Organisations with strong AI governance, data quality, explainability and accountability are 15 times more likely to report strong returns from artificial intelligence investments, according to new global research.
Companies with strong trustworthy AI practices are 15 times more likely to report strong or high returns on their artificial intelligence (AI) investments, according to new global research. The Data and AI Impact Report: The New Economics of Trust, from SAS and featuring research insights from International Data Corporation (IDC), found that 62% of organisations investing strongly in trustworthy AI reported strong or high ROI, compared with just 4% of organisations identified as trustworthy AI laggards.
Organisations with the strongest trustworthy AI practices also achieved 1.85 times greater gains across 13 business outcomes, including revenue growth, cost savings and customer experience. The findings add to growing evidence that the debate around responsible AI is shifting. Governance, explainability and human oversight are increasingly not simply questions of compliance or ethics, but factors that could determine whether organisations achieve meaningful returns from the technology.
That could have significant implications as businesses pour money into AI while facing growing pressure to demonstrate measurable returns. The research, based on a global survey of 2,699 decision-makers across 28 countries, also reveals a significant human trust problem. It found that 97.2% of users override AI-generated recommendations in at least some circumstances. The number-one reason? AI’s inability to explain how it reached its decision.
TRUSTWORTHY AI LEADERS REPORT SIGNIFICANTLY STRONGER RETURNS
The research reveals a striking ROI divide between organisations with mature trustworthy AI practices and those lagging behind. Organisations investing in trustworthy AI measures were 15 times more likely to report strong or high ROI from their AI projects – 62% compared with just 4%.
Those with the strongest trustworthy AI practices also reported 1.85 times greater gains across 13 different business outcomes, including revenue growth, cost savings and customer experience. And organisations already leading on trustworthy AI appear determined to widen that advantage. Some 85% of these AI leaders are increasing investment in trustworthy AI by more than 10% this year.
The findings are particularly significant as organisations face increasing pressure to demonstrate that years of investment in AI are delivering measurable business results. Earlier research revealed that seven in 10 companies could slash AI budgets if returns continue to disappoint. Nearly 70% of executives surveyed were prepared to reduce AI spending if business goals were not achieved, while 73% reported that at least some AI investments had failed to meet expectations.
The latest findings suggest the organisations extracting the greatest value from AI may not necessarily be deploying fundamentally different technology. They may be managing it differently.
RESPONSIBLE AI PART OF THE BUSINESS CASE
That distinction challenges an important assumption surrounding AI adoption. Governance, human oversight, auditability and explainability can be regarded as additional layers of complexity that slow innovation and increase deployment costs. The SAS findings suggest the opposite may be true: organisations with stronger trustworthy AI practices are also substantially more likely to report stronger returns.
Bryan Harris, Chief Technology Officer at SAS, said accuracy and repeatability become particularly important as AI systems undertake increasingly complex work. “When AI works, it’s incredibly impactful,” said Harris. “However, it is well documented that state-of-the-art agents can have error rates that exceed 25% on complex tasks – which is unacceptable in high-stakes decision-making.”
In order to achieve accuracy and repeatability, “organisations must embed domain expertise into agentic workflows, while keeping people at the center of governance and oversight,” added Harris. “Organisations that do this successfully will close the trust gap and gain a competitive advantage in the market with AI.”
The research does not establish that trustworthy AI practices alone cause higher returns. However, the association between stronger governance and stronger reported ROI adds an important commercial dimension to the responsible AI debate.
MANY USERS OVERRIDE AI RECOMMENDATIONS
One of the most revealing findings concerns what happens when employees do not trust the systems they are expected to use. Some 97.2% of users override AI-generated recommendations in at least some cases. Significantly, the number-one reason employees chose to override AI, regardless of whether its output was considered correct – was that the system could not explain how it reached its decision.
Trust also appears to decline as AI becomes more autonomous. The research found trust falls from 76% for generative AI to 66% for agentic AI. That creates a potentially expensive organisational problem. Businesses can invest heavily in AI systems, but if employees do not understand or trust their recommendations, they may manually check, correct or override outputs. That can erode some of the productivity gains the technology was intended to deliver.
Previous research has already highlighted this hidden cost. Two-thirds of employees spend up to six hours each week correcting poor-quality AI-generated work or “workslop”, suggesting poorly implemented AI can shift work around organisations rather than eliminate it. Explainability therefore matters not only for regulators, risk teams and boards. It can directly influence whether employees trust, adopt and effectively use AI at work.
CHALLENGE OF MAINTAINING TRUST
That challenge is likely to intensify as organisations move from generative AI tools towards more autonomous AI agents capable of carrying out multi-stage tasks and making recommendations with less human involvement.
Chris Marshall, Vice President at IDC, said organisations now face the challenge of maintaining trust in increasingly autonomous systems. “As AI becomes more autonomous, organizations face a new challenge: maintaining confidence in systems people don’t fully understand,” said Marshall. “Our findings show that stronger oversight, explainability, accountability and data foundations are becoming prerequisites for scaling AI successfully.”
That conclusion echoes other recent evidence about the risks organisations face as AI assumes a greater role in business processes. Recent research found that one in 10 publicly reported technology incidents now involves AI, with researchers warning that AI capabilities may be evolving faster than many organisations’ governance and resilience frameworks. The issue becomes even more significant when AI begins influencing high-stakes financial, employment, regulatory or strategic decisions.
AI CONFIDENCE AND GOVERNANCE
The findings are particularly timely given growing evidence of a gap between organisational confidence in AI and the controls needed to govern it. Separate global research from Workiva found that 84% of executives are at least somewhat confident in the accuracy of AI-generated output without human review, despite just 11% believing their organisation’s data quality is sufficient for AI use.
Even more significantly, 26% said internal audits had detected AI errors that had already reached external audiences or board members. The contrast between the two studies is striking.
The Workiva findings highlight the potential consequences when confidence in AI runs ahead of data quality and controls. The new SAS research suggests organisations with stronger governance, explainability, accountability and data foundations are also substantially more likely to report stronger returns from AI.
Together, they strengthen the argument that AI governance should not simply be regarded as protection against something going wrong. It may also help create the conditions required for AI to work well.
WEAK DATA FOUNDATIONS REMAIN A MAJOR BARRIER
Yet relatively few organisations appear to have the data infrastructure required for the next generation of AI. Just 17.5% of enterprises surveyed have a fully optimised data infrastructure mature enough for the demands of agentic AI.
The financial implications appear significant. Organisations with optimised data foundations are four times more likely to expect strong ROI from AI projects. They are also six times more likely to mandate the data-quality and explainability controls necessary to build trust. The findings reinforce the importance of data quality as businesses move towards increasingly autonomous systems.
AI systems ultimately depend on the information organisations provide them. Poor, fragmented, outdated or unreliable data can undermine even sophisticated technology – particularly when AI-generated recommendations influence important decisions. That makes data governance a business and leadership issue, not simply an IT responsibility.
SHADOW AI: ANOTHER TRUST CHALLENGE
Organisations also need to understand how employees are actually using AI. As Fair Play Talks recently reported, workers are adopting AI faster than many employers can govern it. Across 72 organisations analysed in that research, employees actively used almost 50 different AI applications, with consumer tools including ChatGPT, Gemini and Claude among those being used.
The growth of so-called “shadow AI” creates potential risks around confidential information, intellectual property, compliance, data protection and inconsistent AI outputs. It also highlights an important limitation of formal governance.
An organisation may establish robust controls around its official enterprise AI systems while employees simultaneously use consumer AI tools outside those environments. Trustworthy AI therefore requires organisations to understand not only the technology they purchase, but how AI is actually being used across their workforce.
TRUST ALSO DEPENDS ON WORKPLACE CULTURE
The human dimension of AI trust extends beyond explainability. Employees also need to feel able to question AI outputs, disclose how they have used the technology and report mistakes when something goes wrong.
As Fair Play Talks has explored, AI transparency increasingly depends on psychological safety in the workplace. This becomes particularly relevant given separate research revealing that half of Gen Z workers feel guilty about using AI, while more than four in 10 are hiding their AI use from employers.
If workers believe admitting AI use will make them appear lazy, less competent or professionally vulnerable, they may hide how they use the technology. Similarly, employees who fear repercussions for questioning an AI-generated recommendation may remain silent even when they believe something is wrong.
That makes workplace culture an important part of trustworthy AI. Responsible governance needs employees who are willing and able to challenge the technology – not simply systems that tell them they are permitted to do so.
AI GOVERNANCE AND FINANCIAL CONTROL
The stakes increase further when AI enters financial processes. Previous research found that 40% of US employees and 29% of UK workers surveyed had used AI to create or manipulate expense receipts, demonstrating how generative AI can challenge established financial controls.
More recent research into finance functions revealed that nearly nine in 10 finance professionals had ignored suspected workplace fraud. That study highlighted another AI governance challenge: while 90% said there was a financial or compliance threshold requiring human approval regardless of an AI system’s accuracy, 46% said those thresholds were understood but not formally documented.
Meanwhile, 45% of finance leaders said their teams often act on AI-generated recommendations without human intervention. Taken together, these findings demonstrate why trustworthy AI depends on more than technical accuracy. Organisations also need clear accountability, documented controls, transparent decision-making and cultures in which people are prepared to challenge questionable outputs.
WHAT MAKES AI TRUSTWORTHY?
The SAS research assessed organisations against five dimensions of trustworthy AI:
- Data quality and governance
- Model governance and oversight
- Explainability and fairness
- Responsible AI policy
- Audit and accountability
Organisations received a score out of 100, with those averaging 80 or above classified as trustworthy AI leaders. The framework is significant because it treats trust as something organisations need to build systematically rather than simply expecting employees or customers to have confidence in AI.
SAS defines trustworthy AI as artificial intelligence designed to be reliable, fair, secure, compliant with relevant regulation and capable of showing how it reached a decision. Users and decision-makers should also be able to hold AI systems to predetermined chains of accountability when outputs are wrong or missing.
EMERGING DIFFERENCES BETWEEN INDUSTRIES
The global research also identified significant differences in how industries are approaching trustworthy AI. In banking, 85% of AI leader banks have established governance frameworks, compared with just 29% of laggards.
The findings suggest leading banks are moving beyond treating AI governance purely as a compliance requirement and increasingly regarding it as an operational and competitive capability. Within the public sector, 41% of leaders are increasing trustworthy AI investment by more than 20% during the year ahead, putting them among the most ambitious organisations surveyed.
Life sciences organisations, meanwhile, appear particularly advanced in scaling the technology, with 23% having deployed AI company-wide – the highest proportion among the four industries examined. The findings suggest the trustworthy AI debate is increasingly moving from broad principles towards practical questions about how organisations build governance into everyday AI deployment.
WHAT EMPLOYERS & BUSINESS LEADERS SHOULD DO NEXT
The research provides an important message for organisations trying to move AI projects from experimentation towards measurable business value: trust needs to be designed into AI adoption rather than added afterwards.
Get the data foundation right
Only 17.5% of enterprises surveyed have fully optimised data infrastructure for agentic AI. Before organisations scale autonomous systems, they should understand whether the underlying information is accurate, accessible, governed and appropriate for the decisions AI will be expected to support.
Make explainability a practical requirement
The fact that inability to explain a decision is the number-one reason employees override AI demonstrates that explainability has operational consequences. Employees need enough information to understand why an AI system has made an important recommendation and when they should question it.
Keep meaningful human oversight
Human involvement should not simply mean adding an approval box at the end of an automated process. Organisations should define where human judgement is required, who has authority to challenge AI outputs and who remains accountable when AI contributes to a consequential decision.
Establish clear AI accountability
Businesses should document who owns AI-related decisions and failures. That includes establishing what AI systems are permitted to decide, where mandatory human approval begins, how decisions are recorded and audited and who ultimately carries responsibility when something goes wrong.
Understand why employees override AI
Overrides should not automatically be treated as resistance to technology. They may reveal poor explainability, unreliable data, inappropriate use cases or a lack of confidence in how a system operates. Tracking why employees override AI could therefore provide organisations with valuable information about where technology and governance need to improve.
Address shadow AI rather than iignoring it
Employees are already using consumer AI platforms, whether organisations formally approve them or not. Businesses need clear acceptable-use policies, secure alternatives, appropriate training and cultures in which employees can discuss how they use AI openly.
Build psychological safety alongside technical controls
Employees should be able to say “I don’t trust this output”, “I don’t understand how the AI reached this decision” or “the AI may have made a mistake” without fearing professional consequences. Governance systems work only when people feel able to use them.
Build AI literacy across leadership
The challenge is not confined to employees. Previous research highlighted by Fair Play Talks found that 70% of workers believe they are ready for AI, while only 39% of leaders believe their employees are prepared. Other global research has warned that businesses are still underestimating the people challenges associated with AI transformation. Leaders therefore need sufficient AI literacy to ask difficult questions about outputs, data, risk and accountability rather than treating AI governance solely as the responsibility of technology teams.
Measure ROI alongside trust
Organisations should assess AI performance using more than adoption rates. Measures could include productivity, error rates, employee overrides, time spent correcting outputs, customer outcomes, compliance incidents and the quality of decisions. The key question is not simply how much AI an organisation uses, but whether people can trust it enough for that use to create sustainable value.
WHY THIS MATTERS FOR RESPONSIBLE BUSINESS
For much of the AI boom, businesses have faced an apparent tension: move quickly enough to capture the benefits of AI while putting enough governance around the technology to use it responsibly. The latest findings challenge the assumption that organisations necessarily have to choose between the two.
Trustworthy AI leaders are 15 times more likely to report strong or high ROI and achieve 1.85 times greater gains across 13 business outcomes. Meanwhile, 85% are continuing to increase their investment in trustworthy AI practices.
But the findings also provide an important warning against equating AI adoption with progress. Virtually all users override AI recommendations at least sometimes. Trust declines as systems become more autonomous. Only a small minority of organisations have fully optimised data infrastructure for agentic AI. And other recent research has shown AI errors reaching boards, employees quietly using unapproved tools and organisations spending significant amounts of time correcting poor-quality AI outputs.
TRUST: AN ECONOMICAL AND MORAL ASSET
Taken together, the evidence suggests trust is becoming an economic as well as an ethical asset. If employees do not understand AI, they override it. And if businesses cannot trust their data, they cannot confidently scale it. Furthermore, if boards, customers and regulators cannot establish how decisions were reached, organisations assume greater risk. And if employees spend their time correcting AI-generated mistakes, promised productivity gains can quickly disappear.
The biggest question for businesses may therefore no longer be simply how quickly they can deploy AI, but whether they are building the governance, data, accountability and human trust required to make those investments work.
Responsible AI should not be viewed simply as the price businesses have to pay for innovation. It may increasingly be one of the conditions required to make innovation pay.
Access the full Data and AI Impact Report here.
RELATED ARTICLES
A global survey shows that nearly 70% of companies are prepared to scale back AI budgets if projects fail to deliver measurable business results.
Two-thirds of employees say they spend up to six hours a week fixing low-quality, AI-generated work-slop, according to a new study.
One in 10 reported technology incidents now involves artificial intelligence (AI), according to a new report.
Despite 84% of executives expressing confidence in AI-generated output without human review, just 11% believe their organisation’s data quality is sufficient for AI use, exposing a potentially dangerous gap between confidence, governance and reality.
Employees are embracing AI faster than many organisations can govern it, according to new research.
AI transparency policies will only succeed when employees feel psychologically safe to disclose how they use AI at work.
Half of Gen Z workers feel guilty using AI and more than four in 10 are using it without their employer’s knowledge, despite businesses increasingly demanding AI skills from new recruits.
Four in ten US employees admit they have used AI to create a fake business expense receipt, while almost three in ten UK staff say they have used AI to generate or manipulate receipts to supplement their income.
Workplace fraud may be becoming increasingly normalised, with nearly nine in 10 of finance professionals admitting they have ignored a claim they believed was fraudulent.
Sign up for our newsletter
