How enterprises are turning AI adoption into value

Simon Benson, Regional AI Lead for ASPAC at KPMG Australia, spoke with The AI Journal about how enterprises are turning AI adoption into value, drawing on findings from KPMG’s Q3 Global AI Pulse Survey. 

In 2026, leading organisations are making a decisive shift: AI is no longer a standalone technology programme, but a core part of business management. This Q&A explores Simon’s thoughts as well as some of the report’s key findings and their implications for enterprise leaders. 

The AI Journal: What is the central message from KPMG’s Q3 Global AI Pulse 2026? 

Simon Benson: The age of adoption is giving way to the age of value. In 2025, the priority was putting enterprise-safe AI tools into employees’ hands. In 2026, the focus is on translating that access into measurable business outcomes. Organisations are beginning to record both costs and benefits, but relatively few can yet connect a specific AI investment to the value it creates. 

The AI Journal: Before examining those outcomes, what is the scope of the research? 

Simon Benson: The quarterly global study draws on approximately 2,131 respondents across around 20 countries. Q3 is the third global edition, allowing KPMG to identify emerging patterns across regions while avoiding direct comparisons with earlier, predominantly US-focused research. 

The AI Journal: What does that global picture tell us about how enterprise AI spending is changing? 

Simon Benson: Average planned AI investment has risen to about US$210 million, from roughly US$186 million in Q1. The increase reflects both greater confidence and higher costs as organisations move beyond prompting and basic automation towards more complex development and multi-agent orchestration. Spending has also become more consistent across the Americas, Asia-Pacific and Europe. 

The AI Journal: As investment rises, are organisations seeing a return? 

Simon Benson: Some are, but the ability to measure returns remains uneven. In Q2, only about 8% of organisations were identified as achieving ROI from AI. The picture is improving, yet many businesses still track spending and benefits separately rather than linking them. The leading organisations can show how particular investments contribute to productivity, better decisions or other defined outcomes. 

The AI Journal: Where returns are emerging, what does AI ROI look like today? 

Simon Benson: Returns are still predominantly associated with productivity and cost efficiency. These gains range from support for individual tasks to workflow automation and faster, more evidence-based management decisions. Large-scale revenue generation remains less common, so leaders should define value broadly but measure each use case against a clear operational or commercial outcome.  

The AI Journal: What distinguishes the organisations that are translating those gains into measurable value? 

Simon Benson: They embed AI in the management layer. This means assigning clear accountability to an individual or team, integrating AI into the operating model and coordinating the functions required to deploy it responsibly. Cybersecurity, data, technology, procurement, risk and business continuity cannot operate as separate workstreams; they need to be orchestrated around shared business objectives. 

The AI Journal: Conversely, what is holding less mature organisations back? 

Simon Benson: Common barriers include access to the right models in the right jurisdictions, data-sovereignty requirements, security concerns and fragmented decision-making. More fundamentally, some businesses still run AI as a separate programme. Without integration into core processes, governance and management, spending can rise without a clear route to value. 

The AI Journal: Given those cost and governance pressures, can open-source models help enterprises? 

Simon Benson: Interest in open-source models is growing, partly because they can offer greater flexibility and potential cost advantages. However, enterprises must balance those benefits against security, risk and governance requirements. Model choice should therefore be based on the needs and risk profile of each use case rather than cost alone. 

The AI Journal: Beyond model choice, how are organisations addressing data sovereignty and liability? 

Simon Benson: Enterprises are increasingly using cloud and data-platform intermediaries to keep sensitive data within controlled environments while sending only the necessary processing component to an AI model. Model availability is also expanding across jurisdictions, easing some geographic concerns. At the same time, liability and indemnification are becoming more prominent in supplier negotiations, with some providers showing greater flexibility than under traditional standard terms. 

The AI Journal: How do these wider risk considerations extend to cybersecurity? 

Simon Benson: AI is intensifying both the threat and the defence. Organisations need to use AI to test their own environments, identify vulnerabilities and accelerate protection, or risk being overwhelmed by the speed and volume of attacks and patches. Because faster patching can affect stability and user experience, leaders are increasingly prioritising remediation according to the potential impact of each vulnerability.