A future-of-work programme should not be a one-off report that ranks jobs by AI exposure. It should be a repeatable way of learning how work is changing, deciding where to intervene, and helping people move with the change.

The right starting point is modest: a focused pilot in a function with meaningful information work, visible pain points, and leaders willing to redesign work—not merely install a tool. Legal, finance, HR, customer operations, technology, and shared services can all be suitable, depending on the organisation.

Use five practical steps.

1. Select the work, not only the technology.** Choose a role family or service area where the organisation wants a better outcome: faster decisions, better service, less administration, improved control, or capacity for growth. Do not start with a vendor feature and search for a problem.
2. Build and validate activity maps.** Map the 10 to 20 activities that matter most, including time share, systems, skills, pain points, judgment, and risk. Validate the picture with people who perform the work.
3. Assess automation and augmentation separately.** Identify which activities can be improved through embedded assistance, which components may be automated, and which should remain explicitly human-led. Record dependencies such as data quality, policy, integration, and governance.
4. Run measurable interventions.** For augmentation, define the activity, user group, tool, expected benefit, quality controls, and baseline. For automation, define the process boundary, exception model, ownership, and residual work. Measure adoption, quality, time, and employee experience—not only output volume.
5. Convert learning into workforce action.** Update activity maps, refine assumptions, identify skills to build, redesign roles where appropriate, and communicate what has changed. Use pilot evidence to sequence the next investments.

The programme should be augmentation-first in knowledge work, not because automation is undesirable, but because augmentation often reveals where stable automation is genuinely possible and where human judgment remains essential. It also creates a better employee proposition: AI is being used to reduce friction, improve capability, and create capacity—not simply to search for headcount reductions.

Governance should bring HR, business, technology, risk, and process owners together around the same activity-level evidence.

A regular review can ask five questions:

• Which activities are changing fastest?
• Where is value being realised, and how confident are we?
• What new risks or workload effects have appeared?
• Which skills and career pathways need attention?
• What should we automate, augment, redesign, or stop?

The final principle is transparency. Be clear that AI will change work, that not every benefit is immediate, and that estimates are hypotheses to be tested. Employees are more likely to participate when they can see the specific activities in scope, the safeguards in place, and the opportunities to build relevant capability.

The future of work will not be shaped by a single tool or a single forecast. It will be shaped by thousands of choices about individual activities: what we automate, where we augment, what we preserve, and how we help people succeed in the work that remains.