Data Quality and Governance Essential for Scaling AI in Organizations
In a landscape increasingly dominated by artificial intelligence (AI) and machine learning, organizations focusing on robust data governance and accountability are yielding significantly higher returns from their AI investments. A recent survey conducted by SAS and IDC highlights the correlation between effective data management practices and financial performance in the realm of AI.
The findings indicate that companies with diligent practices surrounding trustworthy AI are an astounding 15 times more likely to report noteworthy or high returns on their investments compared to those with weaker methodologies. This was concluded based on an extensive survey analyzing the responses of 2,699 data and AI decision-makers across 28 countries. The survey delved into various areas, including data quality, model oversight, explainability, responsible AI initiatives, and accountability frameworks.
Among the organizations categorized as leaders in the report—those that achieved a weighted index score of 80 out of 100 or higher in the five surveyed domains—62% reported returns of at least $2 for every $1 invested. In stark contrast, a strikingly low 4% of respondents from the lowest-scoring firms reported similar levels of success. This disparity emphasizes the crucial role of structured governance in leveraging AI for business gains.
However, Udo Sglavo, the vice president of applied AI and modeling research and development at SAS, cautions against interpreting these findings as definitive proof of a direct causal relationship. According to Sglavo, organizations that govern AI effectively also tend to excel in its deployment, monitoring, and application of insights drawn from AI systems. "I don’t think we make the claim that this is a causal relationship," he remarked during an interview with Information Security Media Group (ISMG). This statement underscores the complexity of deriving connections between governance and return on investments in AI.
Furthermore, the findings reveal a mixed level of trust in different types of AI. Approximately 76% of respondents indicated trust in generative AI, while only 66% expressed confidence in agentic AI, despite a noteworthy 89% acknowledgment that AI agents have already taken on roles in organizational decision-making.
User trust plays a pivotal role in the adoption of AI solutions, heavily influenced by the clarity and accuracy of AI outputs. The survey indicates that 79% of respondents often override AI recommendations in over 10% of their instances. The primary reasons for these overrides include a lack of explanations for AI decisions, cited by 34.5% of the respondents, while only 14.8% pertained to factual errors in AI outputs.
Acknowledging the need for transparency, Sglavo stated that while users might not need to grasp every detail of the calculations within a complex AI model, it is crucial that the ultimate decisions derived from it are comprehensible to employees, customers, and regulators. This aspect becomes even more critical in regulated industries, where institutions need to assume responsibility for outcomes driven by AI. For instance, if an AI system influences a credit decision, the financial institution cannot simply defer blame to the AI, as highlighted by Sglavo’s comments.
Moreover, while human oversight is deemed necessary for significant decisions, it’s important to note that intervention does not automatically guarantee improved results. "We shouldn’t automatically assume that because the human gets involved, the result gets improved," he cautioned. This necessitates organizations tracking instances of overrides to assess when human input adds value and when it might inadvertently introduce inconsistencies or costs.
Data optimization poses another hurdle for many organizations as per the survey results. A mere 17.5% of respondents reported having fully optimized data environments that could support the lineage, validation, and explainability essential for agentic AI systems. The study also found that entities with mature data environments tend to exhibit stronger validation practices and have heightened expectations regarding returns on AI.
Sglavo noted a distinct pattern among the companies recognized as leaders: they prioritize problem identification first, define metrics for success second, and select tools for implementation only thereafter. This structured approach provides a strategic roadmap for successfully harnessing AI capabilities.
Ultimately, he recommended a practical approach to AI deployment—starting with tasks that are repetitive, easily measurable, and typically carry low risks before advancing to more consequential decisions. "Let’s apply AI on the boring stuff first," Sglavo advised. His concluding remark on governance underscores the notion that effective governance should not serve as a bottleneck but rather as a framework for making the most of AI capabilities, bolstering confidence when navigating a predominantly AI-infused future.
In conclusion, the study underscores the importance of data quality, governance, and accountability as foundational elements for organizations aiming to amplify their AI investment returns. As businesses continue to explore the vast potential of AI technologies, a well-structured approach will be crucial in ensuring that they remain competitive and responsive to evolving market dynamics.