on months ago, and nobody can point to work that runs differently.
• Three teams ran pilots, each with its own tool, and nobody owns the decision to scale any of them.
• Managers were told to use AI, and their weekly routines stayed exactly the same.
• An agent drafted something for a customer, and nobody could say who had decided it was allowed to.
• There's plenty of appetite for AI, but no agreed view of where it would pay off in the business, or who owns it.

The AI work ladder sorts each task by how much of it AI does:

• Human-only: AI isn't used. The call needs a person's judgement and accountability, like telling someone their role is redundant or deciding whether to close a site.
• AI-assisted: AI helps one person with a task they still own, like an HRBP preparing for a difficult performance conversation or a finance lead drafting a board paper.
• AI-embedded: AI is built into the workflow and completes a significant portion of it, and people do the rest. It ranks each job application for a recruiter to shortlist, or reads every supplier invoice so accounts payable only handles the mismatches.
• AI-run: AI owns the work within limits you set, and people oversee it and handle the exceptions. It books interviews around candidates' and managers' diaries, or chases overdue invoices and passes the stubborn ones to a person.

We design the workflow, the autonomy steps, the governance and the measures. Your technical team or tool vendor configures and connects the tools, and has them ready by the end of week 2 of each quarter.

• Month 1, recommend: the AI finds evidence, frames options and shows its uncertainty. People make the call. By month end, the task sort, the baseline and the first measures are in place.
• Month 2, draft: the AI drafts work products. People review them, and we measure the rework.
• Month 3, act: the AI acts within set limits, wherever the evidence supports it. By month end, we test the eight gates and decide whether to scale.

From week 3, a typical week has the manager routine with the workflow owner, a primary day inside the team, the team running the redesigned workflow between sessions, and the measures updated against the baseline.

A workflow scales only once it passes eight gates, covering evidence, quality, authority, risk, value, adoption, recovery and ownership.

Each quarter tracks four measures from a baseline. Each has a leading signal and a measure of what changed.

• Outcome: cycle time, then the business result the workflow exists for.
• Quality: accuracy, then rework and exceptions.
• Risk: policy adherence, then incidents and overrides.
• Adoption: active use, then work that runs differently.

Logins and prompt counts are supporting signals only.

Advice on choosing AI tools, and training beyond the team and its managers, can be added and priced when needed. Licence costs sit with you.

• Leadership session: your leaders score six foundations in a short pre-read survey, then work through the scores in a half-day session with up to 10 people and choose the first workflow. A written playback within a week names that workflow, the foundations to fix first and the ask.
• Building on either review: the largest opportunities from the business operations or people function review go into the first quarter.
• Fixed first quarter: the first quarter's steps as a fixed project for procurement that needs an end date. It covers one workflow in one team, about 18 days over 12 weeks, paid half on signing and half on delivery.
• Organisation design for an AI strategy: scoped through the leadership session or a short scoping step, then quoted: the operating model, workforce choices and AI governance your strategy needs.

The six foundations are strategy and alignment, trust in AI, change readiness and capability, operating model and process, leadership and accountability, and technology and data.