As companies grow, they tend to add coordinating headcount faster than they add output. They set up shared services, add governance layers, and blur who decides what. Before long, indirect labour is the biggest cost of running the business, particularly in service-led industries.
Global estimates vary, but research from Deloitte (2024) and the Conference Board (2023) suggests 40-60 per cent of total labour in large enterprises sits in support or administrative functions, once you include consultants, compliance, planning, technology operations and programme management.
For years, companies accepted that as a cost of size. AI changes that.
McKinsey's Global AI Workforce Outlook (2024) projects that up to 70 per cent of current administrative workloads may be automated or heavily augmented within this decade. The early effects are already visible. This is happening now, not in some future you can deal with later.
Where AI hits first: the work customers do not see
For 20 years, automation has focused on the front line: robotics in manufacturing, self-checkout in retail, chatbots in customer support. But most of the effort in an organisation sits behind the scenes.
That work shares three traits. Customers rarely notice it, it is judged on accuracy and timeliness, and until now the effort needed has risen in step with complexity.
Art or science: a practical test for what to change
Harvard Business Review's "When Should a Process Be Art?" (Benson et al., 2007) draws a useful distinction, and it applies well to AI today. Some work is science: the best version is faster and more accurate, and the answer is repeatable. Some work is art: it needs judgement, and the outcome depends on the person doing it.
Many enterprises treat science-like work as if it needed artistry, adding approvals, committees and consultants just in case. AI makes the difference visible, and expensive to get wrong.
Results already in the market
Recent case studies from Bain (2024) and BCG (2023), along with transformations I have supported, show the same pattern.
A Southeast Asian bank redesigned credit operations with AI-driven routing and cut cycle times by 45 per cent and errors by 30 per cent (internal transformation results, 2023).
A US B2B SaaS firm automated analyst reporting, so weekly insights now arrive every morning (Bain Intelligence Automation Report, 2024).
A European insurer put agents on claims validation and avoided millions in outsourced processing costs (BCG Service Automation Case Library, 2023).
Global benchmarks show agent-augmented teams holding service levels while cutting workload by 25-40 per cent (McKinsey Automation Index, 2024). The cost structure shifts. The quality does not.
Where humans stay essential
Gartner's Future of Work Survey (2024) points to the areas where humans keep a proven advantage: relationship-led influence and negotiation; problem solving in ambiguous, novel situations; ethical judgement with accountability; and multi-stakeholder alignment in regulated settings.
AI does the known and scalable work well. People are still needed for the new and the difficult. You need both to get the full economic benefit.
How to start: a responsible roadmap
Across Deloitte (2024) and McKinsey (2023) rollouts, the organisations that succeed follow a similar path.
Map the work by value, and ask where customers directly feel the impact. Sort each process into art or science: if the best version is faster and more accurate, AI belongs there.
Prototype fast, running 4-6 week agent pilots against clear KPIs such as cycle time, error rate and cost per case.
Upskill people inside the work, so analysts become orchestrators who validate and refine what the agents produce. And build governance in from the start, with transparent audit trails and accountable escalation.
The markers of success are consistent: a payback period under 12 months, error reduction above 25 per cent, and 20-30 per cent of time redeployed to customer-facing work (Deloitte Global Shared Services Survey, 2024). When you hit those, scale with confidence.
A better model of operational excellence
AI lets organisations correct something we have always known: bureaucracy grows unless you prevent it (Parkinson's Law; confirmed in Bain Org. Efficiency Study, 2022).
The organisations that win over the next cycle will streamline coordination, put talent where customers feel the difference, use AI to remove effort that never became value, and improve quality rather than only cut cost. The result is a business that spends more of its time creating and less of it processing.
If you would like help working out which processes are ready, even a first review, I am happy to support. It starts with one conversation and one workflow.
Sort every support process into science (repeatable work where the best version is faster and more accurate) and art (work that needs human judgement); automate the science with AI and redeploy people to the ambiguous, relationship-led, high-accountability work where they are worth more.