Author: Paul Harman, Partner nem Australasia
Last updated: 30 October 2026
A note from Paul: This series reflects my own views, experience and judgement on leadership in the AI age.
Part 3. Turning AI Ambition into Responsible Action
In Part 1, we explored judgement and responsibility. Part 2 examined how to involve people as their work changes. In this final article, we bring those ideas together: how leaders can turn AI ambition into a practical plan and test whether it creates measurable value.
On Monday morning, the leadership team has heard several proposals for using AI. Marketing wants faster content production. Operations sees an opportunity to reduce administration. Finance wants better reporting. Each idea sounds promising, and each competes for attention.
The leadership challenge is to decide where to begin and how to recognise progress. Start with a problem that matters to the business, has a manageable level of risk and can be tested with the people and information available.
An AI strategy provides direction, but a first experiment needs a specific problem to solve. “Our account managers spend too much time preparing for customer reviews” creates something concrete to investigate.
That problem can be examined before a tool is chosen. Which preparation tasks take the most time? Where does the information come from? Which parts require interpretation? What would a better review look like from the customer’s perspective?
These questions help establish whether AI is appropriate and what success should mean.
Suppose the team decides to test AI-assisted meeting briefs. A useful pilot would have a clear scope, a named owner and a limited duration. It would also have a baseline: how long preparation currently takes, how often information is missing and how useful account managers find the existing briefs. Agree in advance what improvement would justify continuing, and which standards of accuracy and service must be maintained.
Without that starting point, a polished demonstration can easily become the main evidence of success.
The risks of weak planning are familiar from earlier IT projects. In a 2012 article, McKinsey described a bank transformation in which the finance department became involved only a few months before launch. Late changes delayed the launch by more than three months and cost more than US$8 million. For leaders introducing AI, the lesson is to involve the people affected early and agree the requirements, responsibilities and review points before work begins. A plan should make assumptions visible and provide a way to revise them as the team learns.
Measurement should cover the whole task. A brief generated in two minutes may still require substantial checking and correction. Time saved on drafting matters, but so do accuracy, relevance and the quality of the resulting conversation. The aim is to improve the work from beginning to end. Include the cost of tools, setup, training and ongoing support when assessing whether the improvement is worthwhile.
Clear operating boundaries make the experiment easier to manage. The team should know which information may be used, which tool is approved, who checks the output and where human approval is required. They should also know what to do when the system produces something unreliable.
Those boundaries belong in everyday work. Translate the agreed boundaries into a short, usable checklist covering what people may enter into the tool, how they check the output and when they must seek approval.
Make clear who has authority to resolve problems, assess results and decide whether the pilot should continue. The people doing the work should contribute to that assessment, including those who spend time correcting mistakes. Leaders also need to make it safe to report disappointing results. Ask what the team has learnt and what needs to change, so that problems surface early.
At the review point, leaders need to be willing to expand, adjust or stop. A useful experiment can show that the benefit is smaller than expected, or that a simpler process change would solve the problem. That knowledge has value if it informs the next decision.
Expanding the pilot requires a separate assessment. An approach that works for an experienced team handling straightforward cases may need additional support elsewhere. Broader use should follow evidence about the conditions in which it works well.
AI ambition becomes credible when leaders connect it to a real problem, observable results and clear responsibility. That gives the organisation a basis for learning and investing with purpose.
On Monday morning, put three things in place:
Select one problem and establish a baseline. Describe the current difficulty in plain language. Record the time, quality or service measure you want to improve.
Summarise the pilot on one page. Identify the problem, scope, owner, decision-maker, budget and review date. Set out the approved tools and information, success criteria and reasons to pause. Link to supporting technical or risk assessments.
Schedule the review before starting. Examine benefits, costs, correction effort and employee feedback. Decide whether to expand, adjust, gather further evidence or stop.
Across this series, the Monday morning challenge is consistent: exercise judgement, involve people and turn intent into accountable action. Each becomes more important as AI takes a larger role in everyday work.
Source: Bloch, M., Blumberg, S., & Laartz, J. (2012, October 1). Delivering large-scale IT projects on time, on budget, and on value. McKinsey & Company.