Why can AI do so much while business change takes so long?

AI capability can advance faster than a business’s ability to put it to work. My view is that the gap involves more than access or technical skill. Leaders have to make sense of the change, decide what is worth pursuing and alter the routines through which work gets done.

Earlier research helps explain why these are separate challenges. It provides a useful starting point, while leaving important questions about AI unresolved.

Spending money and changing work are different

Clark Gilbert’s 2005 study of eight newspaper organisations distinguished changes in resource commitments from changes in operating routines. A strong sense of threat could unlock investment while making established ways of working more rigid. Spending more did not necessarily mean adapting well. Gilbert, 2005.

Imagine a firm buying AI licences, appointing a lead and funding pilots while leaving decision rights, incentives and everyday processes largely intact.

For example, producing a draft faster may achieve little if it then waits in the same approval queue. The model has improved one task; the surrounding process still determines how much of that improvement reaches the customer.

What leaders think the change means matters

Tripsas and Gavetti’s study of Polaroid offers a related lesson. The company invested in digital imaging and developed technical expertise, yet senior managers’ attachment to its established business model constrained how that expertise was commercialised. Recognising a technology and understanding what it might require of the business were different achievements. Tripsas and Gavetti, 2000.

Later research by König and colleagues, across book retailing and telecommunications, separated expectations of gain or loss from perceived control. Seeing a possible benefit does not necessarily mean believing the firm can secure it; recognising a threat does not establish that leaders see a workable response. Their findings also made strategic relevance important to understanding resource commitment. König, Graf-Vlachy and Schöberl, 2021.

For AI, this suggests looking beyond whether managers are enthusiastic or worried. What do they think will change in their own business? Which decisions seem available to them? What would make an experiment worth the effort? Those questions connect awareness to action.

Direct AI evidence adds detail, rather than one universal answer

There is evidence of substantial gains in particular settings. Brynjolfsson, Li and Raymond’s 2025 study followed the staggered introduction of an AI assistant among 5,172 customer-support agents serving one company. They estimated a 15% average increase in issues resolved per hour, with larger benefits for less experienced and lower-skilled workers. The system offered suggestions within an existing workflow, and agents retained responsibility for the conversation. This was evidence about a specific deployment, rather than a return every business should expect from current models. Brynjolfsson, Li and Raymond, 2025.

The OECD’s 2025 report, drawing on a 2024 survey of SMEs in seven countries, found reported benefits alongside limited preparation: a third or fewer of users had undertaken measures such as staff training or internal guidelines. Australia was outside the sample. Use and preparation can therefore differ, but the survey cannot establish which support would improve Australian outcomes. OECD, 2025.

Why the question still needs research

The historical studies concern particular industry transitions, largely in established organisations. They cannot tell us how common those mechanisms are among Australian SMEs, or how they operate when AI can assist both everyday tasks and the thinking involved in changing a business.

A smaller firm may have fewer layers of approval but less time and specialist support for experimentation. Neither greater agility nor greater inertia should be assumed. Some uses may justify immediate action; others may remain unreliable, uneconomic or irrelevant.

This is why my research examines how Australian SME leaders interpret AI and how those interpretations relate to their accounts of decisions, resource commitments and changes in practice. Existing theory helps identify worthwhile questions. The explanation must remain open to what the cases reveal.

What follows in practice

My position is that useful adoption deserves practical support. That means helping firms connect a worthwhile use to the changes needed around it, with affected staff involved and someone clearly responsible for the result.

Evaluation should include checking and rework, service quality, staff learning and the distribution of workload alongside time and cost savings. As tools change, firms need to reassess their uses and consider whether their information and workflows can move to another provider.

The objective is sustained capability: being able to do useful work reliably, learn from experience and change course when an approach fails. A licence, a pilot or an adoption statistic tells us only part of that story.