AI could deliver double-digit productivity gains — but governance needs to catch up: Report
AI is delivering 8.2%-12.3% average net productivity gains across industries, but companies are still building the governance frameworks needed to manage its risks. Morgan Stanley says regulatory uncertainty, execution challenges and gaps in AI controls could become bigger concerns as adoption accelerates.
- Aug 28, 2026,
- Updated Aug 28, 2026 1:27 PM IST
Companies across a range of industries are reporting significant productivity gains from artificial intelligence, with average net improvements reaching as high as 12.3%, even as corporate governance frameworks and model controls remain a work in progress, according to a Morgan Stanley Institute for Sustainable Investing report.
The report, Governance and AI Adoption, draws on Morgan Stanley Research interviews with C-suite professionals and a separate survey of 200 executives involved in AI governance at global companies.
AI productivity gains reach 12.3%
Morgan Stanley Research’s Alphawise team conducted 808 online interviews with C-suite professionals in April 2026 covering banks, software and services, technology hardware and equipment, semiconductors and professional services. This was combined with an earlier set of 935 interviews conducted in October 2025 across transportation, autos, consumer staples, healthcare and real estate.
Across 10 industry groups covered by the two surveys, companies estimated average net productivity gains from AI implementation over the previous 12 months at between 8.2% and 12.3%.
MUST READ: Big victory for Anthropic: US court strikes down Pentagon blacklist over 'illegal retaliation'
The gains were not restricted to technology-heavy industries. Morgan Stanley said productivity improvements were also identified in sectors such as consumer staples distribution and retail, where board-level representation of AI skills is relatively limited.
The areas benefiting from AI are similarly broad. Customer service, marketing, operations and supply chain, sales and human resources were among the functions cited by companies.
Governance struggles to keep pace
While the productivity opportunity is becoming clearer, companies are still developing the systems needed to manage AI-related risks.
Among 200 executives surveyed globally, 48% said most of their existing risk-management structure should be sufficient with some changes, while another 39% said their existing framework should work but that they were still assessing it.
Nearly 90% said their companies had either defined or were in the process of defining where responsibility for AI governance and risk management sits within the organisation.
However, more specific model controls have lower implementation rates. Only 41% of companies said a formal inventory of AI models with assigned ownership was fully in place, while the same proportion had formal testing and validation protocols.
Defined standards for AI model design and data use were fully implemented at 39% of companies, while 33% had a model-risk classification and tiering system. Structured mechanisms to identify unintended or emerging risks had the lowest full implementation rate, at 27%.
ALSO READ: India’s new AI export could be human motion data
Regulation emerges as future risk
The governance challenge could become more complex as AI adoption accelerates. Data-related risks, including privacy, cybersecurity and discrimination, are currently the biggest concern, with 56% of executives calling them the most material AI risk.
Looking ahead two to three years, however, 35% expect regulation and legal risks to be the biggest concern, while 32% point to execution risks, including difficulties in achieving returns from AI investments.
Morgan Stanley also found that regulatory complexity is the biggest barrier to effective AI governance, cited as the main obstacle by 38% of respondents.
The findings suggest that while AI is already producing measurable productivity benefits, companies may need to strengthen governance frameworks at the same pace as they scale adoption.
MUST READ: OpenAI loses another executive: Chris Malone, head of data centres exits amid executive shake-up
For Unparalleled coverage of India's Businesses and Economy – Subscribe to Business Today Magazine
Companies across a range of industries are reporting significant productivity gains from artificial intelligence, with average net improvements reaching as high as 12.3%, even as corporate governance frameworks and model controls remain a work in progress, according to a Morgan Stanley Institute for Sustainable Investing report.
The report, Governance and AI Adoption, draws on Morgan Stanley Research interviews with C-suite professionals and a separate survey of 200 executives involved in AI governance at global companies.
AI productivity gains reach 12.3%
Morgan Stanley Research’s Alphawise team conducted 808 online interviews with C-suite professionals in April 2026 covering banks, software and services, technology hardware and equipment, semiconductors and professional services. This was combined with an earlier set of 935 interviews conducted in October 2025 across transportation, autos, consumer staples, healthcare and real estate.
Across 10 industry groups covered by the two surveys, companies estimated average net productivity gains from AI implementation over the previous 12 months at between 8.2% and 12.3%.
MUST READ: Big victory for Anthropic: US court strikes down Pentagon blacklist over 'illegal retaliation'
The gains were not restricted to technology-heavy industries. Morgan Stanley said productivity improvements were also identified in sectors such as consumer staples distribution and retail, where board-level representation of AI skills is relatively limited.
The areas benefiting from AI are similarly broad. Customer service, marketing, operations and supply chain, sales and human resources were among the functions cited by companies.
Governance struggles to keep pace
While the productivity opportunity is becoming clearer, companies are still developing the systems needed to manage AI-related risks.
Among 200 executives surveyed globally, 48% said most of their existing risk-management structure should be sufficient with some changes, while another 39% said their existing framework should work but that they were still assessing it.
Nearly 90% said their companies had either defined or were in the process of defining where responsibility for AI governance and risk management sits within the organisation.
However, more specific model controls have lower implementation rates. Only 41% of companies said a formal inventory of AI models with assigned ownership was fully in place, while the same proportion had formal testing and validation protocols.
Defined standards for AI model design and data use were fully implemented at 39% of companies, while 33% had a model-risk classification and tiering system. Structured mechanisms to identify unintended or emerging risks had the lowest full implementation rate, at 27%.
ALSO READ: India’s new AI export could be human motion data
Regulation emerges as future risk
The governance challenge could become more complex as AI adoption accelerates. Data-related risks, including privacy, cybersecurity and discrimination, are currently the biggest concern, with 56% of executives calling them the most material AI risk.
Looking ahead two to three years, however, 35% expect regulation and legal risks to be the biggest concern, while 32% point to execution risks, including difficulties in achieving returns from AI investments.
Morgan Stanley also found that regulatory complexity is the biggest barrier to effective AI governance, cited as the main obstacle by 38% of respondents.
The findings suggest that while AI is already producing measurable productivity benefits, companies may need to strengthen governance frameworks at the same pace as they scale adoption.
MUST READ: OpenAI loses another executive: Chris Malone, head of data centres exits amid executive shake-up
For Unparalleled coverage of India's Businesses and Economy – Subscribe to Business Today Magazine