Turning AI into Productivity: Lessons from 110 Innovation Ecosystems

4 hours ago

6 min read

John H Howard, 6 October 2026

Governments and businesses are investing heavily in artificial intelligence, while the productivity gains remain hard to find in the statistics and uneven across firms, sectors and regions.

Turning AI into Productivity: The Role of Innovation and Industrial Ecosystems asks why AI lifts productivity in some places and stalls in others, and what decision-makers can do about it (Green & Howard, 2026).

The report is the product of a nine-month research project at the University of Technology Sydney, funded by the Google Foundation and written by science, research and innovation policy experts Roy Green and John Howard.

The Minister for Industry and Innovation, and Minister for Science, Senator the Hon. Tim Ayres, launched the report at an event at the Google Foundation in Sydney on 22 September. 2026

This Insight summarises the project's purpose, main findings, and the recommendations that policymakers and practitioners can act upon and implement.

Turning AI into Productivity: The Role of Innovation and Industrial Ecosystems is available in Australia and internationally in paperback and Kindle from the Amazon Bookstore.

The purpose of the project

The project began with a proposition that economists have tested on earlier general-purpose technologies, from the electric motor to enterprise computing.

A new technology lifts productivity only when firms combine it with complementary assets: infrastructure, skills, data, management capability and supporting institutions. The report calls this the complementarity principle, and the authors framed the research program around it.

Three questions followed.

Which ecosystem capabilities best predict success in AI adoption?
How do governance and financing models affect the speed at which AI pilots scale?
Which policy levers produce the largest effect on adoption?

The authors wrote the report for the people who make these decisions in governments, universities, precinct authorities and investment funds.

An evidence base built from the ground up

The research team assembled a case base of 110 Australian and international innovation ecosystems, each profile running between 5,000 and 7,000 words. The case base covers urban innovation districts, metropolitan ecosystems, regional and agtech hubs and energy transition zones across 35 countries.

The team assessed each ecosystem against a common four-domain framework developed by John Howard and published last year in the Handbook of Innovation Ecosystems, and seven performance indicators developed for the project, allowing like-for-like comparison.

Four weeks of site visits during May and June 2026 tested the desk analysis against the judgement of the people who run these places. The team travelled through six countries, from London and Cambridge to Eindhoven, Dortmund, Kaiserslautern, Heilbronn, Munich, Copenhagen and Stockholm, interviewing institute directors, cluster managers, researchers and company representatives.

The European and Nordic focus was deliberate. These are middle-power economies, positioned as Australia is between the United States and China, and they are responding by applying AI to physical products and systems.

The authors are careful about what the evidence supports. They report consistent correlations across a large and diverse case base and do not claim proven causation.

What the research found

The central finding is that AI becomes productive as "industrial AI", applied to a defined purpose in manufacturing, mining, agriculture, logistics and the energy transition. Every country can buy access to the general-purpose layer of models, chips and data centres. Firms and places capture the returns when they combine that layer with their own operational data, skills and redesigned processes.

Five patterns recurred across the field sites:

Large, research-intensive firms such as ASML, Ericsson, Siemens and Bosch anchor the strongest ecosystems and generate the industrial demand for research.
Translation institutes such as Fraunhofer, VTT and DFKI are the connective tissue between research and ordinary firms.
Shared testbeds let firms trial AI in production-like conditions and fail cheaply.
Governance succeeds as facilitation and orchestration.
Proximity and patience count, since every ecosystem visited took decades to build.

The research team also confirmed four findings with a long lineage in Australian policy debate. Australia's gap between research and innovation reflects industrial structure more than research quality, because the country lacks the large research-intensive firms that convert collaboration into demand. Additional public research funding alone is unlikely to close that gap.

Firms run successful AI pilots and then struggle to move into production. The constraint practitioners cited most often was management capability, ahead of capital or technology. Firms tend to build that capability inside ecosystems, where they can draw on shared facilities, peer learning and translation institutes.

The report then distinguishes assistive adoption, where AI amplifies professional judgement, from agentic adoption, where systems act autonomously. The field evidence suggests that firms with deep expertise may gain more from the assistive path. The authors present this as a working hypothesis, implying that the productive path and the human-centred path may be the same.

The fourth finding concerns who gains, which the report calls the "migration of value".

People, firms, and regions with the complements capture the returns from AI. Regions without them take part in the same economy as consumers of AI services, and the value accrues elsewhere.

What the findings mean for Australia

Applying the frameworks to Australian precincts such as Parkville, Tonsley, Tech Central, Lot Fourteen and Bradfield produces a consistent profile. Placemaking and research are strong. The precincts score lower on translation capability, procurement pathways and systems integration, which are the indicators the field evidence identifies as decisive.

The authors treat this profile as encouraging, because it identifies where effort is likely to pay. Australian precincts do not need rebuilding from the ground up. They need the connective layer that converts existing assets into adoption. The report makes seven recommendations to build it:

making industrial AI a central focus of the National AI Plan
targeting areas of competitive advantage through smart specialisation
attaching compute access conditions to data centre approvals
building a national network of collaborative, industry-led innovation ecosystems on the existing precinct portfolio
committing universities and CSIRO to industrial AI and to translation partnerships with mid-sized firms
building management and workforce capability through extension-style services
mobilising patient capital and public demand to grow anchor firms.

Insights for policymakers and practitioners

Several insights from the report seem especially useful for people deciding where to act.

The Australian public debate on AI has concentrated on safety, security and sovereignty. The report approaches AI as a question of productivity, industrial capability and place, and asks what Australia would need to build as well as what it needs to guard against.
The least glamorous intervention may be the most productive. Hands-on advisory services that build management capability in mid-sized firms, delivered by translation institutes with engineering credibility, may return more than further technology subsidies. Programs that subsidise technology acquisition alone risk funding pilots that firms can't scale.
Public demand can substitute for the anchor firms Australia largely lacks. A young firm's first major customer is often a public one. Offtake commitments in energy, defence, health, water and transport create the recurring industrial problems from which capable firms and translation institutes grow, provided procurement pathways are open to young and mid-sized firms.
The design of compute access depends on the type of facility. Hyperscale investment in Australia leans towards remote training capacity, while Australian firms mostly use inference, which metropolitan facilities close to users supply. A reservation scheme framed around inference capacity, combined with vouchers and tool access, could reach the firms and researchers it is meant to help.
Measurement needs to start early. Policymakers can't yet tell from the data now collected whether adoption is raising productivity. The report proposes a short annual return attached to public investment in AI, and precinct-scale measurement built from linked administrative data, so that governments can observe diffusion from a recorded baseline.

The last insight is about time and confidence. Eindhoven rebuilt after a corporate crisis, Dortmund after coal and steel, and Kaiserslautern from a sewing-machine works. Practitioners in each place doubted any plan that promised results inside a single political or funding cycle.

The report's closing message is that "places make the difference, and places can be made" (Green & Howard, 2026).

References

Green, R., & Howard, J. H. (2026). Turning AI into productivity: The role of innovation and industrial ecosystems. University of Technology Sydney.

Howard, J. H. (2025). Handbook of Innovation Ecosystems: Placemaking. Economics. Business. Governance. Acton Institute for Policy Research and Innovation